From 0f8f23d1d0d1295821ff6383eec768130a7231a0 Mon Sep 17 00:00:00 2001 From: Apurba Bose <44209735+apbose@users.noreply.github.com> Date: Fri, 30 Aug 2024 09:05:52 -0700 Subject: [PATCH 01/14] Dynamic shape index (#3039) --- .../dynamo/conversion/impl/select.py | 66 ++++- tests/py/dynamo/conversion/test_index_aten.py | 269 +++++++++--------- 2 files changed, 181 insertions(+), 154 deletions(-) diff --git a/py/torch_tensorrt/dynamo/conversion/impl/select.py b/py/torch_tensorrt/dynamo/conversion/impl/select.py index 48193dbe11..0d55c5f014 100644 --- a/py/torch_tensorrt/dynamo/conversion/impl/select.py +++ b/py/torch_tensorrt/dynamo/conversion/impl/select.py @@ -17,6 +17,8 @@ ) from torch_tensorrt.dynamo.conversion.impl.elementwise import convert_binary_elementwise from torch_tensorrt.dynamo.conversion.impl.shape import get_shape_with_dynamic_shape +from torch_tensorrt.dynamo.conversion.impl.shape import shape as get_shape +from torch_tensorrt.dynamo.utils import DYNAMIC_DIM from torch_tensorrt.fx.converters.converter_utils import ( has_dynamic_shape, set_layer_name, @@ -111,17 +113,18 @@ def index( else: input_shape = input.shape _LOGGER.debug(f"The input shape is {input.shape}") - if dynamic_shape: - input_shape = get_shape_with_dynamic_shape( - ctx.net, target, source_ir, name, input_shape, input - ) rank = len(input_shape) adv_indx_count = len(adv_indx_indices) dim_tensor_list = [] for i in range(rank): - dim = input_shape[i] - dim_tensor = get_trt_tensor(ctx, dim, name + f"_individual_dim_{i}") + if input_shape[i] != DYNAMIC_DIM: + dim = input_shape[i] + dim_tensor = get_trt_tensor(ctx, dim, name + f"_individual_dim_{i}") + else: + dim_tensor = get_shape( + ctx, target, source_ir, name + f"_individual_dim_dyn_{i}", input, i + ) # dim_tensor_list is a list of tensors dim_tensor_list.append(dim_tensor) @@ -150,12 +153,53 @@ def index( # transpose_tensor_shape = ctx.net.add_shape(transpose_tensor) transpose_tensor_shape = transpose_tensor.shape _LOGGER.debug(f"The shape of transpose tensor is {transpose_tensor_shape}") + mult_d0 = 1 + dim_tensor_shape_mult_d0 = 1 for i in range(adv_indx_count): - mult_d0 = mult_d0 * transpose_tensor_shape[i] + if transpose_tensor_shape[i] == DYNAMIC_DIM: + dim_tensor_shape_mult_d0 = get_shape( + ctx, + target, + source_ir, + name + f"_transpose_tensor_shape_mult_d0_{i}", + transpose_tensor, + i, + ) + else: + dim_tensor_shape_mult_d0 = transpose_tensor_shape[i] + mult_d0 = convert_binary_elementwise( + ctx, + target, + source_ir, + name + f"_shape_{i}", + trt.ElementWiseOperation.PROD, + mult_d0, + dim_tensor_shape_mult_d0, + ) mult_d1 = 1 + dim_tensor_shape_mult_d1 = 1 for i in range(adv_indx_count, rank): - mult_d1 = mult_d1 * transpose_tensor_shape[i] + if transpose_tensor_shape[i] == DYNAMIC_DIM: + dim_tensor_shape_mult_d1 = get_shape( + ctx, + target, + source_ir, + name + f"_transpose_tensor_shape_mult_d0_{i}", + transpose_tensor, + i, + ) + else: + dim_tensor_shape_mult_d1 = transpose_tensor_shape[i] + mult_d1 = convert_binary_elementwise( + ctx, + target, + source_ir, + name + f"_shape_{i}", + trt.ElementWiseOperation.PROD, + mult_d1, + dim_tensor_shape_mult_d1, + ) concat_tensor_layer = ctx.net.add_concatenation( [ @@ -185,11 +229,7 @@ def index( ctx, cum_adv_index, name + "_index_sum_intermediate" ) else: - multiplier = get_trt_tensor( - ctx, - dim_tensor_list[adv_indx_indices[adv_indx_count - 1]], - name + "_dim_last", - ) + multiplier = dim_tensor_list[adv_indx_indices[adv_indx_count - 1]] cum_adv_index = tensor_indices[adv_indx_count - 1] for i in range(adv_indx_count - 2, -1, -1): adv_index = convert_binary_elementwise( diff --git a/tests/py/dynamo/conversion/test_index_aten.py b/tests/py/dynamo/conversion/test_index_aten.py index bf7769c608..8e21f945dc 100644 --- a/tests/py/dynamo/conversion/test_index_aten.py +++ b/tests/py/dynamo/conversion/test_index_aten.py @@ -1,28 +1,94 @@ import torch import torch.nn as nn +from parameterized import parameterized from torch.testing._internal.common_utils import run_tests +from torch_tensorrt import Input from .harness import DispatchTestCase -class TestIndexConverter(DispatchTestCase): - def test_index_zero_two_dim(self): - class TestModule(nn.Module): +class TestIndexConstantConverter(DispatchTestCase): + @parameterized.expand( + [ + ( + "index_zero_two_dim_indices_input", + [None, torch.randint(0, 1, (1, 1))], + torch.randn(2, 2), + ), + ( + "index_zero_three_dim_indices_input", + [None, torch.randint(0, 1, (1, 1)), None], + torch.randn(2, 2, 2), + ), + ( + "index_zero_index_one_three_dim_indices_input", + [None, torch.randint(0, 1, (1, 1)), torch.randint(0, 1, (1, 1))], + torch.randn(2, 2, 2), + ), + ( + "index_zero_index_one_four_dim_indices_input", + [None, torch.tensor([0, 0, 1, 1]), torch.tensor([0, 0, 1, 1]), None], + torch.randn(2, 4, 4, 2), + ), + ( + "index_zero_index_one_four_dim_indices_input_SD", + [ + None, + torch.tensor([0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7]), + torch.tensor([0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7]), + None, + ], + torch.randn(2, 1280, 8, 8), + ), + ( + "index_zero_index_one_four_dim_indices_input_SD_unsqueeze", + [ + None, + torch.tensor([0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7]) + .unsqueeze(0) + .T.long(), + torch.tensor([0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7]) + .unsqueeze(0) + .T.long(), + None, + ], + torch.randn(2, 1280, 8, 8), + ), + ( + "index_zero_index_one_four_dim_indices_input_SD_unsqueeze_broadcast", + [ + None, + torch.tensor([0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7]), + torch.tensor([0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7]) + .unsqueeze(0) + .T.long(), + None, + ], + torch.randn(2, 1280, 8, 8), + ), + ( + "index_zero_index_one_four_dim_indices_input_non_continuous", + [None, torch.tensor([0, 0, 1, 1]), None, torch.tensor([0, 0, 1, 1])], + torch.randn(2, 4, 4, 2), + ), + ] + ) + def test_index_constant(self, _, index, input): + class TestModule(torch.nn.Module): def __init__(self): - self.index0 = torch.randint(0, 1, (1, 1)) super().__init__() - def forward(self, x): - indices = [None, self.index0] - out = torch.ops.aten.index.Tensor(x, indices) - return out + def forward(self, input): + return torch.ops.aten.index.Tensor(input, index) + + inputs = [input] + self.run_test(TestModule(), inputs) - input = [torch.randn(2, 2)] - self.run_test( - TestModule(), - input, - ) +# The below tests cannot be included in the parameterized +# [None, index0] cannot be passed as torch.Tensor to DispatchTestCase.run_test() +# tensorrt.Input requires the input to be torch Tensor +class TestIndexConverter(DispatchTestCase): def test_index_zero_two_dim_ITensor(self): class TestModule(nn.Module): def forward(self, x, index0): @@ -38,23 +104,6 @@ def forward(self, x, index0): [input, index0], ) - def test_index_zero_index_three_dim(self): - class TestModule(nn.Module): - def __init__(self): - self.index0 = torch.randint(0, 1, (1, 1)) - super().__init__() - - def forward(self, x): - indices = [None, self.index0, None] - out = torch.ops.aten.index.Tensor(x, indices) - return out - - input = [torch.randn(2, 2, 2)] - self.run_test( - TestModule(), - input, - ) - def test_index_zero_index_three_dim_ITensor(self): class TestModule(nn.Module): def forward(self, x, index0): @@ -67,121 +116,59 @@ def forward(self, x, index0): index0 = index0.to(torch.int32) self.run_test(TestModule(), [input, index0]) - def test_index_zero_index_one_index_two_three_dim(self): - class TestModule(nn.Module): - def __init__(self): - self.index0 = torch.randint(0, 1, (1, 1)) - self.index1 = torch.randint(0, 1, (1, 1)) - super().__init__() - - def forward(self, x): - indices = [None, self.index0, self.index1] - out = torch.ops.aten.index.Tensor(x, indices) - return out - input = [torch.randn(2, 2, 2)] - self.run_test( - TestModule(), - input, - ) - - def test_index_zero_index_one_four_dim(self): - class TestModule(nn.Module): +class TestIndexDynamicConstantConverter(DispatchTestCase): + @parameterized.expand( + [ + ( + "index_zero_two_dim_indices_input_min_opt_max", + [None, torch.randint(0, 1, (1, 1))], + (2, 1), + (2, 2), + (2, 2), + ), + ( + "index_zero_three_dim_indices_input_min_opt_max", + [None, torch.randint(0, 1, (1, 1)), None], + (2, 1, 2), + (2, 2, 2), + (2, 2, 2), + ), + ( + "index_zero_index_one_three_dim_indices_input_min_opt_max", + [None, torch.randint(0, 1, (1, 1)), torch.randint(0, 1, (1, 1))], + (2, 1, 2), + (2, 2, 2), + (2, 2, 2), + ), + ( + "index_zero_index_one_four_dim_indices_input_min_opt_max", + [None, torch.tensor([0, 0, 1, 1]), torch.tensor([0, 0, 1, 1]), None], + (2, 1, 4, 2), + (2, 4, 4, 2), + (2, 4, 4, 2), + ), + ] + ) + def test_index_constant_dynamic( + self, _, index, input_min_shape, input_opt_shape, input_max_shape + ): + class TestModule(torch.nn.Module): def __init__(self): - self.index0 = torch.tensor([0, 0, 1, 1]) - self.index1 = torch.tensor([0, 0, 1, 1]) super().__init__() - def forward(self, x): - indices = [None, self.index0, self.index1, None] - out = torch.ops.aten.index.Tensor(x, indices) - return out - - input = [torch.randn(2, 4, 4, 2)] - self.run_test( - TestModule(), - input, - ) - - def test_index_zero_index_one_four_dim_SD(self): - class TestModule(nn.Module): - def __init__(self): - self.index0 = torch.tensor( - [0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7] - ) - self.index1 = torch.tensor( - [0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7] - ) - super().__init__() - - def forward(self, x): - indices = [None, self.index0, self.index1, None] - out = torch.ops.aten.index.Tensor(x, indices) - return out - - input = [torch.randn(2, 1280, 8, 8)] - self.run_test( - TestModule(), - input, - ) - - def test_index_one_SD_unsqueeze_four_dim(self): - class TestModule(nn.Module): - def __init__(self): - self.index0 = torch.tensor( - [0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7] - ) - self.index1 = self.index0.unsqueeze(0).T.long() - super().__init__() - - def forward(self, x): - indices = [None, None, self.index1, self.index1] - out = torch.ops.aten.index.Tensor(x, indices) - return out - - input = [torch.randn(2, 1280, 8, 8)] - self.run_test( - TestModule(), - input, - ) - - def test_index_zero_index_one_index_two_SD_unsqueeze_four_dim_broadcast(self): - class TestModule(nn.Module): - def __init__(self): - self.index0 = torch.tensor( - [0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7] - ) - self.index1 = self.index0.unsqueeze(0).T.long() - super().__init__() - - def forward(self, x): - indices = [None, None, self.index0, self.index1] - out = torch.ops.aten.index.Tensor(x, indices) - return out - - input = [torch.randn(2, 1280, 8, 8)] - self.run_test( - TestModule(), - input, - ) - - def test_index_zero_index_one_index_four_dim_non_continuous(self): - class TestModule(nn.Module): - def __init__(self): - self.index0 = torch.tensor([0, 0, 1, 1]) - self.index1 = torch.tensor([0, 0, 1, 1]) - super().__init__() - - def forward(self, x): - indices = [None, self.index0, None, self.index1] - out = torch.ops.aten.index.Tensor(x, indices) - return out - - input = [torch.randn(2, 4, 4, 2)] - self.run_test( - TestModule(), - input, - ) + def forward(self, input): + return torch.ops.aten.index.Tensor(input, index) + + input_specs = [ + Input( + min_shape=input_min_shape, + opt_shape=input_opt_shape, + max_shape=input_max_shape, + dtype=torch.float32, + ), + ] + self.run_test_with_dynamic_shape(TestModule(), input_specs) if __name__ == "__main__": From d75f588e49ca241cf44bbdb8aad0ef3d9578da75 Mon Sep 17 00:00:00 2001 From: HolyWu Date: Tue, 3 Sep 2024 22:45:55 +0800 Subject: [PATCH 02/14] feat: Support `aten.gelu` dynamo converter (#3134) --- .../dynamo/conversion/aten_ops_converters.py | 20 +++++++- .../dynamo/conversion/impl/activation/ops.py | 25 +++++++++- .../dynamo/lowering/_decomposition_groups.py | 1 - tests/py/dynamo/conversion/test_gelu_aten.py | 46 +++++++++++++------ 4 files changed, 75 insertions(+), 17 deletions(-) diff --git a/py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py b/py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py index 1d734bca03..92dfddc44f 100644 --- a/py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py +++ b/py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py @@ -19,7 +19,7 @@ get_positive_dim, is_only_operator_on_placeholder, ) -from torch_tensorrt.fx.types import TRTTensor +from torch_tensorrt.dynamo.types import TRTTensor _LOGGER: logging.Logger = logging.getLogger(__name__) @@ -548,6 +548,24 @@ def aten_ops_hard_sigmoid( ) +@dynamo_tensorrt_converter(torch.ops.aten.gelu.default, supports_dynamic_shapes=True) +def aten_ops_gelu( + ctx: ConversionContext, + target: Target, + args: Tuple[Argument, ...], + kwargs: Dict[str, Argument], + name: str, +) -> Union[TRTTensor, Sequence[TRTTensor]]: + return impl.activation.gelu( + ctx, + target, + SourceIR.ATEN, + name, + args[0], + kwargs.get("approximate", "none"), + ) + + @dynamo_tensorrt_converter(torch.ops.aten.matmul, supports_dynamic_shapes=True) @dynamo_tensorrt_converter(torch.ops.aten.dot.default, supports_dynamic_shapes=True) @dynamo_tensorrt_converter(torch.ops.aten.mm.default, supports_dynamic_shapes=True) diff --git a/py/torch_tensorrt/dynamo/conversion/impl/activation/ops.py b/py/torch_tensorrt/dynamo/conversion/impl/activation/ops.py index f578351ef2..a563118526 100644 --- a/py/torch_tensorrt/dynamo/conversion/impl/activation/ops.py +++ b/py/torch_tensorrt/dynamo/conversion/impl/activation/ops.py @@ -7,7 +7,7 @@ from torch_tensorrt.dynamo._SourceIR import SourceIR from torch_tensorrt.dynamo.conversion._ConversionContext import ConversionContext from torch_tensorrt.dynamo.conversion.impl.activation.base import convert_activation -from torch_tensorrt.fx.types import TRTTensor +from torch_tensorrt.dynamo.types import TRTTensor def relu( @@ -327,3 +327,26 @@ def thresholded_relu_fn(x: float) -> float: alpha=alpha, dyn_range_fn=thresholded_relu_dyn_range_fn, ) + + +def gelu( + ctx: ConversionContext, + target: Target, + source_ir: Optional[SourceIR], + name: str, + input_val: TRTTensor, + approximate: str, +) -> TRTTensor: + if approximate == "none": + operation_type = trt.ActivationType.GELU_ERF + elif approximate == "tanh": + operation_type = trt.ActivationType.GELU_TANH + + return convert_activation( + ctx, + target, + source_ir, + name, + operation_type, + input_val, + ) diff --git a/py/torch_tensorrt/dynamo/lowering/_decomposition_groups.py b/py/torch_tensorrt/dynamo/lowering/_decomposition_groups.py index ae3e7e1ffa..a84a550a1e 100644 --- a/py/torch_tensorrt/dynamo/lowering/_decomposition_groups.py +++ b/py/torch_tensorrt/dynamo/lowering/_decomposition_groups.py @@ -42,7 +42,6 @@ aten.fill, aten.frac, aten._fused_moving_avg_obs_fq_helper, - aten.gelu, aten.gelu_backward, aten.glu_backward, aten.hardshrink, diff --git a/tests/py/dynamo/conversion/test_gelu_aten.py b/tests/py/dynamo/conversion/test_gelu_aten.py index df0a0eca5f..dac33a9ae6 100644 --- a/tests/py/dynamo/conversion/test_gelu_aten.py +++ b/tests/py/dynamo/conversion/test_gelu_aten.py @@ -1,49 +1,67 @@ -import pytest import torch import torch.nn as nn +from parameterized import parameterized from torch.testing._internal.common_utils import run_tests from torch_tensorrt import Input from .harness import DispatchTestCase -@pytest.mark.skip(reason="This test will be skipped.") -class TestGeLUConverter(DispatchTestCase): - def test_gelu(self): +class TestGELUConverter(DispatchTestCase): + @parameterized.expand( + [ + ("none",), + ("tanh",), + ] + ) + def test_gelu(self, approximate): class TestModule(nn.Module): def forward(self, x): - return torch.ops.aten.gelu.default(x) + return torch.ops.aten.gelu.default(x, approximate=approximate) inputs = [torch.randn(1, 10)] self.run_test(TestModule(), inputs) - def test_gelu_with_dynamic_shape(self): + @parameterized.expand( + [ + ("none",), + ("tanh",), + ] + ) + def test_gelu_with_dynamic_shape(self, approximate): class TestModule(nn.Module): def forward(self, x): - return torch.ops.aten.gelu.default(x) + return torch.ops.aten.gelu.default(x, approximate=approximate) input_specs = [ Input( - shape=(-1, -1, -1), + min_shape=(1, 1, 1), + opt_shape=(1, 2, 3), + max_shape=(3, 3, 3), dtype=torch.float32, - shape_ranges=[((1, 1, 1), (1, 2, 3), (3, 3, 3))], ), ] self.run_test_with_dynamic_shape(TestModule(), input_specs) - def test_gelu_with_dynamic_shape_four_dimensions(self): + @parameterized.expand( + [ + ("none",), + ("tanh",), + ] + ) + def test_gelu_with_dynamic_shape_four_dimensions(self, approximate): class TestModule(nn.Module): def forward(self, x): - return torch.ops.aten.gelu.default(x) + return torch.ops.aten.gelu.default(x, approximate=approximate) input_specs = [ Input( - shape=(-1, -1, -1, -1), + min_shape=(1, 1, 1, 5), + opt_shape=(1, 2, 3, 5), + max_shape=(3, 3, 3, 5), dtype=torch.float32, - shape_ranges=[((1, 1, 1, 5), (1, 2, 3, 5), (3, 3, 3, 5))], ), ] - self.run_test_with_dynamic_shape(TestModule(), input_specs) From ae7e6c87dfee8cb037546e198860be3a620dcc54 Mon Sep 17 00:00:00 2001 From: Jiwoong Date: Tue, 3 Sep 2024 23:52:59 +0900 Subject: [PATCH 03/14] fix: get_padded_shape_tensors can now handle dynamic pads (#3123) --- .../dynamo/conversion/impl/pad.py | 33 ++++++++++++------- 1 file changed, 22 insertions(+), 11 deletions(-) diff --git a/py/torch_tensorrt/dynamo/conversion/impl/pad.py b/py/torch_tensorrt/dynamo/conversion/impl/pad.py index 66cc09684b..8cc6bd42c8 100644 --- a/py/torch_tensorrt/dynamo/conversion/impl/pad.py +++ b/py/torch_tensorrt/dynamo/conversion/impl/pad.py @@ -26,7 +26,7 @@ def get_padded_shape_tensors( source_ir: Optional[SourceIR], name: str, input: TRTTensor, - pad: Sequence[int], + pad: Sequence[Union[int, TRTTensor]], ) -> TRTTensor: rank = len(input.shape) if len(pad) // 2 > rank: @@ -47,11 +47,11 @@ def get_padded_shape_tensors( start_list = [0] * rank for i in range(len(pad) // 2): dim_index = rank - (i + 1) - pad_before = pad[i * 2] - pad_after = pad[i * 2 + 1] + pad_before = get_trt_tensor(ctx, pad[i * 2], f"{name}_pad_before_{i}") + pad_after = get_trt_tensor(ctx, pad[i * 2 + 1], f"{name}_pad_after_{i}") - pad_sum = get_trt_tensor( - ctx, pad_before + pad_after, f"{name}_pad_sum_{i}", dtype=np.int32 + pad_sum = impl.elementwise.add( + ctx, target, source_ir, f"{name}_pad_sum_{i}", pad_before, pad_after ) dim_shape = ctx.net.add_slice( input_shape_tensor, @@ -63,7 +63,9 @@ def get_padded_shape_tensors( new_dim_shape = impl.elementwise.add( ctx, target, source_ir, f"{name}_shape_dim_{i}", dim_shape, pad_sum ) - start_list[dim_index] = -pad_before + start_list[dim_index] = impl.elementwise.sub( + ctx, target, source_ir, f"{name}_pad_before_neg_{i}", 0, pad_before + ) slices = [] for j in range(rank): @@ -79,14 +81,23 @@ def get_padded_shape_tensors( ).get_output(0) ) padded_shape_tensor = impl.cat.cat( - ctx, target, source_ir, f"{name}_cat_dim_{i}", slices, 0 + ctx, + target, + source_ir, + f"{name}_cat_dim_{i}", + slices, + 0, + cast_dtype=padded_shape_tensor.dtype, ) - start_indices_tensor = get_trt_tensor( + start_indices_tensor = impl.cat.cat( ctx, - np.array(start_list, dtype=np.int32), + target, + source_ir, f"{name}_start_indices_tensor", - dtype=np.int32, + start_list, + 0, + cast_dtype=padded_shape_tensor.dtype, ) return start_indices_tensor, padded_shape_tensor @@ -98,7 +109,7 @@ def constant_padNd( source_ir: Optional[SourceIR], name: str, input: TRTTensor, - pad: Sequence[int], + pad: Sequence[Union[int, TRTTensor]], value: Union[int, float] = 0, ) -> TRTTensor: From 8e750394879b887dd732d7483c88ea2dbf78ae6f Mon Sep 17 00:00:00 2001 From: Jiwoong Date: Tue, 3 Sep 2024 23:53:19 +0900 Subject: [PATCH 04/14] bugfix: allow empty tuple for `inputs` or `arg_inputs` (#3122) --- py/torch_tensorrt/dynamo/_compiler.py | 4 +- .../dynamo/models/test_export_kwargs_serde.py | 47 +++++++++++++++++++ 2 files changed, 49 insertions(+), 2 deletions(-) diff --git a/py/torch_tensorrt/dynamo/_compiler.py b/py/torch_tensorrt/dynamo/_compiler.py index c28702f451..6cd3cf5f5f 100644 --- a/py/torch_tensorrt/dynamo/_compiler.py +++ b/py/torch_tensorrt/dynamo/_compiler.py @@ -604,10 +604,10 @@ def convert_exported_program_to_serialized_trt_engine( DeprecationWarning, stacklevel=2, ) - if not arg_inputs and not inputs: + if arg_inputs is None and inputs is None: raise AssertionError("'arg_inputs' and 'inputs' should not both be None.") - elif arg_inputs and inputs: + elif arg_inputs is not None and inputs is not None: raise AssertionError( "'arg_inputs' and 'inputs' should not be used at the same time." ) diff --git a/tests/py/dynamo/models/test_export_kwargs_serde.py b/tests/py/dynamo/models/test_export_kwargs_serde.py index 52a927e518..91ee59c0f4 100644 --- a/tests/py/dynamo/models/test_export_kwargs_serde.py +++ b/tests/py/dynamo/models/test_export_kwargs_serde.py @@ -525,3 +525,50 @@ def forward(self, x, b=5, c=None, d=None): engine = convert_exported_program_to_serialized_trt_engine( exp_program, **compile_spec ) + + +def test_custom_model_compile_engine_with_pure_kwarg_inputs(): + class net(nn.Module): + def __init__(self): + super().__init__() + self.conv1 = nn.Conv2d(3, 12, 3, padding=1) + self.bn = nn.BatchNorm2d(12) + self.conv2 = nn.Conv2d(12, 12, 3, padding=1) + self.fc1 = nn.Linear(12 * 56 * 56, 10) + + def forward(self, x, b=5, c=None, d=None): + x = self.conv1(x) + x = F.relu(x) + x = self.bn(x) + x = F.max_pool2d(x, (2, 2)) + x = self.conv2(x) + x = F.relu(x) + x = F.max_pool2d(x, (2, 2)) + x = torch.flatten(x, 1) + x = x + b + if c is not None: + x = x * c + if d is not None: + x = x - d["value"] + return self.fc1(x) + + model = net().eval().to("cuda") + kwargs = { + "x": torch.rand((1, 3, 224, 224)).to("cuda"), + "b": torch.tensor(6).to("cuda"), + "d": {"value": torch.tensor(8).to("cuda")}, + } + + compile_spec = { + "arg_inputs": (), + "kwarg_inputs": kwargs, + "device": torchtrt.Device("cuda:0"), + "enabled_precisions": {torch.float}, + "pass_through_build_failures": True, + "optimization_level": 1, + "min_block_size": 1, + "ir": "dynamo", + } + + exp_program = torch.export.export(model, args=(), kwargs=kwargs) + _ = convert_exported_program_to_serialized_trt_engine(exp_program, **compile_spec) From 8759736843c108d3c8edd95a8e8fc632d8372b3b Mon Sep 17 00:00:00 2001 From: Naren Dasan <1790613+narendasan@users.noreply.github.com> Date: Wed, 4 Sep 2024 10:30:10 -0600 Subject: [PATCH 05/14] docs: Adding words to the refit and engine caching tutorials (#3141) Signed-off-by: Naren Dasan --- core/runtime/Platform.cpp | 1 - docsrc/conf.py | 1 + docsrc/index.rst | 2 + examples/dynamo/README.rst | 2 + .../dynamo/engine_caching_bert_example.py | 10 + examples/dynamo/engine_caching_example.py | 182 +++++++++++++++--- examples/dynamo/refit_engine_example.py | 84 ++++++-- py/torch_tensorrt/dynamo/_compiler.py | 6 +- .../{_engine_caching.py => _engine_cache.py} | 9 +- py/torch_tensorrt/dynamo/_refit.py | 6 +- .../dynamo/conversion/_TRTInterpreter.py | 4 +- .../dynamo/conversion/_conversion.py | 5 +- py/torch_tensorrt/dynamo/utils.py | 36 ++-- setup.py | 4 +- tests/py/dynamo/models/test_engine_cache.py | 2 +- 15 files changed, 273 insertions(+), 81 deletions(-) rename py/torch_tensorrt/dynamo/{_engine_caching.py => _engine_cache.py} (96%) diff --git a/core/runtime/Platform.cpp b/core/runtime/Platform.cpp index a20159cd91..03d9e7580b 100644 --- a/core/runtime/Platform.cpp +++ b/core/runtime/Platform.cpp @@ -36,7 +36,6 @@ Platform::Platform() : _platform{Platform::PlatformEnum::kUNKNOWN} {} Platform::Platform(Platform::PlatformEnum val) : _platform{val} {} Platform::Platform(const std::string& platform_str) { - LOG_ERROR("Platform constructor: " << platform_str); auto name_map = get_name_to_platform_map(); auto it = name_map.find(platform_str); if (it != name_map.end()) { diff --git a/docsrc/conf.py b/docsrc/conf.py index 2e782358cb..daa1a30100 100644 --- a/docsrc/conf.py +++ b/docsrc/conf.py @@ -93,6 +93,7 @@ sphinx_gallery_conf = { "examples_dirs": "../examples", "gallery_dirs": "tutorials/_rendered_examples/", + "ignore_pattern": "utils.py", } # Setup the breathe extension diff --git a/docsrc/index.rst b/docsrc/index.rst index da5ee3d690..d1a91beabc 100644 --- a/docsrc/index.rst +++ b/docsrc/index.rst @@ -51,6 +51,8 @@ User Guide user_guide/using_dla tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage tutorials/_rendered_examples/dynamo/vgg16_fp8_ptq + tutorials/_rendered_examples/dynamo/engine_caching_example + tutorials/_rendered_examples/dynamo/refit_engine_example Dynamo Frontend ---------------- diff --git a/examples/dynamo/README.rst b/examples/dynamo/README.rst index 22ae16ec0c..ff3563cffe 100644 --- a/examples/dynamo/README.rst +++ b/examples/dynamo/README.rst @@ -15,3 +15,5 @@ a number of ways you can leverage this backend to accelerate inference. * :ref:`refit_engine_example`: Refitting a compiled TensorRT Graph Module with updated weights * :ref:`mutable_torchtrt_module_example`: Compile, use, and modify TensorRT Graph Module with MutableTorchTensorRTModule * :ref:`vgg16_fp8_ptq`: Compiling a VGG16 model with FP8 and PTQ using ``torch.compile`` +* :ref:`engine_caching_example`: Utilizing engine caching to speed up compilation times +* :ref:`engine_caching_bert_example`: Demonstrating engine caching on BERT diff --git a/examples/dynamo/engine_caching_bert_example.py b/examples/dynamo/engine_caching_bert_example.py index 43cfc5f15a..428c414a06 100644 --- a/examples/dynamo/engine_caching_bert_example.py +++ b/examples/dynamo/engine_caching_bert_example.py @@ -1,3 +1,13 @@ +""" + +.. _engine_caching_bert_example: + +Engine Caching (BERT) +======================= + +Small caching example on BERT. +""" + import numpy as np import torch import torch_tensorrt diff --git a/examples/dynamo/engine_caching_example.py b/examples/dynamo/engine_caching_example.py index 2d1018bb6e..5154dc1e2c 100644 --- a/examples/dynamo/engine_caching_example.py +++ b/examples/dynamo/engine_caching_example.py @@ -1,12 +1,38 @@ +""" + +.. _engine_caching_example: + +Engine Caching +======================= + +As model sizes increase, the cost of compilation will as well. With AOT methods +like ``torch.dynamo.compile``, this cost is paid upfront. However if the weights +change, the session ends or you are using JIT methods like ``torch.compile``, as +graphs get invalidated they get re-compiled, this cost will get paid repeatedly. +Engine caching is a way to mitigate this cost by saving constructed engines to disk +and re-using them when possible. This tutorial demonstrates how to use engine caching +with TensorRT in PyTorch. Engine caching can significantly speed up subsequent model +compilations reusing previously built TensorRT engines. + +We'll explore two approaches: + + 1. Using torch_tensorrt.dynamo.compile + 2. Using torch.compile with the TensorRT backend + +The example uses a pre-trained ResNet18 model and shows the +differences between compilation without caching, with caching enabled, +and when reusing cached engines. +""" + import os -from typing import Optional +from typing import Dict, Optional import numpy as np import torch import torch_tensorrt as torch_trt import torchvision.models as models from torch_tensorrt.dynamo._defaults import TIMING_CACHE_PATH -from torch_tensorrt.dynamo._engine_caching import BaseEngineCache +from torch_tensorrt.dynamo._engine_cache import BaseEngineCache np.random.seed(0) torch.manual_seed(0) @@ -23,6 +49,80 @@ def remove_timing_cache(path=TIMING_CACHE_PATH): os.remove(path) +# %% +# Engine Caching for JIT Compilation +# ---------------------------------- +# +# The primary goal of engine caching is to help speed up JIT workflows. ``torch.compile`` +# provides a great deal of flexibility in model construction which makes it a good +# first tool to try when looking to speed up your workflow. However, historically +# the cost of compilation and in particular recompilation has been a barrier to entry +# for many users. If for some reason a subgraph gets invalidated, that graph is reconstructed +# scratch prior to the addition of engine caching. Now as engines are constructed, with ``cache_built_engines=True``, +# engines are saved to disk tied to a hash of their corresponding PyTorch subgraph. If +# in a subsequent compilation, either as part of this session or a new session, the cache will +# pull the built engine and **refit** the weights which can reduce compilation times by orders of magnitude. +# As such, in order to insert a new engine into the cache (i.e. ``cache_built_engines=True``), +# the engine must be refitable (``make_refittable=True``). See :ref:`refit_engine_example` for more details. + + +def torch_compile(iterations=3): + times = [] + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + + # The 1st iteration is to measure the compilation time without engine caching + # The 2nd and 3rd iterations are to measure the compilation time with engine caching. + # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration. + # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine. + for i in range(iterations): + inputs = [torch.rand((100, 3, 224, 224)).to("cuda")] + # remove timing cache and reset dynamo just for engine caching messurement + remove_timing_cache() + torch._dynamo.reset() + + if i == 0: + cache_built_engines = False + reuse_cached_engines = False + else: + cache_built_engines = True + reuse_cached_engines = True + + start.record() + compiled_model = torch.compile( + model, + backend="tensorrt", + options={ + "use_python_runtime": True, + "enabled_precisions": enabled_precisions, + "debug": debug, + "min_block_size": min_block_size, + "make_refitable": True, + "cache_built_engines": cache_built_engines, + "reuse_cached_engines": reuse_cached_engines, + }, + ) + compiled_model(*inputs) # trigger the compilation + end.record() + torch.cuda.synchronize() + times.append(start.elapsed_time(end)) + + print("----------------torch_compile----------------") + print("disable engine caching, used:", times[0], "ms") + print("enable engine caching to cache engines, used:", times[1], "ms") + print("enable engine caching to reuse engines, used:", times[2], "ms") + + +torch_compile() + +# %% +# Engine Caching for AOT Compilation +# ---------------------------------- +# Similarly to the JIT workflow, AOT workflows can benefit from engine caching. +# As the same architecture or common subgraphs get recompiled, the cache will pull +# previously built engines and refit the weights. + + def dynamo_compile(iterations=3): times = [] start = torch.cuda.Event(enable_timing=True) @@ -73,42 +173,72 @@ def dynamo_compile(iterations=3): print("enable engine caching to reuse engines, used:", times[2], "ms") +dynamo_compile() + +# %% # Custom Engine Cache -class MyEngineCache(BaseEngineCache): +# ---------------------- +# +# By default, the engine cache is stored in the system's temporary directory. Both the cache directory and +# size limit can be customized by passing ``engine_cache_dir`` and ``engine_cache_size``. +# Users can also define their own engine cache implementation by extending the ``BaseEngineCache`` class. +# This allows for remote or shared caching if so desired. +# +# The custom engine cache should implement the following methods: +# - ``save``: Save the engine blob to the cache. +# - ``load``: Load the engine blob from the cache. +# +# The hash provided by the cache systen is a weight agnostic hash of the originating PyTorch subgraph (post lowering). +# The blob contains a serialized engine, calling spec data, and weight map information in the pickle format +# +# Below is an example of a custom engine cache implementation that implents a ``RAMEngineCache``. + + +class RAMEngineCache(BaseEngineCache): def __init__( self, - engine_cache_dir: str, ) -> None: - self.engine_cache_dir = engine_cache_dir + """ + Constructs a user held engine cache in memory. + """ + self.engine_cache: Dict[str, bytes] = {} def save( self, hash: str, blob: bytes, - prefix: str = "blob", ): - if not os.path.exists(self.engine_cache_dir): - os.makedirs(self.engine_cache_dir, exist_ok=True) + """ + Insert the engine blob to the cache. - path = os.path.join( - self.engine_cache_dir, - f"{prefix}_{hash}.bin", - ) - with open(path, "wb") as f: - f.write(blob) + Args: + hash (str): The hash key to associate with the engine blob. + blob (bytes): The engine blob to be saved. - def load(self, hash: str, prefix: str = "blob") -> Optional[bytes]: - path = os.path.join(self.engine_cache_dir, f"{prefix}_{hash}.bin") - if os.path.exists(path): - with open(path, "rb") as f: - blob = f.read() - return blob - return None + Returns: + None + """ + self.engine_cache[hash] = blob + def load(self, hash: str) -> Optional[bytes]: + """ + Load the engine blob from the cache. -def torch_compile(iterations=3): + Args: + hash (str): The hash key of the engine to load. + + Returns: + Optional[bytes]: The engine blob if found, None otherwise. + """ + if hash in self.engine_cache: + return self.engine_cache[hash] + else: + return None + + +def torch_compile_my_cache(iterations=3): times = [] - engine_cache = MyEngineCache("/tmp/your_dir") + engine_cache = RAMEngineCache() start = torch.cuda.Event(enable_timing=True) end = torch.cuda.Event(enable_timing=True) @@ -141,7 +271,7 @@ def torch_compile(iterations=3): "make_refitable": True, "cache_built_engines": cache_built_engines, "reuse_cached_engines": reuse_cached_engines, - "custom_engine_cache": engine_cache, # use custom engine cache + "custom_engine_cache": engine_cache, }, ) compiled_model(*inputs) # trigger the compilation @@ -155,6 +285,4 @@ def torch_compile(iterations=3): print("enable engine caching to reuse engines, used:", times[2], "ms") -if __name__ == "__main__": - dynamo_compile() - torch_compile() +torch_compile_my_cache() diff --git a/examples/dynamo/refit_engine_example.py b/examples/dynamo/refit_engine_example.py index c8cd5590d3..1feb033a3a 100644 --- a/examples/dynamo/refit_engine_example.py +++ b/examples/dynamo/refit_engine_example.py @@ -1,19 +1,26 @@ """ .. _refit_engine_example: -Refit TenorRT Graph Module with Torch-TensorRT +Refitting Torch-TensorRT Programs with New Weights =================================================================== -We are going to demonstrate how a compiled TensorRT Graph Module can be refitted with updated weights. - -In many cases, we frequently update the weights of models, such as applying various LoRA to Stable Diffusion or constant A/B testing of AI products. -That poses challenges for TensorRT inference optimizations, as compiling the TensorRT engines takes significant time, making repetitive compilation highly inefficient. -Torch-TensorRT supports refitting TensorRT graph modules without re-compiling the engine, considerably accelerating the workflow. +Compilation is an expensive operation as it involves many graph transformations, translations +and optimizations applied on the model. In cases were the weights of a model might be updated +occasionally (e.g. inserting LoRA adapters), the large cost of recompilation can make it infeasible +to use TensorRT if the compiled program needed to be built from scratch each time. Torch-TensorRT +provides a PyTorch native mechanism to update the weights of a compiled TensorRT program without +recompiling from scratch through weight refitting. In this tutorial, we are going to walk through -1. Compiling a PyTorch model to a TensorRT Graph Module -2. Save and load a graph module -3. Refit the graph module + + 1. Compiling a PyTorch model to a TensorRT Graph Module + 2. Save and load a graph module + 3. Refit the graph module + +This tutorial focuses mostly on the AOT workflow where it is most likely that a user might need to +manually refit a module. In the JIT workflow, weight changes trigger recompilation. As the engine +has previously been built, with an engine cache enabled, Torch-TensorRT can automatically recognize +a previously built engine, trigger refit and short cut recompilation on behalf of the user (see: :ref:`engine_caching_example`). """ # %% @@ -36,10 +43,17 @@ # %% -# Compile the module for the first time and save it. -# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ - -model = models.resnet18(pretrained=True).eval().to("cuda") +# Make a Refitable Compilation Program +# --------------------------------------- +# +# The inital step is to compile a module and save it as with a normal. Note that there is an +# additional parameter `make_refitable` that is set to `True`. This parameter is used to +# indicate that the engine being built should support weight refitting later. Engines built without +# these setttings will not be able to be refit. +# +# In this case we are going to compile a ResNet18 model with randomly initialized weights and save it. + +model = models.resnet18(pretrained=False).eval().to("cuda") exp_program = torch.export.export(model, tuple(inputs)) enabled_precisions = {torch.float} debug = False @@ -59,16 +73,20 @@ ) # Output is a torch.fx.GraphModule # Save the graph module as an exported program -# This is only supported when use_python_runtime = False torch_trt.save(trt_gm, "./compiled.ep", inputs=inputs) # %% -# Refit the module with update model weights -# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +# Refit the Program with Pretrained Weights +# ------------------------------------------ +# +# Random weights are not useful for inference. But now instead of recompiling the model, we can +# refit the model with the pretrained weights. This is done by setting up another PyTorch module +# with the target weights and exporting it as an ExportedProgram. Then the ``refit_module_weights`` +# function is used to update the weights of the compiled module with the new weights. # Create and compile the updated model -model2 = models.resnet18(pretrained=False).eval().to("cuda") +model2 = models.resnet18(pretrained=True).eval().to("cuda") exp_program2 = torch.export.export(model2, tuple(inputs)) @@ -91,8 +109,32 @@ print("Refit successfully!") # %% -# Alternative Workflow using Python Runtime +# +# Advanced Usage # ----------------------------- - -# Currently python runtime does not support engine serialization. So the refitting will be done in the same runtime. -# This usecase is more useful when you need to switch different weights in the same runtime, such as using Stable Diffusion. +# +# There are a number of settings you can use to control the refit process +# +# Weight Map Cache +# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +# +# Weight refitting works by matching the weights of the compiled module with the new weights from +# the user supplied ExportedProgram. Since 1:1 name matching from PyTorch to TensorRT is hard to accomplish, +# the only gaurenteed way to match weights at *refit-time* is to pass the new ExportedProgram through the +# early phases of the compilation process to generate near identical weight names. This can be expensive +# and is not always necessary. +# +# To avoid this, **At initial compile**, Torch-TensorRt will attempt to cache a direct mapping from PyTorch +# weights to TensorRT weights. This cache is stored in the compiled module as metadata and can be used +# to speed up refit. If the cache is not present, the refit system will fallback to rebuilding the mapping at +# refit-time. Use of this cache is controlled by the ``use_weight_map_cache`` parameter. +# +# Since the cache uses a heuristic based system for matching PyTorch and TensorRT weights, you may want to verify the refitting. This can be done by setting +# ``verify_output`` to True and providing sample ``arg_inputs`` and ``kwarg_inputs``. When this is done, the refit +# system will run the refitted module and the user supplied module on the same inputs and compare the outputs. +# +# In-Place Refit +# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +# +# ``in_place`` allows the user to refit the module in place. This is useful when the user wants to update the weights +# of the compiled module without creating a new module. diff --git a/py/torch_tensorrt/dynamo/_compiler.py b/py/torch_tensorrt/dynamo/_compiler.py index 6cd3cf5f5f..2e6ff039b4 100644 --- a/py/torch_tensorrt/dynamo/_compiler.py +++ b/py/torch_tensorrt/dynamo/_compiler.py @@ -18,7 +18,7 @@ dryrun_stats_display, parse_non_trt_nodes, ) -from torch_tensorrt.dynamo._engine_caching import BaseEngineCache, DiskEngineCache +from torch_tensorrt.dynamo._engine_cache import BaseEngineCache, DiskEngineCache from torch_tensorrt.dynamo.conversion import ( CompilationSettings, UnsupportedOperatorException, @@ -85,8 +85,8 @@ def compile( lazy_engine_init: bool = _defaults.LAZY_ENGINE_INIT, cache_built_engines: bool = _defaults.CACHE_BUILT_ENGINES, reuse_cached_engines: bool = _defaults.REUSE_CACHED_ENGINES, - engine_cache_dir: Optional[str] = _defaults.ENGINE_CACHE_DIR, - engine_cache_size: Optional[int] = _defaults.ENGINE_CACHE_SIZE, + engine_cache_dir: str = _defaults.ENGINE_CACHE_DIR, + engine_cache_size: int = _defaults.ENGINE_CACHE_SIZE, custom_engine_cache: Optional[BaseEngineCache] = _defaults.CUSTOM_ENGINE_CACHE, **kwargs: Any, ) -> torch.fx.GraphModule: diff --git a/py/torch_tensorrt/dynamo/_engine_caching.py b/py/torch_tensorrt/dynamo/_engine_cache.py similarity index 96% rename from py/torch_tensorrt/dynamo/_engine_caching.py rename to py/torch_tensorrt/dynamo/_engine_cache.py index c8ff7aba50..7a33a81521 100644 --- a/py/torch_tensorrt/dynamo/_engine_caching.py +++ b/py/torch_tensorrt/dynamo/_engine_cache.py @@ -144,6 +144,10 @@ def get_dir_size(path: str) -> int: if engine_cache_dir not in DiskEngineCache.dir2hash2size_map: DiskEngineCache.dir2hash2size_map[engine_cache_dir] = {} + _LOGGER.info( + f"Disk engine cache initialized (cache directory:{self.engine_cache_dir}, max size: {self.total_engine_cache_size})" + ) + def has_available_cache_size(self, needed_size: int) -> bool: """Check if the cache has available space for saving object @@ -184,7 +188,7 @@ def LRU() -> None: engine_hash, 0 ) ) - _LOGGER.info( + _LOGGER.debug( f"Removed the engine cache at {engine_path}, available cache size: {self.available_engine_cache_size} bytes." ) except Exception as e: @@ -228,7 +232,7 @@ def save( try: with open(blob_path, "wb") as f: f.write(blob) - _LOGGER.info(f"The blob was saved to {blob_path}") + _LOGGER.debug(f"The engine added to cache, saved to {blob_path}") except Exception as e: del DiskEngineCache.dir2hash2size_map[self.engine_cache_dir][hash] self.available_engine_cache_size += blob_size @@ -247,5 +251,6 @@ def load(self, hash: str) -> Optional[bytes]: if os.path.exists(blob_path): with open(blob_path, "rb") as f: blob = f.read() + _LOGGER.debug(f"Engine found in cache, loaded from {blob_path}") return blob return None diff --git a/py/torch_tensorrt/dynamo/_refit.py b/py/torch_tensorrt/dynamo/_refit.py index 4ce7d0b150..8b0d7c3e20 100644 --- a/py/torch_tensorrt/dynamo/_refit.py +++ b/py/torch_tensorrt/dynamo/_refit.py @@ -6,7 +6,6 @@ from typing import Any, List, Optional, Sequence, Tuple import numpy as np -import tensorrt as trt import torch from torch.export import ExportedProgram from torch_tensorrt._enums import dtype @@ -35,6 +34,7 @@ ) from torch_tensorrt.dynamo.utils import ( check_module_output, + get_model_device, get_torch_inputs, set_log_level, to_torch_device, @@ -42,6 +42,8 @@ ) from torch_tensorrt.logging import TRT_LOGGER +import tensorrt as trt + logger = logging.getLogger(__name__) @@ -146,7 +148,7 @@ def _refit_single_trt_engine_with_gm( """ refitted = set() - torch_device = list(new_gm.state_dict().values())[0].device.type + torch_device = get_model_device(new_gm) refitter = trt.Refitter(old_engine, TRT_LOGGER) weight_list = refitter.get_all_weights() diff --git a/py/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py b/py/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py index 3c97c8347a..84fe345137 100644 --- a/py/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py +++ b/py/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py @@ -18,7 +18,6 @@ ) import numpy as np -import tensorrt as trt import torch import torch.fx from torch.fx.node import _get_qualified_name @@ -27,7 +26,7 @@ from torch_tensorrt._enums import dtype from torch_tensorrt._Input import Input from torch_tensorrt.dynamo import _defaults -from torch_tensorrt.dynamo._engine_caching import BaseEngineCache +from torch_tensorrt.dynamo._engine_cache import BaseEngineCache from torch_tensorrt.dynamo._settings import CompilationSettings from torch_tensorrt.dynamo.conversion._ConversionContext import ConversionContext from torch_tensorrt.dynamo.conversion._ConverterRegistry import ( @@ -44,6 +43,7 @@ from torch_tensorrt.fx.observer import Observer from torch_tensorrt.logging import TRT_LOGGER +import tensorrt as trt from packaging import version _LOGGER: logging.Logger = logging.getLogger(__name__) diff --git a/py/torch_tensorrt/dynamo/conversion/_conversion.py b/py/torch_tensorrt/dynamo/conversion/_conversion.py index cd38ce56e6..f0b65b3a6e 100644 --- a/py/torch_tensorrt/dynamo/conversion/_conversion.py +++ b/py/torch_tensorrt/dynamo/conversion/_conversion.py @@ -3,14 +3,13 @@ import logging from typing import Any, List, Optional, Sequence -import tensorrt as trt import torch from torch.fx.experimental.proxy_tensor import unset_fake_temporarily from torch_tensorrt._Device import Device from torch_tensorrt._enums import dtype from torch_tensorrt._features import ENABLED_FEATURES from torch_tensorrt._Input import Input -from torch_tensorrt.dynamo._engine_caching import BaseEngineCache +from torch_tensorrt.dynamo._engine_cache import BaseEngineCache from torch_tensorrt.dynamo._settings import CompilationSettings from torch_tensorrt.dynamo.conversion._TRTInterpreter import ( TRTInterpreter, @@ -19,6 +18,8 @@ from torch_tensorrt.dynamo.runtime import PythonTorchTensorRTModule, TorchTensorRTModule from torch_tensorrt.dynamo.utils import get_model_device, get_torch_inputs +import tensorrt as trt + logger = logging.getLogger(__name__) diff --git a/py/torch_tensorrt/dynamo/utils.py b/py/torch_tensorrt/dynamo/utils.py index 66192d59a0..2af7922cd1 100644 --- a/py/torch_tensorrt/dynamo/utils.py +++ b/py/torch_tensorrt/dynamo/utils.py @@ -3,19 +3,19 @@ import logging from dataclasses import fields, replace from enum import Enum -from typing import Any, Callable, Dict, Optional, Sequence, Tuple, Union +from typing import Any, Callable, Dict, List, Optional, Sequence, Tuple, Union import numpy as np -import tensorrt as trt import torch from torch._subclasses.fake_tensor import FakeTensor from torch_tensorrt._Device import Device from torch_tensorrt._enums import dtype from torch_tensorrt._Input import Input from torch_tensorrt.dynamo import _defaults -from torch_tensorrt.dynamo._engine_caching import BaseEngineCache +from torch_tensorrt.dynamo._engine_cache import BaseEngineCache from torch_tensorrt.dynamo._settings import CompilationSettings +import tensorrt as trt from packaging import version from .types import TRTDataType @@ -148,10 +148,10 @@ def get_torch_tensor( def get_torch_inputs( - inputs: Sequence[Input] | Dict[Any, Any], + inputs: Sequence[Input] | Dict[str, Any], device: Union[Device, torch.device, str], mode: str = "", -) -> Sequence[torch.Tensor] | Dict[str, torch.Tensor]: +) -> Sequence[Union[int, torch.Tensor]] | Dict[str, Union[int, torch.Tensor]]: """ Return the torch_tensor from the Input object. If mode is set, this implies user is using dynamic shaped inputs and return the corresponding input based @@ -160,26 +160,26 @@ def get_torch_inputs( device = to_torch_device(device) if isinstance(inputs, dict): - result = {} + result_dict: Dict[str, Union[int, torch.Tensor]] = {} for k, v in inputs.items(): if isinstance(v, (list, tuple, dict)): - result[k] = get_torch_inputs(v, device) + result_dict[k] = get_torch_inputs(v, device) elif isinstance(v, Input): - result[k] = get_torch_tensor(v, device, mode) + result_dict[k] = get_torch_tensor(v, device, mode) + return result_dict else: - result = [] + result_list: List[Union[int, torch.Tensor]] = [] for input in inputs: if isinstance(input, Input): - result.append(get_torch_tensor(input, device, mode)) + result_list.append(get_torch_tensor(input, device, mode)) elif isinstance(input, torch.Tensor): - result.append(input.to(device)) + result_list.append(input.to(device)) else: raise AssertionError(f"Input type {type(input)} is not a valid type") + return result_list - return result - -def get_model_device(module: torch.fx.GraphModule) -> Union[Device, torch.device, str]: +def get_model_device(module: torch.fx.GraphModule) -> torch.device: """ Returns the device on which the module parameters exist. """ @@ -306,7 +306,7 @@ def contains_sym_int(tensor: torch.Tensor) -> bool: return any(isinstance(dim, torch.SymInt) for dim in tensor) -def extract_var_range_info(symbolic_integer: torch.SymInt) -> Dict[str, Any]: +def extract_var_range_info(symbolic_integer: torch.SymInt) -> Dict[str, int]: """ This function returns the min, max, opt values of a symbolic integer. """ @@ -335,14 +335,14 @@ def extract_var_range_info(symbolic_integer: torch.SymInt) -> Dict[str, Any]: def unwrap_tensor_shape( tensor: Union[torch.Tensor, FakeTensor, torch.SymInt] -) -> Sequence[Any]: +) -> Sequence[Union[int, Tuple[int, int]]]: """ This is a helper function used to print/return the shape of the tensor. For regular torch.tensor's, it returns the static shape. For symbolic tensors, eg:(1, s0, 4), this function returns [1, [min, max], 4]. The min and max correspond to the lower and upper values of s0 symbolic dimension. """ - tensor_shape = [] + tensor_shape: List[Union[int, Tuple[int, int]]] = [] # for dimension in tensor.shape: if isinstance(tensor, int): tensor_shape.append(tensor) @@ -509,7 +509,7 @@ def parse_dynamo_kwargs( if kwargs.get("custom_engine_cache") is not None: engine_cache = kwargs.get("custom_engine_cache") else: - from torch_tensorrt.dynamo._engine_caching import DiskEngineCache + from torch_tensorrt.dynamo._engine_cache import DiskEngineCache engine_cache_dir = kwargs.get( "engine_cache_dir", _defaults.ENGINE_CACHE_DIR diff --git a/setup.py b/setup.py index 06b163c51c..de532d9071 100644 --- a/setup.py +++ b/setup.py @@ -484,7 +484,7 @@ def run(self): if not (PY_ONLY or NO_TS): tensorrt_linux_external_dir = ( lambda: subprocess.check_output( - ["bazel", "query", "@tensorrt//:nvinfer", "--output", "location"] + [BAZEL_EXE, "query", "@tensorrt//:nvinfer", "--output", "location"] ) .decode("ascii") .strip() @@ -492,7 +492,7 @@ def run(self): ) tensorrt_windows_external_dir = ( lambda: subprocess.check_output( - ["bazel", "query", "@tensorrt_win//:nvinfer", "--output", "location"] + [BAZEL_EXE, "query", "@tensorrt_win//:nvinfer", "--output", "location"] ) .decode("ascii") .strip() diff --git a/tests/py/dynamo/models/test_engine_cache.py b/tests/py/dynamo/models/test_engine_cache.py index 189a492d4e..770e057a36 100644 --- a/tests/py/dynamo/models/test_engine_cache.py +++ b/tests/py/dynamo/models/test_engine_cache.py @@ -10,7 +10,7 @@ import torchvision.models as models from torch.testing._internal.common_utils import TestCase from torch_tensorrt.dynamo._defaults import ENGINE_CACHE_DIR -from torch_tensorrt.dynamo._engine_caching import BaseEngineCache +from torch_tensorrt.dynamo._engine_cache import BaseEngineCache from torch_tensorrt.dynamo.utils import COSINE_THRESHOLD, cosine_similarity assertions = unittest.TestCase() From 49771b520be13bb2be7d275aae30c03f7f5de9eb Mon Sep 17 00:00:00 2001 From: Torch-TensorRT Github Bot Date: Wed, 4 Sep 2024 16:45:57 +0000 Subject: [PATCH 06/14] docs: [Automated] Regenerating documenation for 8759736 Signed-off-by: Torch-TensorRT Github Bot --- .../classtorch__tensorrt_1_1DataType.html | 6 +- ...rch__tensorrt_1_1Device_1_1DeviceType.html | 6 +- .../classtorch__tensorrt_1_1TensorFormat.html | 6 +- ...ensorrt_1_1ptq_1_1Int8CacheCalibrator.html | 6 +- ...ch__tensorrt_1_1ptq_1_1Int8Calibrator.html | 6 +- ...8h_1a18d295a837ac71add5578860b55e5502.html | 6 +- ...8h_1a282fd3c0b1c3a215148ae372070e1268.html | 6 +- ...8h_1a31398a6d4d27e28817afb0f0139e909e.html | 6 +- ...8h_1a35703561b26b1a9d2738ad7d58b27827.html | 6 +- ...8h_1abd1465eb38256d3f22cc1426b23d516b.html | 6 +- ...8h_1abe87b341f562fd1cf40b7672e4d759da.html | 6 +- ...8h_1ad19939408f7be171a74a89928b36eb59.html | 6 +- ...8h_1adad592a7b1b7eed529cdf6acd584c883.html | 6 +- docs/_cpp_api/dir_cpp.html | 6 +- docs/_cpp_api/dir_cpp_include.html | 6 +- .../dir_cpp_include_torch_tensorrt.html | 6 +- ...ng_1a130f65408ad8cbaee060f05e8db69558.html | 6 +- ...rt_1a3fbe5d72e4fc624dbd038853079620eb.html | 6 +- ..._cpp_include_torch_tensorrt_logging.h.html | 6 +- ...e_cpp_include_torch_tensorrt_macros.h.html | 6 +- ...file_cpp_include_torch_tensorrt_ptq.h.html | 6 +- ...clude_torch_tensorrt_torch_tensorrt.h.html | 6 +- ...ng_1a0593f776f469c20469e2f729fc7861a3.html | 6 +- ...ng_1a0c012cb374addd90eb1f42eaec570650.html | 6 +- ...ng_1a56e110feaaba2c3fd44bd201fd21a76a.html | 6 +- ...ng_1a7cb50492421ea9de4e3db895819df6f2.html | 6 +- ...ng_1ac46ac0901cb97e3ae6e93b45f24e90b8.html | 6 +- ...ng_1ad2efd47b6c3689e58ccc595680579ae5.html | 6 +- ...ng_1af8f3443813315af7901903d25dd495cc.html | 6 +- ...tq_1a226e3c83379d1012cde8578c1c86b16c.html | 6 +- ...tq_1a6186e305f47c1d94b6130ef6c7f7e178.html | 6 +- ...pt_1a5b405fd3bf3c8fc2e2a54cbbab979797.html | 6 +- ...pt_1a6e19490a08fb1553c9dd347a5ae79db9.html | 6 +- ...pt_1a81f9783517335dda877d8cfcf38987c9.html | 6 +- ...pt_1ae8d56472106eeef37fbe51ff7f40c9b2.html | 6 +- ...rt_1ac4ab8313ae72c2c899ea31548b528528.html | 6 +- ...rt_1ad1acd06eaeaffbbcf6e7ebf426891384.html | 6 +- ...rt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html | 6 +- docs/_cpp_api/namespace_torch_tensorrt.html | 6 +- .../namespace_torch_tensorrt__logging.html | 6 +- .../namespace_torch_tensorrt__ptq.html | 6 +- ...namespace_torch_tensorrt__torchscript.html | 6 +- ..._cpp_include_torch_tensorrt_logging.h.html | 6 +- ...e_cpp_include_torch_tensorrt_macros.h.html | 6 +- ...file_cpp_include_torch_tensorrt_ptq.h.html | 6 +- ...clude_torch_tensorrt_torch_tensorrt.h.html | 6 +- .../structtorch__tensorrt_1_1Device.html | 6 +- .../structtorch__tensorrt_1_1GraphInputs.html | 6 +- .../structtorch__tensorrt_1_1Input.html | 6 +- ...ensorrt_1_1torchscript_1_1CompileSpec.html | 6 +- docs/_cpp_api/torch_tensort_cpp.html | 8 +- docs/_cpp_api/unabridged_orphan.html | 6 +- .../engine_caching_bert_example.ipynb | 43 + .../engine_caching_example.py | 288 +++++ .../torch_export_gpt2.py | 86 ++ .../torch_export_gpt2.ipynb | 122 ++ .../engine_caching_example.ipynb | 97 ++ .../_rendered_examples_jupyter.zip | Bin 74709 -> 103522 bytes .../_rendered_examples_python.zip | Bin 55819 -> 78451 bytes .../torch_export_llama2.py | 90 ++ .../refit_engine_example.py | 84 +- .../torch_export_llama2.ipynb | 129 ++ .../refit_engine_example.ipynb | 23 +- .../engine_caching_bert_example.py | 75 ++ ..._glr_engine_caching_bert_example_thumb.png | Bin 0 -> 26794 bytes .../sphx_glr_engine_caching_example_thumb.png | Bin 0 -> 26794 bytes .../sphx_glr_torch_export_gpt2_thumb.png | Bin 0 -> 26794 bytes .../sphx_glr_torch_export_llama2_thumb.png | Bin 0 -> 26794 bytes docs/_modules/index.html | 6 +- docs/_modules/torch_tensorrt/_Device.html | 6 +- docs/_modules/torch_tensorrt/_Input.html | 6 +- docs/_modules/torch_tensorrt/_compile.html | 8 +- docs/_modules/torch_tensorrt/_enums.html | 6 +- .../torch_tensorrt/dynamo/_compiler.html | 101 +- .../torch_tensorrt/dynamo/_exporter.html | 6 +- .../torch_tensorrt/dynamo/_refit.html | 12 +- .../torch_tensorrt/dynamo/_settings.html | 14 +- .../torch_tensorrt/dynamo/_tracer.html | 6 +- .../runtime/_MutableTorchTensorRTModule.html | 6 +- .../runtime/_PythonTorchTensorRTModule.html | 6 +- .../dynamo/runtime/_TorchTensorRTModule.html | 6 +- docs/_modules/torch_tensorrt/fx/fx2trt.html | 6 +- .../torch_tensorrt/fx/input_tensor_spec.html | 6 +- docs/_modules/torch_tensorrt/fx/lower.html | 6 +- .../torch_tensorrt/fx/trt_module.html | 6 +- docs/_modules/torch_tensorrt/logging.html | 6 +- .../runtime/_multi_device_safe_mode.html | 6 +- .../torch_tensorrt/ts/_compile_spec.html | 6 +- .../_modules/torch_tensorrt/ts/_compiler.html | 6 +- docs/_modules/torch_tensorrt/ts/ptq.html | 6 +- docs/_sources/index.rst.txt | 2 + .../engine_caching_bert_example.rst.txt | 127 ++ .../dynamo/engine_caching_example.rst.txt | 361 ++++++ .../_rendered_examples/dynamo/index.rst.txt | 84 +- .../dynamo/refit_engine_example.rst.txt | 94 +- .../dynamo/torch_export_gpt2.rst.txt | 168 +++ .../dynamo/torch_export_llama2.rst.txt | 175 +++ .../_rendered_examples/index.rst.txt | 78 +- docs/_static/documentation_options.js | 2 +- docs/cli/torchtrtc.html | 6 +- docs/contributors/conversion.html | 6 +- docs/contributors/dynamo_converters.html | 6 +- docs/contributors/lowering.html | 6 +- docs/contributors/partitioning.html | 6 +- docs/contributors/phases.html | 6 +- docs/contributors/runtime.html | 6 +- docs/contributors/system_overview.html | 6 +- docs/contributors/ts_converters.html | 6 +- docs/contributors/useful_links.html | 6 +- .../writing_dynamo_aten_lowering_passes.html | 6 +- docs/dynamo/dynamo_export.html | 6 +- docs/dynamo/torch_compile.html | 14 +- docs/fx/getting_started_with_fx_path.html | 6 +- docs/genindex.html | 6 +- docs/getting_started/installation.html | 6 +- docs/getting_started/quick_start.html | 6 +- docs/index.html | 6 +- docs/indices/supported_ops.html | 6 +- docs/objects.inv | Bin 31073 -> 31865 bytes docs/py-modindex.html | 6 +- docs/py_api/dynamo.html | 17 +- docs/py_api/fx.html | 6 +- docs/py_api/logging.html | 6 +- docs/py_api/ptq.html | 6 +- docs/py_api/runtime.html | 10 +- docs/py_api/torch_tensorrt.html | 10 +- docs/py_api/ts.html | 8 +- docs/search.html | 6 +- docs/searchindex.js | 2 +- docs/sg_execution_times.html | 6 +- .../pytorch-sphinx-theme/docs/changelog.html | 6 +- .../docs/configuring.html | 6 +- .../pytorch-sphinx-theme/docs/demo/api.html | 6 +- .../pytorch-sphinx-theme/docs/demo/demo.html | 8 +- .../docs/demo/lists_tables.html | 6 +- .../pytorch-sphinx-theme/docs/demo/long.html | 6 +- .../docs/demo/structure.html | 6 +- docs/src/pytorch-sphinx-theme/docs/index.html | 6 +- .../pytorch-sphinx-theme/docs/installing.html | 6 +- ...creating_torchscript_module_in_python.html | 6 +- docs/ts/getting_started_with_cpp_api.html | 6 +- docs/ts/getting_started_with_python_api.html | 6 +- docs/ts/ptq.html | 6 +- .../ts/torchscript_frontend_from_pytorch.html | 6 +- .../dynamo/custom_kernel_plugins.html | 6 +- .../dynamo/engine_caching_bert_example.html | 868 +++++++++++++ .../dynamo/engine_caching_example.html | 1093 +++++++++++++++++ .../_rendered_examples/dynamo/index.html | 28 +- .../mutable_torchtrt_module_example.html | 6 +- .../dynamo/refit_engine_example.html | 114 +- .../dynamo/torch_compile_advanced_usage.html | 10 +- .../dynamo/torch_compile_resnet_example.html | 6 +- .../torch_compile_stable_diffusion.html | 6 +- .../torch_compile_transformers_example.html | 6 +- .../dynamo/torch_export_cudagraphs.html | 6 +- .../dynamo/torch_export_gpt2.html | 887 +++++++++++++ .../dynamo/torch_export_llama2.html | 893 ++++++++++++++ .../_rendered_examples/dynamo/vgg16_ptq.html | 6 +- docs/tutorials/_rendered_examples/index.html | 28 +- docs/tutorials/notebooks.html | 6 +- .../serving_torch_tensorrt_with_triton.html | 6 +- docs/user_guide/dynamic_shapes.html | 6 +- docs/user_guide/runtime.html | 6 +- docs/user_guide/saving_models.html | 6 +- docs/user_guide/torch_tensorrt_explained.html | 6 +- docs/user_guide/using_dla.html | 6 +- 166 files changed, 6526 insertions(+), 455 deletions(-) create mode 100644 docs/_downloads/06a1dddfb8c2b5515b697700d863a453/engine_caching_bert_example.ipynb create mode 100644 docs/_downloads/1c759c0181fe2845e5579cc82e5b7a7a/engine_caching_example.py create mode 100644 docs/_downloads/2a9ac10f2667047a7f398d1593b7ca33/torch_export_gpt2.py create mode 100644 docs/_downloads/34421db2f2a82ea2b3d9a9cc85624784/torch_export_gpt2.ipynb create mode 100644 docs/_downloads/3454ee6d4b68e83cdf0c757f0059986b/engine_caching_example.ipynb create mode 100644 docs/_downloads/7b7004dc2ea6f839be532665e16e0426/torch_export_llama2.py create mode 100644 docs/_downloads/9e148ac48490c84d381ee281904f3226/torch_export_llama2.ipynb create mode 100644 docs/_downloads/fdd0cb7713d049345adec03926d28414/engine_caching_bert_example.py create mode 100644 docs/_images/sphx_glr_engine_caching_bert_example_thumb.png create mode 100644 docs/_images/sphx_glr_engine_caching_example_thumb.png create mode 100644 docs/_images/sphx_glr_torch_export_gpt2_thumb.png create mode 100644 docs/_images/sphx_glr_torch_export_llama2_thumb.png create mode 100644 docs/_sources/tutorials/_rendered_examples/dynamo/engine_caching_bert_example.rst.txt create mode 100644 docs/_sources/tutorials/_rendered_examples/dynamo/engine_caching_example.rst.txt create mode 100644 docs/_sources/tutorials/_rendered_examples/dynamo/torch_export_gpt2.rst.txt create mode 100644 docs/_sources/tutorials/_rendered_examples/dynamo/torch_export_llama2.rst.txt create mode 100644 docs/tutorials/_rendered_examples/dynamo/engine_caching_bert_example.html create mode 100644 docs/tutorials/_rendered_examples/dynamo/engine_caching_example.html create mode 100644 docs/tutorials/_rendered_examples/dynamo/torch_export_gpt2.html create mode 100644 docs/tutorials/_rendered_examples/dynamo/torch_export_llama2.html diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1DataType.html b/docs/_cpp_api/classtorch__tensorrt_1_1DataType.html index 31e6b0f3e3..5261a19ab4 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1DataType.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1DataType.html @@ -10,7 +10,7 @@ - Class DataType — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Class DataType — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
- v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
@@ -316,6 +316,8 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Engine Caching
  • +
  • Refitting Torch-TensorRT Programs with New Weights
  • Dynamo Frontend

      diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.html b/docs/_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.html index ec2f7172ba..db11858bb9 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.html @@ -10,7 +10,7 @@ - Class Device::DeviceType — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Class Device::DeviceType — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1TensorFormat.html b/docs/_cpp_api/classtorch__tensorrt_1_1TensorFormat.html index 70c50627f4..a886b42cc9 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1TensorFormat.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1TensorFormat.html @@ -10,7 +10,7 @@ - Class TensorFormat — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Class TensorFormat — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.html b/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.html index b8489e5efa..9f2904306f 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.html @@ -10,7 +10,7 @@ - Template Class Int8CacheCalibrator — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Template Class Int8CacheCalibrator — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.html b/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.html index c85615e854..524ac6c6e5 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.html @@ -10,7 +10,7 @@ - Template Class Int8Calibrator — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Template Class Int8Calibrator — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.html b/docs/_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.html index a433a4c113..d81a35efa7 100644 --- a/docs/_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.html +++ b/docs/_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.html @@ -10,7 +10,7 @@ - Define STR — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Define STR — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.html b/docs/_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.html index 9ffed532b5..5585b7b9bb 100644 --- a/docs/_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.html +++ b/docs/_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.html @@ -10,7 +10,7 @@ - Define TORCH_TENSORRT_PATCH_VERSION — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Define TORCH_TENSORRT_PATCH_VERSION — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.html b/docs/_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.html index f5b76d65ec..be9d363205 100644 --- a/docs/_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.html +++ b/docs/_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.html @@ -10,7 +10,7 @@ - Define TORCH_TENSORRT_MAJOR_VERSION — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Define TORCH_TENSORRT_MAJOR_VERSION — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.html b/docs/_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.html index 516d81e51e..ac1ae177d2 100644 --- a/docs/_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.html +++ b/docs/_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.html @@ -10,7 +10,7 @@ - Define TORCH_TENSORRT_MINOR_VERSION — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Define TORCH_TENSORRT_MINOR_VERSION — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.html b/docs/_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.html index 9ea0983bbb..481e0df5f5 100644 --- a/docs/_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.html +++ b/docs/_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.html @@ -10,7 +10,7 @@ - Define TORCHTRT_API — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Define TORCHTRT_API — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.html b/docs/_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.html index 1dc0c8ad7c..820561ec0a 100644 --- a/docs/_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.html +++ b/docs/_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.html @@ -10,7 +10,7 @@ - Define XSTR — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Define XSTR — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.html b/docs/_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.html index 16d42190a2..e4b2984c99 100644 --- a/docs/_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.html +++ b/docs/_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.html @@ -10,7 +10,7 @@ - Define TORCHTRT_HIDDEN — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Define TORCHTRT_HIDDEN — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.html b/docs/_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.html index 02ccfdfef0..6a24f61376 100644 --- a/docs/_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.html +++ b/docs/_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.html @@ -10,7 +10,7 @@ - Define TORCH_TENSORRT_VERSION — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Define TORCH_TENSORRT_VERSION — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/_cpp_api/dir_cpp_include_torch_tensorrt.html b/docs/_cpp_api/dir_cpp_include_torch_tensorrt.html index 2b6ec936a2..adfae6b652 100644 --- a/docs/_cpp_api/dir_cpp_include_torch_tensorrt.html +++ b/docs/_cpp_api/dir_cpp_include_torch_tensorrt.html @@ -10,7 +10,7 @@ - Directory torch_tensorrt — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Directory torch_tensorrt — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.html b/docs/_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.html index 88d1bf933f..b5e17cc9f0 100644 --- a/docs/_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.html +++ b/docs/_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.html @@ -10,7 +10,7 @@ - Enum Level — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Enum Level — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.html b/docs/_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.html index 2bd0fe9493..46efbbf20c 100644 --- a/docs/_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.html +++ b/docs/_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.html @@ -10,7 +10,7 @@ - Enum EngineCapability — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Enum EngineCapability — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/file_cpp_include_torch_tensorrt_logging.h.html b/docs/_cpp_api/file_cpp_include_torch_tensorrt_logging.h.html index 14dd18617e..26994ee744 100644 --- a/docs/_cpp_api/file_cpp_include_torch_tensorrt_logging.h.html +++ b/docs/_cpp_api/file_cpp_include_torch_tensorrt_logging.h.html @@ -10,7 +10,7 @@ - File logging.h — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + File logging.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/file_cpp_include_torch_tensorrt_macros.h.html b/docs/_cpp_api/file_cpp_include_torch_tensorrt_macros.h.html index 96c429fa9d..ae2e3cad77 100644 --- a/docs/_cpp_api/file_cpp_include_torch_tensorrt_macros.h.html +++ b/docs/_cpp_api/file_cpp_include_torch_tensorrt_macros.h.html @@ -10,7 +10,7 @@ - File macros.h — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + File macros.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.html b/docs/_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.html index 2ad5cb396e..ef1d5cd486 100644 --- a/docs/_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.html +++ b/docs/_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.html @@ -10,7 +10,7 @@ - File ptq.h — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + File ptq.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.html b/docs/_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.html index 407cab6a57..3a7dbe4cfa 100644 --- a/docs/_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.html +++ b/docs/_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.html @@ -10,7 +10,7 @@ - File torch_tensorrt.h — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + File torch_tensorrt.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.html index f7a5ad663a..cf75405f74 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::get_logging_prefix — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::logging::get_logging_prefix — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.html index e552eadb27..ab72cb40b8 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::get_reportable_log_level — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::logging::get_reportable_log_level — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.html index 25a9acee0a..060361232f 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::get_is_colored_output_on — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::logging::get_is_colored_output_on — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.html index ab7ca3c3d9..6164b6828e 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::set_reportable_log_level — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::logging::set_reportable_log_level — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.html index c91a7c1318..dd6d140e6e 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::log — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::logging::log — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.html index 6539a9fa08..15c8fd97ef 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::set_is_colored_output_on — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::logging::set_is_colored_output_on — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.html index cb3438fbd4..f0d0f268d8 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::set_logging_prefix — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::logging::set_logging_prefix — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.html index 39eb779e61..bd2afc3fdd 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.html @@ -10,7 +10,7 @@ - Template Function torch_tensorrt::ptq::make_int8_cache_calibrator — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Template Function torch_tensorrt::ptq::make_int8_cache_calibrator — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.html index 68a2191bfa..8dcdf7e780 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.html @@ -10,7 +10,7 @@ - Template Function torch_tensorrt::ptq::make_int8_calibrator — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Template Function torch_tensorrt::ptq::make_int8_calibrator — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.html index 7a611c9e68..3b763f23e3 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::torchscript::check_method_operator_support — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::torchscript::check_method_operator_support — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.html index 95bb8b9f20..af1aa92340 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::torchscript::compile — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::torchscript::compile — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.html index 5596faa1e8..b5bdaacb80 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::torchscript::embed_engine_in_new_module — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::torchscript::embed_engine_in_new_module — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.html index a77675787a..0c01a5133c 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::torchscript::convert_method_to_trt_engine — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::torchscript::convert_method_to_trt_engine — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.html index 88bd6c2ced..fd1c397c98 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::get_build_info — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::get_build_info — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.html index a1ed2686c1..1fc6a140b4 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::set_device — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::set_device — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html index 82be1de9b6..1313a1cae4 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::dump_build_info — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Function torch_tensorrt::dump_build_info — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/_cpp_api/namespace_torch_tensorrt__logging.html b/docs/_cpp_api/namespace_torch_tensorrt__logging.html index 34d0fe3a5a..a9d7faf7b0 100644 --- a/docs/_cpp_api/namespace_torch_tensorrt__logging.html +++ b/docs/_cpp_api/namespace_torch_tensorrt__logging.html @@ -10,7 +10,7 @@ - Namespace torch_tensorrt::logging — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Namespace torch_tensorrt::logging — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/namespace_torch_tensorrt__ptq.html b/docs/_cpp_api/namespace_torch_tensorrt__ptq.html index 7527a31589..751ff237ff 100644 --- a/docs/_cpp_api/namespace_torch_tensorrt__ptq.html +++ b/docs/_cpp_api/namespace_torch_tensorrt__ptq.html @@ -10,7 +10,7 @@ - Namespace torch_tensorrt::ptq — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Namespace torch_tensorrt::ptq — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/namespace_torch_tensorrt__torchscript.html b/docs/_cpp_api/namespace_torch_tensorrt__torchscript.html index 674ebdbd62..9b1540b94c 100644 --- a/docs/_cpp_api/namespace_torch_tensorrt__torchscript.html +++ b/docs/_cpp_api/namespace_torch_tensorrt__torchscript.html @@ -10,7 +10,7 @@ - Namespace torch_tensorrt::torchscript — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Namespace torch_tensorrt::torchscript — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.html b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.html index c95192e7b3..c0eaedc8f1 100644 --- a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.html +++ b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.html @@ -10,7 +10,7 @@ - Program Listing for File logging.h — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Program Listing for File logging.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.html b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.html index 4d26e6ee09..db77c27520 100644 --- a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.html +++ b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.html @@ -10,7 +10,7 @@ - Program Listing for File macros.h — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Program Listing for File macros.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.html b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.html index 8934a73239..0215df7549 100644 --- a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.html +++ b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.html @@ -10,7 +10,7 @@ - Program Listing for File ptq.h — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Program Listing for File ptq.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.html b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.html index d536fb5050..1062cc16ed 100644 --- a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.html +++ b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.html @@ -10,7 +10,7 @@ - Program Listing for File torch_tensorrt.h — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Program Listing for File torch_tensorrt.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/structtorch__tensorrt_1_1Device.html b/docs/_cpp_api/structtorch__tensorrt_1_1Device.html index b0b5055f12..9c7609b1aa 100644 --- a/docs/_cpp_api/structtorch__tensorrt_1_1Device.html +++ b/docs/_cpp_api/structtorch__tensorrt_1_1Device.html @@ -10,7 +10,7 @@ - Struct Device — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Struct Device — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_cpp_api/structtorch__tensorrt_1_1GraphInputs.html b/docs/_cpp_api/structtorch__tensorrt_1_1GraphInputs.html index e151e4ab5a..f997950fda 100644 --- a/docs/_cpp_api/structtorch__tensorrt_1_1GraphInputs.html +++ b/docs/_cpp_api/structtorch__tensorrt_1_1GraphInputs.html @@ -10,7 +10,7 @@ - Struct GraphInputs — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Struct GraphInputs — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.html b/docs/_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.html index 12201ef875..007335249b 100644 --- a/docs/_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.html +++ b/docs/_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.html @@ -10,7 +10,7 @@ - Struct CompileSpec — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Struct CompileSpec — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

    diff --git a/docs/_cpp_api/unabridged_orphan.html b/docs/_cpp_api/unabridged_orphan.html index 8125ef370b..78d659f647 100644 --- a/docs/_cpp_api/unabridged_orphan.html +++ b/docs/_cpp_api/unabridged_orphan.html @@ -10,7 +10,7 @@ - Full API — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Full API — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
    @@ -314,6 +314,8 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Engine Caching
  • +
  • Refitting Torch-TensorRT Programs with New Weights
  • Dynamo Frontend

      diff --git a/docs/_downloads/06a1dddfb8c2b5515b697700d863a453/engine_caching_bert_example.ipynb b/docs/_downloads/06a1dddfb8c2b5515b697700d863a453/engine_caching_bert_example.ipynb new file mode 100644 index 0000000000..fe7a070b26 --- /dev/null +++ b/docs/_downloads/06a1dddfb8c2b5515b697700d863a453/engine_caching_bert_example.ipynb @@ -0,0 +1,43 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "\n\n# Engine Caching (BERT)\n\nSmall caching example on BERT.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "import numpy as np\nimport torch\nimport torch_tensorrt\nfrom engine_caching_example import remove_timing_cache\nfrom transformers import BertModel\n\nnp.random.seed(0)\ntorch.manual_seed(0)\n\nmodel = BertModel.from_pretrained(\"bert-base-uncased\", return_dict=False).cuda().eval()\ninputs = [\n torch.randint(0, 2, (1, 14), dtype=torch.int32).to(\"cuda\"),\n torch.randint(0, 2, (1, 14), dtype=torch.int32).to(\"cuda\"),\n]\n\n\ndef compile_bert(iterations=3):\n times = []\n start = torch.cuda.Event(enable_timing=True)\n end = torch.cuda.Event(enable_timing=True)\n\n # The 1st iteration is to measure the compilation time without engine caching\n # The 2nd and 3rd iterations are to measure the compilation time with engine caching.\n # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration.\n # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine.\n for i in range(iterations):\n # remove timing cache and reset dynamo for engine caching messurement\n remove_timing_cache()\n torch._dynamo.reset()\n\n if i == 0:\n cache_built_engines = False\n reuse_cached_engines = False\n else:\n cache_built_engines = True\n reuse_cached_engines = True\n\n start.record()\n compilation_kwargs = {\n \"use_python_runtime\": False,\n \"enabled_precisions\": {torch.float},\n \"truncate_double\": True,\n \"debug\": False,\n \"min_block_size\": 1,\n \"make_refitable\": True,\n \"cache_built_engines\": cache_built_engines,\n \"reuse_cached_engines\": reuse_cached_engines,\n \"engine_cache_dir\": \"/tmp/torch_trt_bert_engine_cache\",\n \"engine_cache_size\": 1 << 30, # 1GB\n }\n optimized_model = torch.compile(\n model,\n backend=\"torch_tensorrt\",\n options=compilation_kwargs,\n )\n optimized_model(*inputs)\n end.record()\n torch.cuda.synchronize()\n times.append(start.elapsed_time(end))\n\n print(\"-----compile bert-----> compilation time:\\n\", times, \"milliseconds\")\n\n\nif __name__ == \"__main__\":\n compile_bert()" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.11.0" + } + }, + "nbformat": 4, + "nbformat_minor": 0 +} \ No newline at end of file diff --git a/docs/_downloads/1c759c0181fe2845e5579cc82e5b7a7a/engine_caching_example.py b/docs/_downloads/1c759c0181fe2845e5579cc82e5b7a7a/engine_caching_example.py new file mode 100644 index 0000000000..5154dc1e2c --- /dev/null +++ b/docs/_downloads/1c759c0181fe2845e5579cc82e5b7a7a/engine_caching_example.py @@ -0,0 +1,288 @@ +""" + +.. _engine_caching_example: + +Engine Caching +======================= + +As model sizes increase, the cost of compilation will as well. With AOT methods +like ``torch.dynamo.compile``, this cost is paid upfront. However if the weights +change, the session ends or you are using JIT methods like ``torch.compile``, as +graphs get invalidated they get re-compiled, this cost will get paid repeatedly. +Engine caching is a way to mitigate this cost by saving constructed engines to disk +and re-using them when possible. This tutorial demonstrates how to use engine caching +with TensorRT in PyTorch. Engine caching can significantly speed up subsequent model +compilations reusing previously built TensorRT engines. + +We'll explore two approaches: + + 1. Using torch_tensorrt.dynamo.compile + 2. Using torch.compile with the TensorRT backend + +The example uses a pre-trained ResNet18 model and shows the +differences between compilation without caching, with caching enabled, +and when reusing cached engines. +""" + +import os +from typing import Dict, Optional + +import numpy as np +import torch +import torch_tensorrt as torch_trt +import torchvision.models as models +from torch_tensorrt.dynamo._defaults import TIMING_CACHE_PATH +from torch_tensorrt.dynamo._engine_cache import BaseEngineCache + +np.random.seed(0) +torch.manual_seed(0) + +model = models.resnet18(pretrained=True).eval().to("cuda") +enabled_precisions = {torch.float} +debug = False +min_block_size = 1 +use_python_runtime = False + + +def remove_timing_cache(path=TIMING_CACHE_PATH): + if os.path.exists(path): + os.remove(path) + + +# %% +# Engine Caching for JIT Compilation +# ---------------------------------- +# +# The primary goal of engine caching is to help speed up JIT workflows. ``torch.compile`` +# provides a great deal of flexibility in model construction which makes it a good +# first tool to try when looking to speed up your workflow. However, historically +# the cost of compilation and in particular recompilation has been a barrier to entry +# for many users. If for some reason a subgraph gets invalidated, that graph is reconstructed +# scratch prior to the addition of engine caching. Now as engines are constructed, with ``cache_built_engines=True``, +# engines are saved to disk tied to a hash of their corresponding PyTorch subgraph. If +# in a subsequent compilation, either as part of this session or a new session, the cache will +# pull the built engine and **refit** the weights which can reduce compilation times by orders of magnitude. +# As such, in order to insert a new engine into the cache (i.e. ``cache_built_engines=True``), +# the engine must be refitable (``make_refittable=True``). See :ref:`refit_engine_example` for more details. + + +def torch_compile(iterations=3): + times = [] + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + + # The 1st iteration is to measure the compilation time without engine caching + # The 2nd and 3rd iterations are to measure the compilation time with engine caching. + # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration. + # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine. + for i in range(iterations): + inputs = [torch.rand((100, 3, 224, 224)).to("cuda")] + # remove timing cache and reset dynamo just for engine caching messurement + remove_timing_cache() + torch._dynamo.reset() + + if i == 0: + cache_built_engines = False + reuse_cached_engines = False + else: + cache_built_engines = True + reuse_cached_engines = True + + start.record() + compiled_model = torch.compile( + model, + backend="tensorrt", + options={ + "use_python_runtime": True, + "enabled_precisions": enabled_precisions, + "debug": debug, + "min_block_size": min_block_size, + "make_refitable": True, + "cache_built_engines": cache_built_engines, + "reuse_cached_engines": reuse_cached_engines, + }, + ) + compiled_model(*inputs) # trigger the compilation + end.record() + torch.cuda.synchronize() + times.append(start.elapsed_time(end)) + + print("----------------torch_compile----------------") + print("disable engine caching, used:", times[0], "ms") + print("enable engine caching to cache engines, used:", times[1], "ms") + print("enable engine caching to reuse engines, used:", times[2], "ms") + + +torch_compile() + +# %% +# Engine Caching for AOT Compilation +# ---------------------------------- +# Similarly to the JIT workflow, AOT workflows can benefit from engine caching. +# As the same architecture or common subgraphs get recompiled, the cache will pull +# previously built engines and refit the weights. + + +def dynamo_compile(iterations=3): + times = [] + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + + example_inputs = (torch.randn((100, 3, 224, 224)).to("cuda"),) + # Mark the dim0 of inputs as dynamic + batch = torch.export.Dim("batch", min=1, max=200) + exp_program = torch.export.export( + model, args=example_inputs, dynamic_shapes={"x": {0: batch}} + ) + + # The 1st iteration is to measure the compilation time without engine caching + # The 2nd and 3rd iterations are to measure the compilation time with engine caching. + # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration. + # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine. + for i in range(iterations): + inputs = [torch.rand((100 + i, 3, 224, 224)).to("cuda")] + remove_timing_cache() # remove timing cache just for engine caching messurement + if i == 0: + cache_built_engines = False + reuse_cached_engines = False + else: + cache_built_engines = True + reuse_cached_engines = True + + start.record() + trt_gm = torch_trt.dynamo.compile( + exp_program, + tuple(inputs), + use_python_runtime=use_python_runtime, + enabled_precisions=enabled_precisions, + debug=debug, + min_block_size=min_block_size, + make_refitable=True, + cache_built_engines=cache_built_engines, + reuse_cached_engines=reuse_cached_engines, + engine_cache_size=1 << 30, # 1GB + ) + # output = trt_gm(*inputs) + end.record() + torch.cuda.synchronize() + times.append(start.elapsed_time(end)) + + print("----------------dynamo_compile----------------") + print("disable engine caching, used:", times[0], "ms") + print("enable engine caching to cache engines, used:", times[1], "ms") + print("enable engine caching to reuse engines, used:", times[2], "ms") + + +dynamo_compile() + +# %% +# Custom Engine Cache +# ---------------------- +# +# By default, the engine cache is stored in the system's temporary directory. Both the cache directory and +# size limit can be customized by passing ``engine_cache_dir`` and ``engine_cache_size``. +# Users can also define their own engine cache implementation by extending the ``BaseEngineCache`` class. +# This allows for remote or shared caching if so desired. +# +# The custom engine cache should implement the following methods: +# - ``save``: Save the engine blob to the cache. +# - ``load``: Load the engine blob from the cache. +# +# The hash provided by the cache systen is a weight agnostic hash of the originating PyTorch subgraph (post lowering). +# The blob contains a serialized engine, calling spec data, and weight map information in the pickle format +# +# Below is an example of a custom engine cache implementation that implents a ``RAMEngineCache``. + + +class RAMEngineCache(BaseEngineCache): + def __init__( + self, + ) -> None: + """ + Constructs a user held engine cache in memory. + """ + self.engine_cache: Dict[str, bytes] = {} + + def save( + self, + hash: str, + blob: bytes, + ): + """ + Insert the engine blob to the cache. + + Args: + hash (str): The hash key to associate with the engine blob. + blob (bytes): The engine blob to be saved. + + Returns: + None + """ + self.engine_cache[hash] = blob + + def load(self, hash: str) -> Optional[bytes]: + """ + Load the engine blob from the cache. + + Args: + hash (str): The hash key of the engine to load. + + Returns: + Optional[bytes]: The engine blob if found, None otherwise. + """ + if hash in self.engine_cache: + return self.engine_cache[hash] + else: + return None + + +def torch_compile_my_cache(iterations=3): + times = [] + engine_cache = RAMEngineCache() + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + + # The 1st iteration is to measure the compilation time without engine caching + # The 2nd and 3rd iterations are to measure the compilation time with engine caching. + # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration. + # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine. + for i in range(iterations): + inputs = [torch.rand((100, 3, 224, 224)).to("cuda")] + # remove timing cache and reset dynamo just for engine caching messurement + remove_timing_cache() + torch._dynamo.reset() + + if i == 0: + cache_built_engines = False + reuse_cached_engines = False + else: + cache_built_engines = True + reuse_cached_engines = True + + start.record() + compiled_model = torch.compile( + model, + backend="tensorrt", + options={ + "use_python_runtime": True, + "enabled_precisions": enabled_precisions, + "debug": debug, + "min_block_size": min_block_size, + "make_refitable": True, + "cache_built_engines": cache_built_engines, + "reuse_cached_engines": reuse_cached_engines, + "custom_engine_cache": engine_cache, + }, + ) + compiled_model(*inputs) # trigger the compilation + end.record() + torch.cuda.synchronize() + times.append(start.elapsed_time(end)) + + print("----------------torch_compile----------------") + print("disable engine caching, used:", times[0], "ms") + print("enable engine caching to cache engines, used:", times[1], "ms") + print("enable engine caching to reuse engines, used:", times[2], "ms") + + +torch_compile_my_cache() diff --git a/docs/_downloads/2a9ac10f2667047a7f398d1593b7ca33/torch_export_gpt2.py b/docs/_downloads/2a9ac10f2667047a7f398d1593b7ca33/torch_export_gpt2.py new file mode 100644 index 0000000000..a26305e4a3 --- /dev/null +++ b/docs/_downloads/2a9ac10f2667047a7f398d1593b7ca33/torch_export_gpt2.py @@ -0,0 +1,86 @@ +""" +.. _torch_export_gpt2: + +Compiling GPT2 using the Torch-TensorRT with dynamo backend +========================================================== + +This interactive script is intended as a sample of the Torch-TensorRT workflow with dynamo backend on a GPT2 model.""" + +# %% +# Imports and Model Definition +# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +import torch +import torch_tensorrt +from transformers import AutoModelForCausalLM, AutoTokenizer +from utils import export_llm, generate + +# %% + +# Define the parameters and initialize the model +MAX_TOKENS = 32 +DEVICE = torch.device("cuda:0") + +# Define the GPT2 model from hugging face +# kv_cache is not supported in Torch-TRT currently. +# CPU is used here so that GPU memory is reserved for TRT compilation. +with torch.no_grad(): + tokenizer = AutoTokenizer.from_pretrained("gpt2") + model = AutoModelForCausalLM.from_pretrained( + "gpt2", + pad_token_id=tokenizer.eos_token_id, + use_cache=False, + attn_implementation="eager", + ).eval() + +# %% +# Tokenize a sample input prompt and get pytorch model outputs +prompt = "I enjoy walking with my cute dog" +model_inputs = tokenizer(prompt, return_tensors="pt") +input_ids = model_inputs["input_ids"] + +# Auto-regressive generation loop for greedy decoding using PyTorch model +# We use a custom generate function which is very similar to the huggingface one. +pyt_gen_tokens = generate(model, input_ids, MAX_TOKENS, tokenizer.eos_token_id) + + +# %% +# Compilation with `Torch-TensorRT` using dynamo backend and generate TensorRT outputs +# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + +# Export the GPT2 model into an ExportedProgram which is input of TRT compilation +gpt2_ep = export_llm(model, input_ids, max_seq_len=1024) +trt_model = torch_tensorrt.dynamo.compile( + gpt2_ep, + inputs=[input_ids], + enabled_precisions={torch.float32}, + truncate_double=True, + device=DEVICE, + disable_tf32=True, +) + +# Auto-regressive generation loop for greedy decoding using TensorRT model +# We use a custom generate function which is very similar to the huggingface one. +# Move inputs to GPU +input_ids = input_ids.to(DEVICE) +trt_gen_tokens = generate(trt_model, input_ids, MAX_TOKENS, tokenizer.eos_token_id) + +# %% +# Decode the output sentences of PyTorch and TensorRT +# ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +print("=============================") +print( + "Pytorch model generated text: ", + tokenizer.decode(pyt_gen_tokens[0], skip_special_tokens=True), +) +print("=============================") +print( + "TensorRT model generated text: ", + tokenizer.decode(trt_gen_tokens[0], skip_special_tokens=True), +) + +# %% +# The output sentences should look like +# ============================= +# Pytorch model generated text: I enjoy walking with my cute dog, but I'm not sure if I'll ever be able to walk with my dog. I'm not sure if I'll ever be able to walk with my +# ============================= +# TensorRT model generated text: I enjoy walking with my cute dog, but I'm not sure if I'll ever be able to walk with my dog. I'm not sure if I'll ever be able to walk with my diff --git a/docs/_downloads/34421db2f2a82ea2b3d9a9cc85624784/torch_export_gpt2.ipynb b/docs/_downloads/34421db2f2a82ea2b3d9a9cc85624784/torch_export_gpt2.ipynb new file mode 100644 index 0000000000..4623ccd105 --- /dev/null +++ b/docs/_downloads/34421db2f2a82ea2b3d9a9cc85624784/torch_export_gpt2.ipynb @@ -0,0 +1,122 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "\n\n# Compiling GPT2 using the Torch-TensorRT with dynamo backend\n\nThis interactive script is intended as a sample of the Torch-TensorRT workflow with dynamo backend on a GPT2 model.\n" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Imports and Model Definition\n\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "import torch\nimport torch_tensorrt\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\nfrom utils import export_llm, generate" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "# Define the parameters and initialize the model\nMAX_TOKENS = 32\nDEVICE = torch.device(\"cuda:0\")\n\n# Define the GPT2 model from hugging face\n# kv_cache is not supported in Torch-TRT currently.\n# CPU is used here so that GPU memory is reserved for TRT compilation.\nwith torch.no_grad():\n tokenizer = AutoTokenizer.from_pretrained(\"gpt2\")\n model = AutoModelForCausalLM.from_pretrained(\n \"gpt2\",\n pad_token_id=tokenizer.eos_token_id,\n use_cache=False,\n attn_implementation=\"eager\",\n ).eval()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "Tokenize a sample input prompt and get pytorch model outputs\n\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "prompt = \"I enjoy walking with my cute dog\"\nmodel_inputs = tokenizer(prompt, return_tensors=\"pt\")\ninput_ids = model_inputs[\"input_ids\"]\n\n# Auto-regressive generation loop for greedy decoding using PyTorch model\n# We use a custom generate function which is very similar to the huggingface one.\npyt_gen_tokens = generate(model, input_ids, MAX_TOKENS, tokenizer.eos_token_id)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Compilation with `Torch-TensorRT` using dynamo backend and generate TensorRT outputs\n\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "# Export the GPT2 model into an ExportedProgram which is input of TRT compilation\ngpt2_ep = export_llm(model, input_ids, max_seq_len=1024)\ntrt_model = torch_tensorrt.dynamo.compile(\n gpt2_ep,\n inputs=[input_ids],\n enabled_precisions={torch.float32},\n truncate_double=True,\n device=DEVICE,\n disable_tf32=True,\n)\n\n# Auto-regressive generation loop for greedy decoding using TensorRT model\n# We use a custom generate function which is very similar to the huggingface one.\n# Move inputs to GPU\ninput_ids = input_ids.to(DEVICE)\ntrt_gen_tokens = generate(trt_model, input_ids, MAX_TOKENS, tokenizer.eos_token_id)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Decode the output sentences of PyTorch and TensorRT\n\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "print(\"=============================\")\nprint(\n \"Pytorch model generated text: \",\n tokenizer.decode(pyt_gen_tokens[0], skip_special_tokens=True),\n)\nprint(\"=============================\")\nprint(\n \"TensorRT model generated text: \",\n tokenizer.decode(trt_gen_tokens[0], skip_special_tokens=True),\n)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "# The output sentences should look like\nPytorch model generated text: I enjoy walking with my cute dog, but I'm not sure if I'll ever be able to walk with my dog. I'm not sure if I'll ever be able to walk with my\n=============================\nTensorRT model generated text: I enjoy walking with my cute dog, but I'm not sure if I'll ever be able to walk with my dog. I'm not sure if I'll ever be able to walk with my\n\n" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.11.0" + } + }, + "nbformat": 4, + "nbformat_minor": 0 +} \ No newline at end of file diff --git a/docs/_downloads/3454ee6d4b68e83cdf0c757f0059986b/engine_caching_example.ipynb b/docs/_downloads/3454ee6d4b68e83cdf0c757f0059986b/engine_caching_example.ipynb new file mode 100644 index 0000000000..5df63748b4 --- /dev/null +++ b/docs/_downloads/3454ee6d4b68e83cdf0c757f0059986b/engine_caching_example.ipynb @@ -0,0 +1,97 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "\n\n# Engine Caching\n\nAs model sizes increase, the cost of compilation will as well. With AOT methods\nlike ``torch.dynamo.compile``, this cost is paid upfront. However if the weights\nchange, the session ends or you are using JIT methods like ``torch.compile``, as\ngraphs get invalidated they get re-compiled, this cost will get paid repeatedly.\nEngine caching is a way to mitigate this cost by saving constructed engines to disk\nand re-using them when possible. This tutorial demonstrates how to use engine caching\nwith TensorRT in PyTorch. Engine caching can significantly speed up subsequent model\ncompilations reusing previously built TensorRT engines.\n\nWe'll explore two approaches:\n\n 1. Using torch_tensorrt.dynamo.compile\n 2. Using torch.compile with the TensorRT backend\n\nThe example uses a pre-trained ResNet18 model and shows the\ndifferences between compilation without caching, with caching enabled,\nand when reusing cached engines.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "import os\nfrom typing import Dict, Optional\n\nimport numpy as np\nimport torch\nimport torch_tensorrt as torch_trt\nimport torchvision.models as models\nfrom torch_tensorrt.dynamo._defaults import TIMING_CACHE_PATH\nfrom torch_tensorrt.dynamo._engine_cache import BaseEngineCache\n\nnp.random.seed(0)\ntorch.manual_seed(0)\n\nmodel = models.resnet18(pretrained=True).eval().to(\"cuda\")\nenabled_precisions = {torch.float}\ndebug = False\nmin_block_size = 1\nuse_python_runtime = False\n\n\ndef remove_timing_cache(path=TIMING_CACHE_PATH):\n if os.path.exists(path):\n os.remove(path)" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Engine Caching for JIT Compilation\n\nThe primary goal of engine caching is to help speed up JIT workflows. ``torch.compile``\nprovides a great deal of flexibility in model construction which makes it a good\nfirst tool to try when looking to speed up your workflow. However, historically\nthe cost of compilation and in particular recompilation has been a barrier to entry\nfor many users. If for some reason a subgraph gets invalidated, that graph is reconstructed\nscratch prior to the addition of engine caching. Now as engines are constructed, with ``cache_built_engines=True``,\nengines are saved to disk tied to a hash of their corresponding PyTorch subgraph. If\nin a subsequent compilation, either as part of this session or a new session, the cache will\npull the built engine and **refit** the weights which can reduce compilation times by orders of magnitude.\nAs such, in order to insert a new engine into the cache (i.e. ``cache_built_engines=True``),\nthe engine must be refitable (``make_refittable=True``). See `refit_engine_example` for more details.\n\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "def torch_compile(iterations=3):\n times = []\n start = torch.cuda.Event(enable_timing=True)\n end = torch.cuda.Event(enable_timing=True)\n\n # The 1st iteration is to measure the compilation time without engine caching\n # The 2nd and 3rd iterations are to measure the compilation time with engine caching.\n # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration.\n # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine.\n for i in range(iterations):\n inputs = [torch.rand((100, 3, 224, 224)).to(\"cuda\")]\n # remove timing cache and reset dynamo just for engine caching messurement\n remove_timing_cache()\n torch._dynamo.reset()\n\n if i == 0:\n cache_built_engines = False\n reuse_cached_engines = False\n else:\n cache_built_engines = True\n reuse_cached_engines = True\n\n start.record()\n compiled_model = torch.compile(\n model,\n backend=\"tensorrt\",\n options={\n \"use_python_runtime\": True,\n \"enabled_precisions\": enabled_precisions,\n \"debug\": debug,\n \"min_block_size\": min_block_size,\n \"make_refitable\": True,\n \"cache_built_engines\": cache_built_engines,\n \"reuse_cached_engines\": reuse_cached_engines,\n },\n )\n compiled_model(*inputs) # trigger the compilation\n end.record()\n torch.cuda.synchronize()\n times.append(start.elapsed_time(end))\n\n print(\"----------------torch_compile----------------\")\n print(\"disable engine caching, used:\", times[0], \"ms\")\n print(\"enable engine caching to cache engines, used:\", times[1], \"ms\")\n print(\"enable engine caching to reuse engines, used:\", times[2], \"ms\")\n\n\ntorch_compile()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Engine Caching for AOT Compilation\nSimilarly to the JIT workflow, AOT workflows can benefit from engine caching.\nAs the same architecture or common subgraphs get recompiled, the cache will pull\npreviously built engines and refit the weights.\n\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "def dynamo_compile(iterations=3):\n times = []\n start = torch.cuda.Event(enable_timing=True)\n end = torch.cuda.Event(enable_timing=True)\n\n example_inputs = (torch.randn((100, 3, 224, 224)).to(\"cuda\"),)\n # Mark the dim0 of inputs as dynamic\n batch = torch.export.Dim(\"batch\", min=1, max=200)\n exp_program = torch.export.export(\n model, args=example_inputs, dynamic_shapes={\"x\": {0: batch}}\n )\n\n # The 1st iteration is to measure the compilation time without engine caching\n # The 2nd and 3rd iterations are to measure the compilation time with engine caching.\n # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration.\n # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine.\n for i in range(iterations):\n inputs = [torch.rand((100 + i, 3, 224, 224)).to(\"cuda\")]\n remove_timing_cache() # remove timing cache just for engine caching messurement\n if i == 0:\n cache_built_engines = False\n reuse_cached_engines = False\n else:\n cache_built_engines = True\n reuse_cached_engines = True\n\n start.record()\n trt_gm = torch_trt.dynamo.compile(\n exp_program,\n tuple(inputs),\n use_python_runtime=use_python_runtime,\n enabled_precisions=enabled_precisions,\n debug=debug,\n min_block_size=min_block_size,\n make_refitable=True,\n cache_built_engines=cache_built_engines,\n reuse_cached_engines=reuse_cached_engines,\n engine_cache_size=1 << 30, # 1GB\n )\n # output = trt_gm(*inputs)\n end.record()\n torch.cuda.synchronize()\n times.append(start.elapsed_time(end))\n\n print(\"----------------dynamo_compile----------------\")\n print(\"disable engine caching, used:\", times[0], \"ms\")\n print(\"enable engine caching to cache engines, used:\", times[1], \"ms\")\n print(\"enable engine caching to reuse engines, used:\", times[2], \"ms\")\n\n\ndynamo_compile()" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Custom Engine Cache\n\nBy default, the engine cache is stored in the system's temporary directory. Both the cache directory and\nsize limit can be customized by passing ``engine_cache_dir`` and ``engine_cache_size``.\nUsers can also define their own engine cache implementation by extending the ``BaseEngineCache`` class.\nThis allows for remote or shared caching if so desired.\n\nThe custom engine cache should implement the following methods:\n - ``save``: Save the engine blob to the cache.\n - ``load``: Load the engine blob from the cache.\n\nThe hash provided by the cache systen is a weight agnostic hash of the originating PyTorch subgraph (post lowering).\nThe blob contains a serialized engine, calling spec data, and weight map information in the pickle format\n\nBelow is an example of a custom engine cache implementation that implents a ``RAMEngineCache``.\n\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "metadata": { + "collapsed": false + }, + "outputs": [], + "source": [ + "class RAMEngineCache(BaseEngineCache):\n def __init__(\n self,\n ) -> None:\n \"\"\"\n Constructs a user held engine cache in memory.\n \"\"\"\n self.engine_cache: Dict[str, bytes] = {}\n\n def save(\n self,\n hash: str,\n blob: bytes,\n ):\n \"\"\"\n Insert the engine blob to the cache.\n\n Args:\n hash (str): The hash key to associate with the engine blob.\n blob (bytes): The engine blob to be saved.\n\n Returns:\n None\n \"\"\"\n self.engine_cache[hash] = blob\n\n def load(self, hash: str) -> Optional[bytes]:\n \"\"\"\n Load the engine blob from the cache.\n\n Args:\n hash (str): The hash key of the engine to load.\n\n Returns:\n Optional[bytes]: The engine blob if found, None otherwise.\n \"\"\"\n if hash in self.engine_cache:\n return self.engine_cache[hash]\n else:\n return None\n\n\ndef torch_compile_my_cache(iterations=3):\n times = []\n engine_cache = RAMEngineCache()\n start = torch.cuda.Event(enable_timing=True)\n end = torch.cuda.Event(enable_timing=True)\n\n # The 1st iteration is to measure the compilation time without engine caching\n # The 2nd and 3rd iterations are to measure the compilation time with engine caching.\n # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration.\n # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine.\n for i in range(iterations):\n inputs = [torch.rand((100, 3, 224, 224)).to(\"cuda\")]\n # remove timing cache and reset dynamo just for engine caching messurement\n remove_timing_cache()\n torch._dynamo.reset()\n\n if i == 0:\n cache_built_engines = False\n reuse_cached_engines = False\n else:\n cache_built_engines = True\n reuse_cached_engines = True\n\n start.record()\n compiled_model = torch.compile(\n model,\n backend=\"tensorrt\",\n options={\n \"use_python_runtime\": True,\n \"enabled_precisions\": enabled_precisions,\n \"debug\": debug,\n \"min_block_size\": min_block_size,\n \"make_refitable\": True,\n \"cache_built_engines\": cache_built_engines,\n \"reuse_cached_engines\": reuse_cached_engines,\n \"custom_engine_cache\": engine_cache,\n },\n )\n compiled_model(*inputs) # trigger the compilation\n end.record()\n torch.cuda.synchronize()\n times.append(start.elapsed_time(end))\n\n print(\"----------------torch_compile----------------\")\n print(\"disable engine caching, used:\", times[0], \"ms\")\n print(\"enable engine caching to cache engines, used:\", times[1], \"ms\")\n print(\"enable engine caching to reuse engines, used:\", times[2], \"ms\")\n\n\ntorch_compile_my_cache()" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.11.0" + } + }, + "nbformat": 4, + "nbformat_minor": 0 +} \ No newline at end of file diff --git a/docs/_downloads/6a6052d9668b2cb8332d349d328e21c1/_rendered_examples_jupyter.zip b/docs/_downloads/6a6052d9668b2cb8332d349d328e21c1/_rendered_examples_jupyter.zip index 9ebd46bac6bea13060449a2ba4e913a101ad86f7..34927a4abd82acfb1dd9f2e4811efe98dc046cd3 100644 GIT binary patch delta 21163 zcmeHPeT-aJb>DRylR%u3IL40cB-gtsnKiTHU4N2QmSC@)#EI=V_8J#F_RQOvH!}~v zd2e{{&8|0Yc+DS0N?Q=pld$?Bm8eur)Kow%RRU4eLMc)!6e*bU2xpicO-`ht%H8wh)d}6W_M~>H2^L|ikP0rh1 z=m#g~)rr7w239*%OHR~M$L%F`#&(*mC@gzrZ_aOb9M_5*-&0PgEKk`NI~eIK+RE?P zf&Rn_l@lq)TlC#UJ5+7UTUMB}(^64jd11p3+VWx?7WET7BUZ4yI_8O-w)3LAfh9T} z*Rk=`Q&G!SZNF~2C3VD8H7ms2OLkz(2TPhr(*K!P~Vf$_Cf~_~( zurc+5YnQ#qSKZK78AqLlOr?inD;>>Y<;AvL$M1L_L@sn4H&TtjZ>z8dM%G%&w(t*_ zVV4GkRQ52(qEiP0EOla8B&a-a3}R^8wU*^MVVm>nov&n&rsX&m&!|kPqS*azRQA%% zQb+p8iduo+ZMMLBVQ^qKoQMoePJD3hy0WM6mHf`s<7?%IuTE```NQs#I>{G>7Of1B zWtaRad-E+QQ`C+8z_DD_@N3{7^afmqoQ&iTJU_3N{NO^v^_SFAi*l!u>DQF&T(I$O z)UqP51j7R(e$TdJvbgL)4Z9+EAy{h())!@>sGmL3TcDV3j=XGwmMCzVO-!0HMYOY| z4&W0EwY{d}VJ9tUekZUO9lslbng!eTAlDdz#e_UgJTUSu)VC%BEvQVrI3Qnq>M0BS z;5n%oqHBKBb6&(MR^J+;HhNN?rj$&wnx`pj`OgKBs&%7&+R4@hyJfizDmFJEOe=)8 zJ)^36z6yziU@P`TtKD(!>O?7-UN@P%|Gn?K(whF-nnYFB#g>*U_o4FU!PpTkUdYOG zBFHw3><)zn-8APR6NDz$f(ny#V}C5=(OTfhANM1kbjTOPQus7KWiPhsb%*u{lXol_ zE*qy;Re3HeB4`p;)dMjkv@zaS)%l=nS8=SUc)elx1zgbT;&^qiKDTBnxdodiLLb+# z75iNmT<&%{KG-~XBw#;4PDvdWp@(KvcEqp~7z)rN=Z9N%U2Is4mM~PC2H8i6oP^zr zNHUEd7E@Mm*j7!S1od(Bq#Yi&qv`h=ODMVr(_8o3P)mve=M1`3Hw1^t{GJ~tOV+mJ z8;`%^N^|cMYm)Eoe>KjXSR+lX#U}zAh8!AS&-;dxGJiTFNV8=tU}&SuLmgn@1|;Xz zaOMs|A3YpoSl`IDVBUMiRW<{CY{#_qJc&FGv`4A=5GSBXq2l7irQ{edU<3}G0)gh`w)qRodaW8+eIjNs-VU(TSF6X)}=JhJPCHCz+NofA~6 zLTg3OTrosdt7Wgz^=hf}G@7=GjmA8`o*t1p#RyZ?+p-6@Y^F`Bd%m1BA3vHL*!%7) z-+uh(YvTH%1s>Pc>CkE#^U4E5f#5iG+Y8t{_)WFLhC;RMcg5gwrTSdK?@8PU8 ztu8b+f}>VT7W|8^Crz2JV2*YOlP2k>chD-uc&THB+A~m419XD1+J9n4WzV((EYpGK zfR{0rSaU^O48a0<&y}0DM}&oAg+U75-xKFxXpt?4(85?72%7gEI@8b!)X}S zZO&{3%VJvQePu2B4tRm_MYUtc0YW&j6saEW*fHR!a7bvetSGYE9Wku%7%>E8sk#GK z8G&1!4rQO0K-o*R0GrB0VV;?ztKB2$4FL+kd`F+GL4ARnfWZf5i!5;;^qVmeAFQ|l zv9LqY;na(e;Oh+*6Yp!-0s9exmm2VCjSrV$mfCuE4fx*$t9F4R2yEK@daBLo?ks^z zFi+Hi80PIXj*8e&QEae}T)ER9`!9Dz+ zOXh@A_VUsRx>_?dx!?lb25luLUTHIooiUgaBoUdA6svr?(S|0BH4&c#|hw z3-;wC{Sa+lwd#144xe!!E^w)w`B5O(vTCs z622pb8HgrwmJ=kaTCl-}tTHK$8Zu4T>0LqACF2{_|r`b8}DOS`0UsIV(*h1 zMn?GkaPsoqo7Ofv(XM2A^QE;AH0A@8;TDeZ47 zuG=nVb}xD1sjtKoO$OMs)OvVE1cx2lZrM9_;5RDs&p&tg_$f83cJC^ChYo+>$lPK4 z0*g{>mFj@$HM>wAt99$v%v5=7f-)Ufhag^H2CmcFl(t{3ZZihZhc$!D`R;{9ePXGl zG^_>&LJJ{jiOMGD5mMK>0iB3TLp!eKPMqfZkaQhK7}^&4SPhQkX?)PO+kOCu7Q)m( z6&CRhkF(6e(DQAsjy#f|{^>)fqm+Z#R3Ar6WN4*_i?B-yR`$qXOju+F4GJeY>uinF z_u&fpKEFCvy&?-%oce6Ga>@3?>~ZdG>`t?0_9@E^?c67D7G4EXcI`HJBDwDm-)C{QzFL*&TPHPY+wpcAikmPN2;bLRLa)}Fskezfi# z@e#oH5Bs!q?gffa)VjUQz{rM$X_m(jAyRXUX%am#3R=)B6(N(T8+esSawQ?i?nJ^< zp$B{kF)BaF+47kAD6WU)v2$9(D5}Z8ZbHJ6I5B>mIThFUJ0fs=XxEoj9SH|9p52f{ zr^Z~?pe~&eR|m!*Vn(!>9Tk-U&83!uB!Ux%3T7HO7^xNHG-#r!;m~L*H1!$h*oJ zcQ#;Cv`h`$$=p^>rt!FeRf7lBVftkyI){Z7KES}x+|xN{XG$Nbvd1G*L3lcA&_1ot zkX^T}isv-G0oBW{^vj zo*Ir9iPH)VLNLmjz3pt#3RXln5 zLxjv4tPTVktPS+?xjgq97(*1zNaaz^LcB7Rb1Pn9)GS zeFK1TEStu_I+6ReLPqFcWx@4p7b+oQ>Dg&B$IVBQ@%4u*eJ;+uLA^`E6^Z;;#w+8u z3rTLz@(HU2JHPTsa)J5qiougvWFIqf&Y`x|VP#R(T-)+0-HryErbrJ2iuf_qk-jn9 zgnrkEL1O@=Avn2Tc~-!qAwYTf_D~b;+g-Nr6g?kZ0iVqoBEo(14V#xI^p4*`w8(e; ze?zn-=0Z39?EA<4dqzh1T|gg=hNn9RD&eS|MageXF7MYBT`7Pue^EGi_~g9clvAiy zyOOBqb5l8Il9X?hl2_mHmE^}Cno7>!@wVxKxzl$|3N%0}a1p^hGBzl|OXffM`j$tN z+V08ZyIUXMo~5Hur;haDn9m7TrUOnrg6ZJ?_v6(E=po}Qo2qH_ z*JLCSWx*%6(NKfpn0pcy>es%7@j|tg& zWKZ7~*wVq!7&A01l&Mym0WZ|77_ggbBF{wXFqbG23fj_yDb8RZ~V?)8|4*Mfx!+dSNQDq@8<%`Ep=qF-U#oo5`!cyR|udE~eqGX)66< z?^QWLCH4#j^}fB`w|!gkQd8ZNPmU+&?=0SxW&)GxJ2xc9?tJhO5?&swqwa_l^^D$D z&OareJ-+#_6!axOX+9>+R(jBo&2Wk#k8>01p~G0kvBRGC4~{{tmV!1}6%>LwvOKmk zYIk;uEve8Y(+n9Y=r=QuP7?SvFcI-BXPQ<|JfU_^6?qt^pFY^*_DlS6ogSM^A5qC4 zyms#f33>WSOn!5D+uM7?k~hY-Cu={kA^GOmW|&JoUO|WJLnN28h2gSSLnWUjth{W+ zik49J#smee0Jdv&Q0<{9w_!Z%6XGbiP@18e_V--}F5hzr^KfOa^q_i{0vh?S! zeCyPI=^Eibo&MHl77PF6bx;0#<&BU1V_|}h^88Wv)2~gBjPQGWHAhJw86VcIZgaYe zjAt4Bav~7aAzp;9LBB=-D{rd1X_sZ**z5iR_IbMK0XxGUIyM3#J<;;9VUK-(s2^*w zXqR0wd#lD(R*m6IWMcfcW1+pF+d;m-i%RMlSYOmcXjf<#P5(g|LZuPpP_|~FL<8Rq zp&DdCF=!}}9A!p`rXWULxQI?RR2JC$8Un+zr2>U5NBU8@0y<3aFt90n z${wwNJ(6YfQraTnf!#r6*RD$$(Xf_hnYI|4v$6&i^lUWoA{tS$&9ei_S`@48qvI;* z)-aJV)8e%1P6%VGjrpYB9jHSG1Pjw0A6!78p``oevPR815DI6*V(6nmHy6Ag{f$2V zDmK)i--B|?06qMSNl2uPajE~te?6!M`WzeRy<^2xWepr*cL5P}7cNAr5CkPG(b2N@l2gCNv3ewa-EVB@`q zfGmsZc{F6RzcTsZ7fZ=I&mKr7cW)UlA&zWVT^F4b2CvQ^Id?IHHeKL9O;ql}Pdw(tYkG=QZF^y+`yMNn< z$vr~YVUU3-%h?rcr@#e=+Yd{8r(|D5vrJgv&*`T$!T4BDE1yXG7NZ(u4j9QJlZa|~ z+~;~amx>fNyp1L*v|*wJ6?H|U{VC@)m~Gi^r>D#LR%5hmrS{8c5e!#fgEwlyNIlkV zJ$=K)Miv~`iP*_$%qh&jKGiftP&X|&u?g`vL|I+udL zA%yk8H_!w!{UApB60Qv3j0E7$k%n+T^a1HYYoMFRVIJv6GNVfS32x~R!F#Tp zSv<53nbvv;4H6|Ea46Fj;#7Ju`ag{L*mu0dZe(LTBz(xtXpUL6%J=Pl6e^JsTmYK< zO#3fg2kbCT8&;;tg&)_ypqtAITe@e=j-3EU8$Bao2X39hNfWr~w#gONOTodMUCfz- zqOxIg!B~(^Ip~?ekJO`K_qsvIJkb75cA;WA9;jd!fh_r>?Woaf3U`O@dEBu>K2E#& zwBFH#qhqg&j)ELL8nCeM4Nw?jB|U8mhCb@n*>(V65O!-V7H^dw^A)m4-_0bKn;NpT z8D6v!-8yI=L9qnAOYFprX{N*{RLp74MN_vn)Rss)9vYi#pLL5vZHctwUPOGmLv4w) z*?g)6ecMBAA|L1QO$oIV>(T>v77Ei-Q^}w0ekd1|j6-di722|?s|4evwNJ{nrdcI&% zh#k@|URi^enZ49|jPQOZB6^j4ad|^>&+GRlAKJe4uJnCI?a5zm->9!h<)e1KftJLT z&V-TgkjDdr=;6bnbdDePTu)%Sv@raoWb_m7ijD26>z10r4qcL(V91?bF%lD;M?+kj zUqHLRXy6=QXXGrwvi_&f_U4|;j}jo0#x(AR4;B|-d;=MbV^x$E-ug@yXiFmtVr?B^ z=O@tSHuWuLJYAQ_$~;G)4&#uYZyL=zSDEj`kO@V2t4_EpKOvE6A0KohfFKttI>BVy9p^H<`Zb)Mjt z;-KT<$#ZQt46^8TpcX`mCEncf#2pmX?xNbYYmfXhF~QzA-M|dZE}G;iDxy{ZWC@p@pFvd_VC~}UuBj=rMpE88mUBzm zY(luraJGLF#k60d61J=kiiDTSV;5&fW%A4opF@4@7IuRW3c z%f9=Q`(NLboWE$`!_@vQS<;{SI0Plk4tdtU#87N7yEc2ZJTNBGCe5$3Hs!d z|Gg32OQypUG1wMUs>oMeaoispG9nMn%=CCZbn? z$}XDqeZcvuRhaS9%rwcIxU7QAOm+nbi5Pzg>Hk=7T9eg(d4)L*v~ELrxir0QwK|aA z#OgH{v|*)rrsU0CF$QR~g-F%oC-_HRbmS782sTbA32~3IUXn8YJmxm{Jo=by>cJUV zLPW#og*47&Dil)qnFMi6KvkQuiYQs^REND-fGRQwx9xIS*#bI4Itk!5zXz6=r6 z$9<=Rbf#PKSxBk#&s1a=9yJ3+3G^QjNiVM=6@xEgzZb7efvEz@ z5_Dszp<|m9!d&F4xCNW-MKzQT-an7~Vr z4X#i!n~H=~)QWNG%&A>KZBBncM0wE0+ERJpUqR8A8-ca5yEuI^%7kM|dD2YEOHdxK z9)iaw4;(uzN;b#XPqb1%s~FjHiMQu-?-m_W{Vx^VYIXK-* z_W{>w+s>32l`eS}R8hV&Xe|kH{)aLH8st#(G@qa!T>pst!Il+jXkwL~5~lPY04mK| z28^c^EkkuSn3zFL3-_Rye7%5b9A?2m^>LvBr&7vR)NI2e(kX``omy#iD+uXks@%Jr zjeQ>2dmZD*4YU49xk%O{OHX(lSFn2SaB$x%_px&ulh(;17AY4J7PSc2B6u z_o0%5Kl#H_&RH#|j4dmre62@4WknyC^}}IIj)luOCONi(OQhIvNRJ)PptKT|k}tB$ zGaOd4;deb$QOQdsf9PQe~33Oyv~ljS8r zA%H^Bm8-sRywYCQC+C`S!}+5IVGo>t6Oi?KwPLA9-cYmFP3hH&SET)KZ)Aq65=|rS zUrJ;a-*~guO^M9zlmqlu2KlWF@<2KA)@G0oYP)UzWm1Y4lGB=*cx4ihH)c5mz+rD? zo__^0&mT&$4kbpA4CnldWS)DjZsN)}PW@c>mc97d^B-Yv+2!x3(GSQEd$x~0uh?AH z|Az|s_YC#dXI}onCj8mLo$}IyueSbh^Tv@8e(y}*z<)pU^3QgEYJ7B4^3dijm)|o! zdXK!__eU=J-adci@(bHX-?fi7wo6WD*;_dA+~}GI?#TX-;s@_P8@}=#{L1f%zOnyS z=1P|vW1~t|>AMHn_vZh6L{>SaulWpLXR!G^o!~Ps`yabXEBfVaZe8P}-Q_dmqmS+%*5B8o<(ofRg_g?? kjgM}T**;2wzVPtqEpm6t*{2Ahb>S=}*MA$=fU_i@M#aqUr%vZ9@=M)uKkpQiKQy$9`Uq;+=6l zE5C?7rAvh!6j5mZoYi+;v_q@+cC$nPxfEm>5q-jeIUoy$89irba~Ii^yJ0&z8YV8=n=;; zXT`~X>A6+S^-6gGL#65~EzGx>b#z3De?|l}0kHa!T zYETLT=r9#QmIR*tgRB)P*;wLB{0xV-$GZfODC^Kc2vfY6Ay;u%_h$C3wncz3t;o!e zz?0x5N!(?;1(&>(iA=&du`mltz@qY&V}xf|7-o_Q4k86#aH%}5VD|9Spin9U6d}x> zfg~mJzGcyjJOzm$0^mnp93yX07!tmTbuO)7#zRbz`v@XYtj%8Y6n%qu5v4*prOuB$ z=I%Z#L3t~bK(ZERAtNgk5CrR_bj1)zF+s%?yM#li%>@y;5_f@d3<<@^f<}G-+(%an z7Y&l^FX*(a|Ktt%FRi;GQ59RO2zOrpO^O>Ycnixi0@hFJ_9qrNLL;7D^reJx2YwSGW-Z8}iVjM(w{zw}#UQb_*y zOf;q+0DHV_?}N!(Kf8ARe_EQ_V}m35+X)kVu;XKUAMDz=Q!@{4d@o=d*r*tEKP`f)Ueo7PmU0&c9^$_KJbN45<-h!X+ zoY?wbw{62Ouf6pb2mbEO2PP)od>~od{X#8ry=ArN2d&QZqT_{raC%Xl3jF23>V;~} zjXLUOXHA`P+~rObmXiP6{p5vj-WHci3x2QfcCE`bV@|1I~j|JRSNBMmx(08m* zxsh_cRlmFHgsNwG>k3QvJ1Pn+FKqikPaX`b`hs3CELon7RU)_NzM=0ctKaXs4z7AC z>co!f`L@%osS}=RSs|8Qa{>o<9Q-1nD!;8P{gmqQidu5ipl@4|W0(9EUIOG+x4W(? zPHnjc7(*uzV!Y&^KB_F+>PJoxP69A2+O>jZN45Mg;<`cXv~UR<_D1Stpky$nFp=^8ASwk0Q3^=`s67Cv z6ZnJWjv==SoVFX0e95WD=eCv3wii2mPjuioOydD@9laxTX1UVZB zZv`wZ2ZJsMc*2W?44*oYuU}1ZAjOseN)))u%UCl-UL>%lj^YkP!SR+|&na~v+Wo*; zb^SpISe6{e)6Bzi)EX*;JPG-1$!ivb!mh>{6OFP_dMr?biqJAZxBO+#eZwhPLtBVw z=tZe6DUcvMcC%I3@q73)h{(Vo04mWGV%_ZmXE9AeS8W3Xm0IZ#5Qce+2sLdS z_EQ#!vFsC~A)=mk!k3+B_6caw9L%DYKp0rt??E=n4yYOg$~G{QoWMiVzm|M||HH}i zpWOY<-3K1KIeFvYrw$mN>UB;94vaNq-cH>Bcs(5pe5YlPd{ux)d_dRsfVv%E^ENo> zwV-^*AetWZ8KyRJESPdw(t>@Va*0)^LIPtlTu4E>8A~pz#XtLT4wbR^hQXkg?R&7K z>q2SLYS0J1K3)qlGl`905U@kGVnh?4GF1}@9u0iv3R+5`smJA#W1ot(_2#C~e1g<$ z3bYNqbi+W^Y+{r4z-y(R(#YH;j4~I>%(7l72hoEkPLWNvlusu9bGIhH*uU?cLHP?e ze+2s+wO1{W+E%ZI)^f7{;c~M5;?~cK3Q=;PV1wQgr~t}S3V~Uu*8KsAO`d6?@}j`+ z3Pzgx@Chv^429ok#JQ@IR$nc^SR|jRmt)GQxh^jV!Dzgp0PAp!@L?jbz{~(IhUh90 z6}N*ONEm<^i1tY&U|12`Mp|p?6&J{>+4)&`!5%~=-C8_X>L8&wnY)9mbgV#puhk;i zyKdO2n%#U4y0dHzAiWX&%9;h2I!!Uv5=hn-P$G{3YhL;o^zXz8Q!x(D13f7@f1F{tH1VmT| zW!Q;24~(gxRu|adSBFbm!Me6Ki@vf}eHZkA&8e!XsiTbD#CW7aIW<)sc3{9?%$OBL zPOmTaRy#aP*)G&O0>S!y3P6zyfF_lNB0tQv3am5M5NSGkFe#f|{;W>de`K;YOPM1Y^Hh3gfjK#UQpl4hX<7kC^7x8ZttG{mz6tl3;y zD_*RrGupUUt@&=7+T#(Z*R5)I@kSPyg4#G8PLVGCi=A_QAN zi^=yiI$`RNr%!b)Seerdh5*8^B_DrcpV+`=)AbsC-XuC*M6TKd$i5;rjMTWI*&FgK zj~mJYLp(~qiGWw5BCnDuqW&AhcWuM}TvufsSa*MTxQp3RJMKKYNWQy2cddSLh#16ND02&j?lB|o|8 zMDqLl%E@ip_rymLGN0t-=lo#78iZE&rIS^~tBXDsMz9l<#NrGhw+r#9? z<98ua%|>?6aw_H4z_#XR%9A5o2(1J&Z5lg+Wn>nxlaL?)v$Cq~Kx(p0)CswU+$?@p zXY5GUTZ4ces!JQL7EZm&=OH&X%mM^B^Z_v9=;v{x=k)vl;a~_O1zlLhGqe$lTqI9t zYsvPfAB>}vh`?rus5KI`0c&Yu!b+I}0_2$KX>9Zxqs-axqP)Sk^a|DN@4jWnkTY484tOzUOGLS|Ri4^7sbRjXSOzCZvD zL0q-u(5MRAbw(^FG!c0^veZm%@}PFD5-mS4(j!_{qSiE!AJ-Y8(U!s{6)9i{(6i1pLg&{T-P;|nK07n_*kt0L-kbd7 z-8-tRzd04r8nc)dYuPfapUp(#^=c{k4R2@srZ!8csaVnBxwrJw5!@yd--hiEklNSj zu=O2nzUtbjnM-cSJhc(E56_t=MSHGF+>o)2k+&=u5$@jrc{`*|`m8f}v3Pgz3$}1b zlQKfpBEO>1oh(IkZ)3%jcM76<)skAz4b8LExngvIbPP!bvKL7K#szDX!`u1+1g=uP zaA98zYoPxU0VtmuGD)dQA@Fg|Mf1uSyj+7s>7B~3E}orvy{f_$x8Fd(+(N~`=%{dd zvRX=aEZ6Jx3t!(e3c@faGqt`F48w%JG7PDck^G$)7@Gg$f1T6OW*RC*_Jt~IQTNvAW69J`SB@z+7gE+TuPE_k`~d^X4;oL#A26W2 zYA>C0Z77_4o;{swk0&n|PKw0LnPlSH14_Nj)rOQM#*_#tAI}5IS|Om!7Q1{%dC!KB zGWm|TC;sqa%CRA3JnS;zO0Rbglf{%{LrUsurad<{rp%?a5K>k)L`BFyZhTC+#BzJX zM3#9nKa}oWI=-x`*2lY!tymXoUM( z_lG|{^7tJS6a0P{LpM6ed~~Fxj;>pj_SH=_Rm!Q4UFJE&zPghvy*3t-D#wnWUYsml zxUln~Qt1`6%Dd86;4Ra$XVH`o)k-MXCU5LJazs}_N480yHf7NOmBCe{^(ft-{V%Nw z6j>im9^C&(vaG#}c}xT}tV~igN&j)uY2`A#&wdBPw0`t66+x z<(gSK@AGhj9Ar+n2-F%<-B*z&KEVf+F{;<7pP`A(!2)EYA({}#O;G{X{e&SBq}!1; zDF%W7CrLQxfCbtXT40x201vcKc}F4~p{i)v;eDEpY);t*x7ydfyd%y~2#o~LMzU2# zsUxi^z@}>mj9HzIbdc*yrE-`7U|L9~v8pQ+<7#6h7=>NlQR-&QB7I@!PFw$_n76yC zURN_(9Mj7JKtm=y8hS?xsT7s7ZTM{9WSNA$;lbq2GkY*x1`2FgiIughSxBDC9u)1U zaZVzzGmL=H@80{Mo~c@S^q~Hwwq{VXKRkZ4v&txIyehC-%w-Kt zbi7{C;^1_xK3Y69*bdC3J&8Dwdz<>;G`~_Kv4oMNNUl7!!Nkpf&MTv^PJ3KM8V;-B zhLAMR4%w|LK~WkFj#4xl)R%Gt-D!wOqpTIj?4a%ZT2`LswlSVHS%05@ZrVFdwU4^_il90HXI&2-xo)!KBabjFcF@wt8% zlPoG)L(`~_>J9xw2$`ks&T?J^x2Qc2uH#(jxSVPAAp-Vb$*8m-Hm5UwVF@NVh$Phni{QE{{#y?nb1be7z%8)1kb#>_j*u#gU;+_YG_@5zB& z@CglbfaY@utL+A`v60WgLiB|Jvq%TqG_(-7md$9v*ag|_cV;fD3MLiiR*y(y#hAjM z4&)allmgAk8YC5NYrsx)I%KMYTMll^w6=vAlpt^!-ufUNfFhtWG=)x(w+Vz%VSwa*6?VlHMe!DYS#he=bYc@r{hC@!tnBa}DZSi#`4Cu-E7|#na&c-tp_`|{=WZ2Qu zM2;I^6HJwaeS`w!Pg*=8;=~(tiQ8BgbVf-#HjUk03C{M+k44#@oHyPtpQ&Vz3oh8)bN0W|u!2u=?3IORahckDG zq`_MP)U#mtqk)aF1k6kTwn3}I0R;UB>7tK?7)XtNNsU&^AI*uXxHb5x%kcosQ$4|I zR_I}rg6WmuAZ^1+vzd>v%aFZc-z3Ib9mmY+%k$dAC7*l$j<~7il|IEr4+VoxH5e1- z;5sliYJH=U@c1f|c5!4kAPL7=3Xp#^3;*zv@R$=A+JkbGo|38={2?Xb6`3g?&B6r^ z(WBJgG6}DZO~R#2EwvaO%JR%Wr7}A+lf1TnZ#^BzM4#4 z-W`{=UpbqnNEK=nd7f-6m8~ahDpOi8Cz+V1C!MHX1$&-hznQE$&Ud`Vv|2K0{m+n% zEnJsh%|xWh$l+_n%*Vs3Ax&v}KFz^p_@&$nhu_sf^{7(8HwpX9b>#1eSPDzcVwPv) zQn3W+W7ZOOUTUqmY^@SbHFEwpbu4-R>9|e~7mUu+3@~9{$Rk&{!0F_t-#W-v#VdM9 z5ZM;+AY+ShDaLs5W_tXOqw^d&_vr8S7{f3Tjqm0vGRAEIKp49g<|i>A#89S(oRMLQ zPKjfa{Tdu3`GsVOE0^S#VFzH`kl5*M^+2$A;$rNPI$m82$Ox2M-K_i_i{!z!T=)9- zCCc9>8BE7b;k7HdwfDr_SJmOFnwxt}{+XO)(^AHk%ac{DG3a<&0m4NP*>-y~44#bd zKzwKB5kw=Umj>kY4tln+_N?2hl;sXYeFxI5`fL@!*SY%K%#2{xyD1>Y>6gV}j=W*&AE{l~X->CV){HWzcm%wWizEXy4hGyzX9c|&53 z*c@OYkvCcSzb=LUlTeatto)CUIp%WdFJ!-$j5*ZDlsgo6jLW>mCFmRE72^x7x|@kU zkfC3u1Mn2fI2$Va^m^&nIOhkM*S&%dtp)JI1CJnvIJhSoV8*>-Jmyg8JemF8j^r`5 zEBRt?Px9>(yBQqZDII{i{G0rlw=cQtQ#-OqFV8_}OGdid#9nAU93|DqTpNM{X%2jo zF3DsalX~&uk(AoQ?`icblIXWGc|0* z761z8Jwy|XEMny0`e*~L%b7eL-lSb=HirZdJ7nk7hHh}dW|Lvat1KGHsyK>*3QKmL z2o)87ZR8UHP{-ual-Qh2VPyit+7MRmO&`!~jubAjeX9%LxYZ(C4Ca&>gDsjw)Zg(N z93xLlQY*OH|5i- z@mNR)f76rnS_Ck*1SW7z&de*GL9I$fK$oNt+rd6zNOnI1Q_C& zoc#9gm~5ZNso@}ZCuAf-M80JZ7!UXf`4|}CkpXs$EN9>zMSz;GLnOF$tk9+nCg-I( zB_g!aeLvtLs_gl}i5qN4{m!t>>;`(nhREec3HFlgE#-%q2!rocP8z-<+9rNffX1aGhY|aKQP0OuQ+TzzT(u2 zH{HDBmh58@U-Q>~^oOT#6p`N_O<%&#i|3C$DVu!tquY|ddVjlYGJJM1fBNG2AHKj% zKBJGzU_5Pfo6+I$^xcc+zxL|oM#YnR-rp%`fA9SVZ@#mU{LA}C1d@BI$-ln0ZDNAo z!(ey%0J}nR$%DW5Kdb7u4@^w(`@VEn!S73PlKkVrBjbke(&sLouYcnb&=D}%;E%ul p`Sw4zCMNieJ+iNh=c|98+#eo%;M9v-Z{^jy`QH{$x^@%(`acN2eOUki delta 1347 zcmZux&ube;6yB8+N>sa6j4C^BoxD^g!Hq53N(gmvC~geJIH8Jd2+75acW1SOS2Nq0 zSqt@|=8{VYqzq*XIrNeq+Y;Q;KcO_xL&>2%6$<$e5-7dp&8%wYM}=nfwDab@@B7}H z_w>n;J3kJMwAP1?R6YB-_}#k?N0yEb;qU(1(PwWs@EE?iQPtNQUk`hpT_?T# z^t=y7j7--6wkQ{JV+%f!B#q&!2(yHm^nTU+`Y7HBL?>kl4Xu}Y%?$#=VTXdG5z`vI zx=hEAr6F;DC~L=BHOl2R4jsaK=&q=mg)W6i($6x=b<#86kB;{{BBTlKs4R$qC?IK? z^cZi$9+6CB3a*Ha70?1UHQ7lhtgYP|1oH)v^^*ySXqNqBAviu z*oujURA7DxVv;13x6zF7I5Im|P-8rzlJbCpNHy!Q&xytq2n!JvLSW?l@3VfjuRxPs z44_z>vxuP%8t@36MWVYlzoaSEOtU>YIO<|SjFKchV4OyHilG7sO~@dgI}li9X)2^v z166IXgT1z^QAb?0MN&>~k!kdtJHQ7zrhx1)#jKNtg;b(wJs?3q6SSQ^)~hSzZj^|w znXEL;waN$PZUxN4%H;mu*loXgxhTDhhxNgPPA+>eVp0`+f$$I%L6%H6%(aW}8NXIH zGjwAAQtd+NWF59*rU1W!BZ2`n+bA*RU>?^_o9@Kw*dY81-D4K+E|}Ys>*h}Tgt=6E zbN|`o7p3=MCDBxJWPk~;56}ZP0&3Li=@|QABjdIX=J#)*UoO|-awf5)*k_Oy4zh57 zP@sTZ=>I?sGy_E?Bgvrl4cHV;fTF`~o-Rug+w-B|NSeD{MLKv1iogv_>PTmOMQy`d z_8lQ9;-pcQ6x4en15Qy0>}iF(vF$_bY1@7p?}xiM{}=u6iWID0y$_A@;XrT$;M3(V zhn+F8ee<|8vUFn98Ta}>pK`|1qk3qKe7oUqI^(B?f5jQCdHGAyj zygv-D4-_|n+~dKm?Gtvq%sMa8{p+h1jvTwr96EUqZk<14gI;i6KP>(BVePo**){W$ z*Dp compilation time:\n", times, "milliseconds") + + +if __name__ == "__main__": + compile_bert() diff --git a/docs/_images/sphx_glr_engine_caching_bert_example_thumb.png b/docs/_images/sphx_glr_engine_caching_bert_example_thumb.png new file mode 100644 index 0000000000000000000000000000000000000000..8a5fed589d17fe6ee62d6a9788519dba471fbd65 GIT binary patch literal 26794 zcmd3N^;cV6&~zU%f(#fBBGMqQ6{mHpumQ z^~%~w0VJvEnR)nlooKF`yb|H}xA|C<^cUE&K&!B{#ar*%RrUI{fOnuSCL$03!--#| zXbLuXBIIR=G00EgrNBb5!{Wg0L?-(}Cu?Haf+0F7%3dlH!I7q$mL$Yecj9<`jSzfT z^D5Zgt0)_Z#Ecth&8aB^q#^tt+jp*g;+l3Kuta%Af{32Sylxr^V1hg&jmnR}mkXfv zAbI2;{|#XPvE=`E0PeA)C<&;HFH;EcKm_ss4Tt~~BwnThPbCg5(+AS2SjXlQXBFG6FYU#Fiz_d=HGDi{5xv>!;RVHaotlES;Y{+4GrXjSnu(wiwX$)Rm{msq2XKT(PDE-xEgi z_(VivIE9l($F=w;5%X502cyd9S;o)?@*@IUt%-#RKi)is3GxWf#e9B1a0p}-K^GkH zE#RcLH-hEomo}wtS-`xST`geg^-vL1i@Xk;hL(*ihFVL8OfGro z7;=ZwLXX%~LDV&-I)%f^Zc@iBUPr)DOdL&SY8Fe4jRr0M^?OY#0}B$nzJ&L1In9Al zQOK9T4G@xxBY?1&=&}M<$uJ+xfTKBW>PRAN(VR;B>we1G_F=pv`)g>`Z^uYRx_jEF zLgunT}hWZL=J)o}-tSc-H43?kvu z2VGc#wT9ESOGB&rE@vy9Zz3|&#)3Ae*}ua?Sht~cbom|GLpPlTSm&X z8g90~9$!qLFeLh!xc`P0$(R$Ja&h&6EezoO__ACMkpyr*-p^zc6Sxuga|EOJ~>i$53|i zk3uPR?4WZH-H$~ya%J$=ipQkS-owRa4tw(P{%YPDc|taynH>AT=I|^o0}49wU~C{J zauK2=H85m@3Fzliq*hcd5LX*{uUE#$CpBIVU20AE#`cp-cs73fa2V=^W|F4HJO04v zh?+`+`1w}s-C)?|2Q&Bi)%tp}=My{T_L}AL?X!5%*(mp1+HX=ICa0VpGHL53#n?|w zrsVN!1h2aM0663#mGa+wr{K=E&1QL;2u+|ml4!Vcv|H~i8eectf>jNDUkq^angTdW z)#m!O%sOuC%7f^!Qht!qCw#tfe81zkyyL#c-mQq;u_=hU8Y&X0P3ZX%SuC=j%cBp_ zr)IOhDL^juo$ys}!om$%Q`tjo6J*I-hLpYq;KY(EwP$7&><+R6Cs-uf#l&-#?&}Uc z2}fsB6#K32O!O58@ksbMFteP}G2(2lnU(Xx0EQ{|6Fyl`z1ErflfT{pGkLUjTO->; z)k=c&e_nqG{-=f;xxac&YcIobydWM*>-WCPDRj}<`M?J$>U-1W=e@wPjOqn`WB6h3 zK~jXSxO+anw%MV<(bYq5#QL@?&$9}E9mOt!yYPhfLyOPk<($)-DALZ(Z<${Ys2e!I zEf@cqbht`&4b(^c=S@?tyU^-%7nsg>47yyUusdaMN|9?WoX_ry|GMNc@rE*&1ZWvhRi z_S3m-n`bk(@O+$m8Kp&CbTqx#!3Ro^Y+#WD0wD>iyApzndvKsR&ejCNHW0D$L?Y{& z-+!(ex(HR2t+z8ezJ13p%e#_pEzeK4BMasjYdnwJ?s{s>RYi;(GFI&ez1i>3C&Ca}f(#daZAeX;DyOc))i?10TL3cyU9coSu% z)362xbDc4*fSbB%;E}Qhj(t3Y;h0y44|>PJ{Wnh8tcJHxpgfb*cWZ`3JyR%I1eX@D zhJN6ES7k-g^nBRA6+Lz{hr;f=v9U%{mDsRr?D`_~Bl#>!q2hGpQrdVLN8GXyT@?Oz z+7-##Pu6db%j-<<)!r^-I*w+`UXXo=8fYr1;~kdxDdbuG7#=k3gnwVoy*Pse*8?CP{EM4h zidh=Lh5n)*Bd=X5LAxl_{94L7yp5$zMl^#3{EGwW`CdAsJZjClo-A{sUr5=r94uSOqPi$Bx2XqwJ0Xj>oqW`wvg3e07J z)B$YP_^e2>9eu-`H$Ilmv=_G3H|}OW)AT=Y3o&4{OAN>c@maT9IRH=16i5c13XIbq z?_;?AkC)0lxBghxk!|c14fTiYk`u>n>GBPA8g4GyLuofED1?CZ_ogTPNTn=*7Rz>` ztYLiEcuec^&_!PYFq(!2r=O|z#Ft@!biey30#@bucj&Ux#$iwD(4dukXbV1&1KhNt zqjE7f0?G2qOI$YC4rS^$^olhr+rM>-n6e_o=%}lSws$wIa?!a=?|&NpC0eU;zsWc{ zKWScWoG(_FP6kwvxFXl0Buq*6*?NA&t1N%CNhYQxAHE>%%=x@2fFKMkfr^ZzsGxgw zHOYm3bg4i(w(q;Sv}7U`6Pvh6Lm7Ptpx?d8)x>^1RQ1|vuG zD|M?Wx8nh#9<4K3^p^ z+tzr;24w)=wRU5Hn4~bhvixzHRwBbhqo*XVUm->z)SC1la7x*{$-z(_ak+M#+ZP-g zeJQP$k`X`41*d?oW^%4JZH z`2Ld2K6XjjT~MILbr{_xL>OfMd9;zSl*--#DwbWDM4Mp%J~N_%)nv|~hmH9@ z)4zT59opehPLJ^KJ6=VVO6Ho)jzT3gR98(NX*61GGVm56{>rj2M18)d^HhPG&iNrA zw3~jo&)!0v{GJE>iXI;ylLdla?z?dM+T(hzjWMg#Y7nngVIY8XR3%u9|L|1IzJP2% ztaEj&!l5_dWLvR#((N&`TQ%Z_aD4=0dIY1oPwr0rTMp@~krDj&qrOo>N7s>&+3e2# zj!pO%Q}P>}{K5VvO(@IX)|C%9o-~d!1M>GPa$u9}~ znp>y9i{nXjS1MjmdxDO!=)mKBA}-}mBp?edtY>D#G4~`p;ncc6ZCuu^9162{V`}LD z1V6Z{i6Rj#_U_ZbKJm)mqBY>vgDo5E5hb5`g#UiRW{W{;Yv3BoGE+jQunVS36Np2enN*G1C~E`p6Rz z{j*O=Hr16UEyfQtZjl0gr^mQuLw~P`!;^`G!)B$eL?HKwuJHg*Av%bEfp+&rI#S6h3Du-ZF@PZN*czynUs7CxrIzi<4CKW$DVUDZp4t@iHd(8;)$gN|7tcwXZ*}bVfvjelr52^Z_wbOI56$NP(NgU%iVHE zF4%~Q^N)t*gCUrPDQc6o=jm9zod2D$s=kAl9(4KKJVtQbESUEoxkjYn)gw9Hf$ zxkomp*C}$oszKqQn1bcxZ?n$jcvSPa-%+!QHfnAo#hP0@4Eu7#;_sn^YAFm}a5$kj zR>ve*17@Gze8vcL$w+E@`m5(=Qy*Q~>OI;1*l|C1v1PMWOw3eL9{)(8=9b%IM}6!U zMcHwGU&b-gd%E#WWQ1W&us{G=3v*Gczbd>nU~XA+RYmPvAK8xKIjItpke-|6DHLM? zV^m(Trt~H3n8Djvl-rxNw5%%KO97+CBR|c6-1xk!_MSCj{y-=Tl)@CO+~o2vmxWhz zY)tX$O0oEU#QKS|IP#9wssiTL8{ zJD#*5d^it313T$)w|h}CIq*lseFB}(#~)6PjENhpg?^8Xm3F>+oFgAQ0$j`746Z8! z4T=vRkY=AQ@V*?NvGxTdbw)Lbpeb#Hwn$jEkRl=kfxy_mFdnIZZ|v?Bk47e(MA6iu z=n4|Yw5OvgD)#Z`yIX8TL)kRf#j_X&X!Xre9^FbrGeT&ua~Ez z;L~rWd6P4b4BJHLCtjp5IQbK@Yn5nt(u(A%E*th>PWT9zmg#$i405 z1u)h5>|%j;2S6YL!d+3OTS>AcGo{_X_PRbtC^cZ~OZKy980j48DF6mKe0`ShOy)%)$Gpx&xYS;yMa=?0Luop+4>{HHg<+3oxDAZ>#%N{Z}PF z*y2V1MAwfF%&zQdI+R~N;zar?o(NK19$nQ&-1jXxMS(~iDLANz^Z7LflHO0Vw(S<{ z-&6fR1v zwYJ=7+&AQJkPZmGr9o>Y;3Dk51T1!Wny3eOhpSrVHP)r8j)n->0clR9O!IoX`MucX z^Zwiy9^o(E6WZ-Mwy{?YbRia8Hzm{^kdB^~ULyM{?&zV&+;rnvab{}4=iyZAU08O5 z@MV3_6vpe|{fU%r_^adQ-fn4Eqe6TqJ&VNr&47I)qv~~M_?mkg8<^CFz(S?YiNc&J zF8sSlt6VZ89ZVI=m8J2(?|e_Ih#V2_^euv=$5qy@pDZ&CU`?d!{!Xjfsm*EywtUo{ z7M`bz>Lm*;&#qzeUCS&pM&I|_VXXi>@ekIJ%6OP-srY6;NEm+rQ*Vm2SP=dpg`9v=8of0 zhOkW8VtL8ieYY31ECO7=oqO;@Z;1Kzm!dv@X`5}nJ?l8#I_`)#C(=D2%+*Mp=89va z5suRsE0+)2#jpPQZ^r2U4e8Tn0Artnx(_PB{~E_^f3atL)FN~J89#@h6>YEJ)SkOU zkVI-6t3OWeJfD)pik~F|nR+pV9ikPTcqm+4>0E|mxI>`U=jt6gulFW~I2NBxSomP5IND~h{YMxG>1blL@4eF^{6 zk#UK;tX9dPQL#jH$V)FC!y~DdwEKH`o^W&`SzLFoKY%L61$Od-y6mNtoA!97 zb$UD;2B^}_laB+luIKy^Uxf-;n)u=5i3a`beLsWhyI-)zg6gb&zCf*CjB)ffyJji% zQ#_YU?1e#Jdx!n7Hh(Ev_|i!PMx4H_`%&raZn?l8`iT!XKN5a0UHxpS!xGFhTNf+u zxpL3g?fDyTpaY9;=4WvgLgnhu2&s^lz|K2HyQC6^rpuT~EyYofcy)zpZBfmg;sx&O zRGn#3W%*3Vl+k#uA^Ks&Nc5APO_6`soi1jM35BHy;tl(LIhwN&a^a`% zJuR6vsJ;uQlY9qsuOj!Xa9BQ;(SH{l%NmW8PkWZ6_$5iF;!BonGJlTSxyBnademO^ z=BG6}B=bz`T7Vp#E}gK#fA=$pv=-a?p8toJvm3oLf8KYg@E~;oi^ZJM)2> z@1e%vtsSw=PX_(a0OjVZhHr*B{g*Z;W~=k%vO2VPhKypEAzx81pO<(tom_IjrLqcK z@zif`H$2E)oZV2FE-3#8Buc{2P?839y z#~{t0*VWy@2eH+)q@uti{*C6u-6YLDb<8RHk){z#wL@!tX{(Ia1#tU zO(0i6^t)Ueq*UJ0(u~F1x&}Xc`0a@7^&Z~y#d7ER`@Xo6>+bhoC<@-^DTsWJz$8O@ zE6q6A|0WUD9M91pzApN8$^s8ZUZ_$??^mFi= z)P@l=?Y7Lm!mt}6Km%zR(K8vzVwM*Cph$R+Xh-GcA2iO3omIER78s5{b=pAhf9(?R z=jTLCaD?TmqKr8OI!4VC|q!DCP=dfL9nazg#oyW^g{NeK5d*yo|ECkDoA zntQ_k_T^y$F%++7ZZFRcq6{D9EFXT)`{k$vq&&@tPha9Z|9HA1$islU8DzR5%*4N_ z8DoCrp?%1ld;gLWA)zb^F16gFV+?4T=W$cQrr1dXp6_NT3tcXsDig~m+Mf7k+r5{4 zaX`hIACv4>RGTPbo0)OUixoJ^mShV!I=(iQq{0#r8*6Kg8efTj75)X$kJM+F@I}UK zUT6J1kRe_tI*P!@&+7A20@t?a1+@I_mNWPu$Azzu(L~LJHNe5}vc>A0`ZhB2cvGWA zAP_fmED`+(u^vqY)Gl~k>@lkPE7s{N6Gk9+#3TqD@!RmIUBVFrL?mpY%{jeTz>Ul# z^|pRz2~Co5>*5Vwzu~?e8!^P_RU&lWU@N*MZCSemvk~1wkWh+dwBQK1rWMVqOV=%v zyWG(np@yt5?UDMjtFamAdo2)+Vqc~!LSR~pYD@3!kZumq_zz#BBQJsnet2!E?A{p`ONN@Od?Dfmf(G#Pe1^hkYDJ<6WdPOs0PO-=ay0Yv^)oC(`N1#}Xo#K?6jVUH7&Y zt>ZCkQuHIG`z{jqel&L+J&lkI(N??MwcMzEWwuRE z#tsWN%OvjLGsAyYLoMu0+RDt)d<{d{D_#C6oX)J(P(ujhKG$ds$ObrB+81&6HEUZK z6)Ny?G2BiOc@i>AX>0zY@kpJT=631_c&}0lHb_-=K+_#y@r3`Ov!QI~A~_f2I%EKS zJ7emxmhe7MD5eWcXU=kDb2?}pJ=Z2ixgDeUC)Cr0Dgk+Xip@@Z?fW>sIrSfsd3iKl zA^E2N1so#Rx-^RoLChMgcI*of<7Fc`lfSl9eUto$^I#>M(!@{8#<(NM6gWSTnNcQu z#F(n@U$X%u^XVuNW%;9#eG{+J3*S0eDbzOK*JU^!PFZiL;WG-ghu)JglFs-`?gx{0 zg^83973w0~@wfomO4tTCaGoAw5ySDvK+M&*B-%3F6upN6405V@Led1c)Ncz<5n^(< zPiF~QP0i%xt-8+Ans2bE6p^{j(LYwh4VMN%1BUMdM~=lxfRktQVj7Tbz4GRfJ5l**g7#AP{Ll zg26O4>0syPBWE01LOh{Vto8R|+hzwxjMgaa+4w4znI|@APC}4r- z_j-3kt^W!$v8L)O)(h**BQ`G7xTp5~*cIxL$Cbd!4BDqq1iG#K=~9AS9TqsXP_zV8f-mKeyL`)N8UjxE6ah@%CHBRjiYY zIAN$UE$g3x=1lf{!xt1ZRgVcz_({~_;ON4pN2KG(x>7`@%V6tAaX}iC*iEyNuh{Z< zP4>B7WT23ha(YzwOadZ+-><%ECjXix zzBw)Eg|m*luIPw7+s=KKGbMdlsp~TSU^^P?aC>~6uI9b%gV=r|dD=H2?R4lgxOouS z4y$=lE(0JSD=3f=cism*huq^ph>fl#eruAl&Qgel9&OAcjRk)uVWD_8t)s7{ATQEaE!nr^~>g=%(4pS8+Bu z5=G_(yS2XmS6Y<<>th8PL}|H=vh_+`rF++VgzfHwF?)POOi$iLaTu-X)KFZf9*u?T z?)}Bqv}%JriAp(#fdiklAyZx}iL7=i?Yhzm53*70oyBuUiOR%j>5$SfU;Qg4%7qnxL^X zxI~FR3wjB?`nEIEh!feQ?EB5fxl_Xbb9SubMyRuRRP*Q$EOTP&drs zY?gCRJmNK?Yq1V9fe#~p=K6_U2~;O55mu0P&seC38CLy+tv~Wp-x`q#x&>ZM7-Q3m zW4eD3)V})uA3+zGj}#kgL19Tb|B00v_HN*#OGe|6oD8pH=^tCuMly2~^7SO4!t)?g zy6@oxDL?L3C~3=371z7zURp$Z`m{P~37?4aXWtd&fX+CBi)^dH8Pk?Ah@fkZ()!`) z1X0mZu{ka}gkauYDfc*`#Qvobr##rtiB{U4PY809iCvoe)JrGm!V8g_?{9}sjEN0= zJ7$GABFPts7amvF?Wxg3FuLN{L~fQO2Ji_C{H?Zr zfx1Mn>%xiT^c%1j?7^kj&;Ai(5oLtxTKUAKq*cME;>S!-zWFU`$Q5Qp+Y6LFt-d-* zI|Yi&1D5sozyVt6Io*3Y(QbMUH6Zt&!WxBbM#)2%rL4sK&J$q7R-gKw`9Ij&^fvO1 zmLCl&t)>E;}-txS})rj3hexI>&3B9Lw&HJr;em;dp7)6hFSx)BpE zDm*>V;wiw)K?Qc**$GE$ELU6tzKjmkL)s zu`UM#;VDh!tL`#Wf4#+=lYGrdZfsUx$JbO|r4oTwXfXkrg4`=3V%Izv8Kj(cTDGhf z58eQ=^xq<-^r*F|u>!{;_Em2^J)jAy{&s94JOl#v*YHux@HFrJNuq_KMj{cd4B!GI zyQy%L+U#GslYhy*L&uRMJLVUA4D-9ylSREQjKgNu|8-2>jqtiW+McHya%Ket_zkuj zm3$1|$Ks(pmrP4EkMi}jCyvG+`NH5bpZ;|q0Y&){coHggK%;uf=ygA9Gjg@-z<06? z?Ix4=xt4ntSo@+9oy6cjU5elzD4ubWSy9ggSr6AII%4s~kygw*$EfUf zCY$#?+H2MPU7Dk3nKJ$X=vT#P&+45%*v>a)aJH8>pmXiNa08~ z#uSsyGl@a`R z?7#3vi8E>bKryMKl1od9W2}*q@jjedoHB;K~H? z;h{>mbHAOQ?yR})5R|r$vm|$Ig$ZnnL#t*8KK5}E+@@Kv8^(9yG?EfT5PU1;lin{4 zLSUjS%c8>N;~$k2tlzG=zm*BM6jW_WNrM*T)#Gz|3{UBi_BFO`gWIeuu7`n7qJ7!+ zoB49j8dB$kFYl%CkKkQAyfMq60G`QsWATY0(n^Qq0MqRiwat%l{@qJH_O!Wp|2ASV zp~0AbnKEINW-+Zod_NkA)8QUU3OqKkv6{WUv7umiWmN6w4y`-SB|;BV;-wz#^2nX) z8^1f-0CkHspK8Awr2)AoVa;dY`-(cwmj-#a^qWnY@txS1{hgb=G>2m7{0c%!)H zbs=y!>P=MRpDu#0WZhGe0hmuS$N;@)mS|*4E~72mu{YNq`RZI2l38VG; z&9cBZbIPwX$TSBPKN?k&W?cZIRHxaR)9Ch}tUXnw4yeb9w9Y0HH=?rzg6qU0_kXX& zikwQyBcGjclOqS5tHEx2(=0Eyz4ua<7wa614t|PXbnX@1^gG5j4}b>F?#wG?KKQUL z>D3%@JC#ak1x48sS1KpB=kC_yJ!7{-np@4coMInIt48ux_LJd7=zo%8N)@-)@Mqn? zCB{TDMLie03HEs~sahW1-369Q7k;$b8PLlKc|e60hl^hIV7!27_os@JOqQ{&^m@It zOU+BY&Kd1I<4B8XNEOGQ1fxpfLKZXxc`gJZVp(4YLUrlS#O3!F;mrX!*VfOSO6<2k839IE#ij3B6`Z33p3V1pDGW)z||*)hzf4| z#5U$te~Ik~F3J|qKe`0sb25<>Dt`sQv!b)&q;?!Q=CJ{$~>URp+ihe!Y>>%EQdotYO8c58TuoTLmkppGqebiz(tbA&68SL)Ym2PBh`wyUU1P7^o-V&xxQ%&I z1`AGU3bH%$YNuf^IXQ&0!Tdb$I7(3)vaau0-GLq!gUc*E{8a*|5H8aRMZy`Da}U;Q zk0R&2Y^ra(MQ4s*7$;THElm5vNno7%&0ev8lfW&rjt8_-k;O9IROA2H67|DhXL*}G z{w0f8xhpks8E$F1j1p#xIWiXZjBfJ5^IaMxafGqW`53v4sVk*MQwPMFnyS<{K}hw@ zd!wvxfN`sRTTVb-x;RGfyoA$6%2mq}{0gGHfQRm0M9P^IUVT{HFSh1`bO^0`(macn znI;iM@Bp<>s_2kCwF=JY z&PSrL)N>3YjsOL_gI3lipNJ2XrOul1Wkw+f?&ac$$3qPNi@V_?<6r^H*bU;p$q=>> z1HvZjJge#_N_vj=kpT^tx;Xz>wFlov3AYhs1rZwWA82QuRk~@988N#nN1&YOzQHa3 zdf&p06k#ycU~rClDyg|%s8er0F|}lCP@)Z{v-VzuK5T>5<=`+i$=M^zg~GidW{KUn8}&u^k7nksXyJs5{jRl`tHwI4aOs5*?ya1z}BK`rSbW4Osuwk zzP!7-44zPQ)BpsgLeCxhiNT!Za_>?f*CXJ;7qQ8#<^K=Qzj13VkbE1||LKI?N#Dgz zIi7fmv}!*%-MEm8-?F^f%-T%3{*>l+mq_<~HYo0Jb>M|QdT}F~I?YSYm@dv}bQ}pW zFz@r0I@tf#7ANC?Q{`j!@F@?F3h&0m;LY#5vfPc>xE`I4U`@X&vyP-}ERVw+47+d* z8XbB2sGh5Hd+{W*%UR)T5f#vG^F8`sWtR@~8)B#C?QQLfjP@T}EoZ+c4rjM1ZjGme zhKp}sp}+!o1JpBW<=)Fa2TtF_-3=QXVIC2VH8rTfxH&^ZfK)EC-4LZ7vc)gt$j;&B zE7W-upbGnrXAHR~IRB-K{$r25*hy70%I93yIb>rw9p~$V7vkmchY;zmv47Gi0=ZcH zG-@I9F(-ru^j;GAl76&@L6#uC9|fE$kv7K>zz5BI7}0z>OgtK##ulGp-w47^V_Q0# zHuCjpJ4->8o$~7oo7i0@Y29LCCw36}NAQbYir!Nzr3tMo>~3U^4m{^j8Rl_h?YAiE zyhGp#8%ummkQ_Q>=-qTpeH?NTNV6{LaTjc#b_*38J~BuhbgN0&^W0t-1FuH~#hhC1 zMn{OGvG&h*`xt>KB~YjZM>i)`Nsmwxb)F$$rXEa__@Kz+APMRe*11R#?(&0!_hWIB<((OfR`TY$wQ>BK1O&x$6u zr8b-zG`lss-&MNJ83H{XmrzrxLCXP3zx616s_fK!t~=CSlew@PbyMR61D!a| zsLmv7%{S${Gp_IQYMXWX0v;SV?H^Pci3At^G0=Y~46@dK00jSvyO&zMzs`y%Yr2nYu6#9Y z*YTghKFZCC2=D}K1OtgE7My;a;3 zsO4r0%Bicd7q4#W5 zm+Xvgm|hj6LwVVX7t5zeMTOVcpsaPSw)$>PxMkvdb`kyB?G;aGW z#kZjiTUtq)U|p}b)@9@V#PZn`{gD1Td0g|}}0wF!xU=e0hy zHyG}uiLnRu=sn7Bh;pjoG`}Fdv!-#0vnYegTi{ni^fepVyOqj1R+r>Tla9W>4-K)UAhR`x5(nWp0izSdW$kHF1DHoWrOr2;ljn~@&VXv z!;to}?n^_JyK~TjZwrA8#KSX8dc!BRrSBlPOK7p;OZB1NX&jQUu>3+zhyr%DF>>dQ*Vj_VXb;zX( zIo_JSRUrlqS!=a;rg_c(E}I>bRiUcdPKf);b;qbq#k^^jYshOCbOAWZUqIUnQ9fUKBEK$&)hXv1&qdgv{1eugEQ%@G{a2OG^*wXuG3b~$Zv(2H3pToh28)cr zYev{el;D$^5k5A9V}#~b=H}vTS^YGz0)O@*21Z`O0#~2 zm-{BE`O_B3P}xZhi&+@oFrY>(h+qjmQno+gN&jLJZ+iW2Lp5>J!M6k+iEQybApKr0+#up4UogAeoM;JMHEB#{VVG0Br(~t z;o<{@RWdi)*lYX)(Ellg;`^QWQ~uKqQPcgOc8HPgMHiIdN%#N6MK-MwOPxW=U$AsA z6hD+|C?>Sb{SdlDXTqSEA{YPm#r(u&eVOc1!~;E`A@Ii?TO5jMV6eX7U#3^=(KbJw z)X$50!WpsTP5Ivi;dG`1(Q1Mn-iO$?#VW{dUm|NOB$ja@QE>~Ph#-mNJ6c1`0#bL8 zb^;qmJmr$E@2Yme^?T-@OSlKhpe$oGY@`wSL5|#*;yAw8FZzoULOkIY8Pjn%67i)} z{l}6W>Jt9?%jS@lTQ|d>jTNm#PJF*w6sHB{j<*k-*5Fdj_2a$#24~eH=jC%yHg2!H z2%3x{t~SIdKK>vjhfqvt_1%^QjO=(dj%>Ar?0bGs(A!D$zPa)z8Xsv49exN|7Z+O6 zQRUPUKl+X}fXYb+dctaPfkRo?R%^7-e7>~5?UUu~$KcRYt<>l?;Xf7hVW&f)4q zKa7KyF0KN+?^GbTpU)om1euxNw5TZ^rg5oVeTlQIZ`^tJb~Uv$0s|b;q%9c&;>51e z`c$)xs>`*eJ`x%qvrd4^9+drc4!I@NZrA_@HNnBsw{F1nVkF8z@ec`yiSEx14LDqy z(CcM#uu}3QSi0JRtqHAV&yzQlJ+KM6c}f*i3M;DrmHQUjI^~nLX$wx^x;4UaN$FVt z&Pmm5(XnW;jlm;3k}hAJAQFW@KVom}CA#H78GhaYQ}k-~GuPz3;-Eyv;kR5brwy=B zfR_?_yPB`{N!wjgu_j5w7@JOz2OZGu;OL@xz{Q;*pimlR!y=pCB{#+t5AL z32+inpvA+yi-KRXMGWNP^XPZAp}vroYP!sQG4E;+EVla%vHfn| z#|EaX2j^1(GZJzEND26^KBVP~l=+=c^IXE0_g9ivk#n7S?_xastD0^LpRc#-hdV_0 z{hMS2L%wspFLfHM`AppRfmS04pWmTbkzSqy*eo0l65!CmA7^`)_Abq90Nw2}UpR@b zjs1LC*2a%m*M;*QVbo&1a^QSx7#|CBpliq@NG4%{O(l;%KbSXgE&km>;pUK}O#LY7 z=hKdHY$ZPGW{US;_!1p?0+d#mw!Ifo*>HlFQT4NYKK3Xx=%8w}&z^)%lHyL5sH{iV zQlF5F9w7Upp0rr!ed-P?%I&>cfx#jb4Y6>zdKeJ+1*oh}oS%=)jNbSgBl(?h{7G(h z3mkpT=d=<%WB8h)flTzOV9O;(hB8mL9kMwruOktKJSwJobi9HGFIkOasl_Se(vrzI z%21W_{aWwfkgZ?Y-lndle*~}utqX+WJJEhlYJk+i+X|(+ms~(05=ByP)FJUvJ{*@t zA0UuB_vEcydLk(s6&v~4(`h))Cc`&Un_zxPiYeT<0))7e6|JO%vqVKb!nHdzRPQar^O`$jL}9z zcyBLNT;7G_UWPng87D7{$!@~X$jtVVDcw-^!O--QX8KHDsde0RJpzB=Dl8CGTWVX; z@z}4;-=5e^MKO4ZMW5ZX8QNt)v*|Dr=8Jk=B(Ie^;aW$=do+aqxqxHb|3JavGWra5 zNAiX(4t5fh7us?>RPI#)^Zbo>R|As>Q{=BSGd7fKNt-!1Ld6HSL-vHePI$O@TM*k}5G zq^A`yH&o-#vlckF*uR?APTJx>E?Zt)SPk|+!z72N4GVeicTyS;X^jnzEuc?(yP3P> zlN08Op$ef~naLYyYId-keedgm&K9%#VT>(izue;kL43eqt%)>!v%o88*#7LwDY|8_^$tgu9%gq<`i_mHPkDUVII;Qtw_RB+^7)5mC zCO)~}!GYsE*TkmRB3s_G>oXw8BYs`AY5sg;!geCmo1pHIex)aefr2!DvA#Du%ja7g zDT(m(jrA81Ja0NuFR4?npk#SW=Tgxj>f>|w4$m(>3*rxA$Nt%CR|a@Y`UTlmzk=`{ zlJa_LK&$aMpOzv$X4@i{)S2$*P^*VjVyi`d>`~!zR2>ac!W2?GYJ3BA%brp-`1T-|#o!YMKf7{>+eEb0 z!+5L{^jYlE?$bB4;fJkXuVVT8R&evJ-b|1^3j?y}_DNB@Qmhb~hIpqim9aAH5+>c)Y#7rpKts zZ(aG`MU5}{u>`^kuxaA4s219GVYtPiXZkO2mN)&g2(3*85 zekL#c#|vwqsPkYpbh%Qbr(NbNcExY?J_YDh8Z^w7#{*G;43yA?l`OC2K^dy!{v6SH zps#MZGnK;JmJV?nmvaY@bw2t4~(UEcWEs7l%MB_4?CHKX#Q!hWaXl6K6-TZ5uZ=I^?gOjBz|5}r&69B4}z_Tzld}wGy z?b%1(T+-Xdvq{HW@~S}iyt|ILpxD5FIltHoy%AD(m<*-9OrjE#=`&No~h z&a5kP)IFE8-cJKe!KPQ3n0;~FRcZ(VX?#sEJ~^5!nlJ&J+T5a$c0~u#e_WY-)+gKH z{YZ+(zrN=FbSpBh_D&uHpO)o?oyPXOrEcIeidyH1wR_`PpE?ueekOtlSq zD?*Um=U0_Q!WP0HmHr;0IDH($;EznIm)1y;*Uw?!R{{+X&=}A!?<5WRid`iRdZFm& zs+(j3E{U~_ZlOKp%p!IEGLh?yA26aZ^Jr+5KW(&=>FBfk$ha?qWhR^MGpKL!pMKr$ z7H_L>>D%j8#_jS#U#seThtMgb!S$HNFps~D>K6`Rg~$(tc6ZusJL_?`wc1}j9}iJbfaut)2sEk--+ z=k|Oi_(_-E>o54eBAm;1UF`l>4h<9QdG{*Q0M!SAvrL_28w3V~MH0;eJIaZKp0v29 z&8L=>zYp8bg30W+;VutS>>Hof2%kjM&0lXT{-?6DY-;oS+IEp5#jUsnEAH+Nh2lYq z6{p2DxVr_~QrtbbyA>}E#e;i+;BI$*|C#3lJTLPmbIoM!wXeO`ah$Pjy|bz4_Md;~ z1pC@Ph0t7v#4(dWv8uN^bN0A5tHRbFqD*Fu_d=(_{>f6i@-qE|xqU7m}YZ4OB1(@)vk7l|OabbzUsXrP!g8cW9xZUgV;}&7FRAIGETMG4Qrc z@Rcqr_DE;dhCVJEZw@J!9ZD-9!<^q9?2uEH5}_nyZc=qL*#3$=-TZa@m4Q!GdR6EU zY4dEtdO*_8qNu$V?|#l`$KVS~E7m$SZ(prQt+X8QNkep)GiGJi&2G26{L^daEZycT z0URxTOx4*qHtXF&(amaJTQ(5U6R32nkJcNKyGF-aYJi~Z%rhUGfjt8)G*Wh_n4ye| zZ5=c1H%zCvqvVahp$pm$Mk{QSf!7M&wdCw*V-Q<*DX?w&RYHw_PTRtKRD@@Cq(zii zt?u$wxSHhJF;K4=a%UZOB&L8KibDYJ`ufhIDyMtYH6BaRStSfkIWkwNG4l&t(SxYJ zQLQF`xN_vv{wyCaZ3~tpf^SRmTuiH&-0V_x6!%xw6rX|$3ILYPtUHWr;FSzYX&~@n6+zWdPH@-*=_8Us~~T#}hK z@N*_gzQ2b^ zx|WPHq$4pL3fGtHr0?`l)0fV5ivu;yzAajO`U^J&U;4RYQ!5P{U%oY7%`9|Wu8HJR z(-Gzn&XqZwD{+HyJ~^i-OJi@j#631hjddE()`yHmt57~FEo>vS_;W7Y*7Lb0Gi9*k zn6ht=38plebV1|SSqhBbd_+mA=y(3l&F!asQT$W6(CKy$(Z6pXFhQ1N`5w<2w)15j z9odK{jC;(c*JjE&C%`l3cN!Le0fv>-!1OqU$tH*A(E5nr5y;}!?O8NyN;W%q`bp56 zU+>u<+@g=_-?kcDzZ^l_9i(>K7B!WJfCp&eEp4?|?8Okn+GS zT>@*2p4(bjyL&ZTzw2()&PmDGE;^QvR2&Le553=b5P>jB87lho$Vtn?l+lg~B_KxR z%i|BP(Tzs6D+sv*OVrpQ)s;aY6pdCX&P7H}O~-L`b=Z(tWHAO; z0SjPieNi@v_gnd&TD;`@&{MWwPoa7#qokn6N#O3+-VD&~dsROYw)rA29lkd|5CR?R z(!8zwUPV>1syn!rUoQL%UwMJqj8@iI;X5^}PCrMF*a9p@%>$|C_N5b0vDn?GY5L>x`NI|S=4kN8UhC-(u8?-#qWPk>GvY!hTsQ3>$elCyZ1wiFoM=6^zt53mj_xLUqaDdBI*2;&ndxKcbao6CnTY5RGuXUdiv!~|-l`MzV}lS+#5;zJ-IVr*U%&TC*u|K0^ zf}G~(aVqCOX;9?~8S=Gm%SmpysOc;p#n2FFL+_-qi6T)j0v^vhY97Ye|3{p8cC3X>-o}cd&TX9WRcHtwxh2 ze~ZRP9M;=crn5a~#rp#|M{`;hwOYO^_%dziUVV)v8hoIco9tODo1YCDXl76J4dkHs zJH|uIgFy6&y2vc=8okc|G~xbf@a2A>Qf0xX(+!2Zk80(AxF-L5UPvX<_+d%5TpcgA zqP!(O(--EWZ=7YKE7&jl`82e-2$#Mk=OIx_xb!V}a)k5)RX6E|4p?EoHZNl`{meSf zElFi(<2+}sSK$)C+7E*{(ZZzzZS(WOzBm*IT!+MBAvn<76h#vfH7~=|ALqjKjTES@TEU0!YgbU~Rgb@O-&BENqDu=QN24{s@~={eReNU6E4=aQBoJ zzI1Eo9rrzlHmvAQ_pT8W)21an2KAa{)?y|A9Zk|x!u5+hSi>x1juF3}FZ0^t#hMmC zTk?1Rl`0FN@ZeBs%}*q{pq``Yw4`h#`F;oqqW#>abcnmkW{@bmgRNqTdHaKWnv?Tp zi<)ShZ8BM%r9)_kR^mH7aE)~=mmKIomZ5(N1(F&ceUC9UU;u$?qz?Ai4JImr~!Cdmrn(% zu2>#`Z36A`a6A`pZ(gfeo5zzY_B*8_>1=Lqr=v;|31BZsE;nqdPa^0o;xzjW-b08E z47SBlhPGEj5|*c#eE35?#=v_aeN<$j4l4(6aT zV&CYNpDouTGUyjCi1~8B)6iSIvL3A4GX(q={PrkFOw>mrUyHOm+EBK(61DJlxa}Y= zvpU3;AD4pDx3XhMir#gllFyHPVBc%u< zIh0{{Q4)H6;%620zIBt{BW8Mru zG0#P>RK_T=)G^~KMfa>G_|GE;A7Ur_*1uo1drGF38p_<^QW3E@9}!zoGx1zVLFb-K zHy13$*N*>Fr3O2qj}Vp>9|2}$wy{d$r|=3x*aCzS9Hh&~kS6BdE*riDGG1ni$r0zd ze<-U=Y|AifYm)6-Eo)bqks@)j6Y`?jS8~tbvd~zfuO)y#p99oYEMYrOrVxuTKXkq% zdQ(gA)Yawm!!m77y7@&vM%W&6u2*-YH5aIZ)MSx_SwZiSXxb0S_J@S_V~BL@<_2(% zkVMalhm5@ruCOY%ErBRG&ASCDhgov5@S83%31k&=CA#qzFb$Y8gbK&gCJP?ur+n6j zScb?LO@*dr>ImjWXMO>#pm{~#CnB(8MdF)1=^*bI3CS#>z+%>Byr~@v=zrpf04!XI zpK=_lh{0lECh@VP*o-AD@r)(1tWldjgD#k zK}+*BMMMV%5!m_}!(R44G`0JHo9DMNDUT~4yOG|iC?^WL^W<wRcpCpbZ&tw1EYJ#W=mWT|tW?TCwVq%jHJ zZ5KS~_XlicZ_m(Ff<)Eoc0DP3&ouUna3KCA;!{(JC%^V#j@Ed2+jk$ox}cAWLT!99 zV{S6`Sf5E!pbkhSHnkJeY92(9pB&KAeQ|F$V7FrHKUySb>|r7t@3da`;o2QjYJ2-E zl0eP3xq|9^GmNV)WVzfw{1vw%PvvYEccJ>Gs@sPY-40sv7BDqlLtb5#lERvjv2ldvj|V@y1W4oI5q?6foBgkH zDYz9~77WRA3H2zcWY}o{$s6S{99wPto*QrJpn(T6L<$P;6w-mrj^)^~3HEc<$@Ar# zu1BGaCecg=ii9d_l>Il^PFOmd`>1Qb~ z3pRU>UF}qc89%|)49*uDH*dNz!cq)(u{KS#VmiE! z#z$qe#Kt&K+kHG1Um>#0Piq}wKO=`$y!#L z5HC`{Dt5pvJN{Q(&OJYw_BM@;#1qD%k!V==&6QG%_nQBu7E_mv2C|uqtcIY9)u)?u z5sMY0Mfw`Z@GsxhOV(Q6|9WU)zdO;8+(8lH8ZB*5YiQ-W8*FdZMDP=0dd^~WiJm*= zt~Pt=;N_Vj$9W{M!4vHQ`Or;(Pp*_=$=ODDBkM23%0^kMV@2tJ)=W7|KG%?Vh5QM~ zBj=eVy1zJvbxCgG*l|2zr?j|uzMwrE+R?2008;tk7UB*)p8qn#u z=9A~A=hKKPEt8%IZ3FM0FhuPIo5z8pGu~ccK>W+co^5~{Mp86<2`P0( zkvNV<88MPGlZyLr@fBNZEM3R{2{76PN$I^S*wI1vKvJB^%mra3@5vK!Vs5)Qz zKxPI(x%&x-3=2WCh;6hn0N5$<-9|5mWU`)a+#XO$g{YP!XE_8PCpi{~AdTh5f5M_O z<*Dk791XVh6f68NJVB?GLtFtlxQp?<{#8(I>vpkiIj+uXRt@t;heRS#P;ovb%XpWa zdJ30Pu{=JH(2$iHQWjNI824mVzTe)$?T-J1fKX*KY)`TS+`t;+iqol z;mIsxo+5kX)byt1*akU;m|zZjd(20OcPS>rCAaW{cUnf6CRwPR2=+ewrIo2y7W-VF zFS7;ROoKQKJNuVuOz-*04Uk0oNVFL3Gs&2L8Gh=7oE5ir@4{k%K!#b1l0~(2u0FP# zRfF1t6rL-;Vpogd0$)qGX)Z3?W*d+_errbl^3cWsSyyFJ2TWMw9<}NIV~* ziQ$ZD{ys+0Bst9nhuif+JFy#Wo^0?g5Y3&j?IsouNsM;Z`kKNO1HtFxceUu1Kd~|^xTRhcc#Vb;5k~SsfFAZZxddhLo0G} z3XQ~ZBoc-(SrF)~O4ph4UCZZ*ti_wXrY-J2jmOlhiis*74U^XGDF#o|;_$zr)y@`6 zw8?qBzn>RP5rBm6&UEAVOJscQxOX$9`!*+qW8l9X zl5jG0%>Q@h5l6c*)llUKpMT03e3@C%=U3Jh3r`HSr_QCkk)ozTX`djmh^}FCtR8ZG zR+$m09T<-8{%9MBmt0E=zEEQ%PkR|m(%sfdJkBxhXuIFF-ULoL!_6H$65z-;kZK!_ z1a9-*N96h_x>)h%1)TQ=r(d)K=^~jl86(6e6q79lPbq~DD9ExIoj{%4*L6b|9@16X zgV|}qEQj`+s8Tw1M}1<0ERG>reUIxh_04_#vknTAnY0br@_k|l!!<*`sA<3Lsk}U! z@UkCvjE;sa=?oLjDp?*_9iegDH>;F$iE8D6Z{!H^dd(}?mLaV<>&K5c>C$U%^qoD!jRaumQ)@)(XA$Dkz zT(^4T&z~#)WDnRDwi%qp4jcf1$$LhSF%#wxlY8}q2Bw6@OQdw8t9G^Y4RJ+9nT-Y6 z%O1u2;bw>3cGDkqMzqvv&h@`@i&zkP*uSDfu@mtLPZfv!NzXCNJ2|zh{CQRV-Lr3M z;Wdz@&=+h5z76>Bb5e|ZO-N`7ZT+cNYk$-(xkE(rpgMI)(I^#%)(-tkDB*fTcPXhp zWvU_bZ*%x?;7BvzwN;Zn+nzX3P)`);;cv~*ed&MKE?5$&W8?2sBO-h#U z`lPO986`baZ#KaN@%x(30G!*1gOlu~1wG&M#uUBg056Bpl^@=B7zSsPAADrwOX|h? z%==Mq`?o=+f0Y>lA=aeN0Z!%Q&y!LnBGVHP0*irT+o@UU`+OPh<7o+Ou`%iv#h|kW z>rLj=$8^D8fw7SK=;7J-+X&I~J{6!$SfdwNi{UOzQ2vT2(#g=~yxi>SjQ?|_q{{q5 z`(FbAoJULXX$^7Sl_^O-myl|I3$tfb>NGU(2xK?Z=KSN7f(@{Lx>MY=MO;G7F>K!z z-Dcl#iD@YGCE^3Y!`V6p4%P_b5{`a{7g`dB{IWl$iz2o|JuJUm%mzTSZ+2dV>L zQEWu8EL9JzW@1$jlXLDdjKH{?Wk)I-;(;o8Q~5Pi!XcEv5Ii zLotVmMB92J%aVoMyogWLL7dnwcfP$mi?8L{hk_~P7fPHD^@5_@hHP|J4WZt}On=i9 zQ%CstnpM(7#dutJlU61^2=elpvTXWQxmp2+<3lTr_E$E4ROWJGg9^c%@OMmNBHc<{ zU%|z;JF8#oEw%{2c-Eoc7O|@&k5f)`zriTO=`BbF2evcx?_>0iIKAimc}4GL zC#NJXniI{}M4=wMb|iedDvu{l?4W@C2L*l}13n4ks6cvT6Er}i ztbj9P%Qkq}1%O40*57v0tMK*wI{r_$dkv)k{N=2RD>8_LDw5s?hpI>D|x;Z$>2{`J`HFhlE-+GF!3m0x2SB_RP1;euJ+`AzN7xNlP#=50gBf~U_7@K{1mI2 z%l-*CsG7lZe?0Vc%nxxy7ubgq_LWae`ohJ+c_|(vIpgQvEO`-UXFKqDq-cE2M|)>m zzL*-%68p9`aaS7UeRc*GKG#jPo!trmsL6+%57dvN1a;4YFHWv`UX~7PJMQ~!&v`9c zGe<(GfF#s6E!3dzZ|fvA`9y;)nQSTiLg$wuS5BF4t$Uh#)CWugKj!wJ13Z+|pW#7e z^IqR_I?SjjS)ji8b#ONdmf}3o-ZU98-y1XE%b4fkvuLpn{y-rujITk4K-N1!D&I^F z2X}VEH>}(xtMXATDh5|O(Fs|QK{`!DR^1N#C|^e$_Z0Zdfx`AJE6KSJY7yNoHOZ#p z+*+C0Y`^QJwY_hiW#DU7{6@s;)ZvIdz*H~r&gZx~PI>-;B+u*Jh?*!*bw2wDO8D&< zis|h{r$JM8TD5)7v*=M{=h%6>m8Skx;#@my@P{oN zedyIcBCmG;^A9>O{?BgRP`{gm>^`q)nAc!Rc55tlI8Sw_CQ9AfAR%hYen|OCA`+(a zgl=SidR`qiAW#GpGdr~@r(DyH_FVyW_>VS+bq!R?fz%R7suiHW@xz{ z8CIvW51-`xL?0(B#qw+|BTTtYii*ml&&Ra*-%ZV$i@9Kl?{#I1i$fnj1<|;AOjut6 zg}a$irou3fbij@gEUR6u`zQr;Mlk=ATDe`mlpPbqRV{iLog(BP2|V{2AFmSLeT!x% z;Z{TE*7G zKtgA&5zq=-5h;Mal$0cS zR=;iPz&p#?@PR*1kXBO`ZKU*9etEeQ~p5yf^s1rwyc+V@2!>n4x#3^on?vQ1R;{;!(WHM2(v8>E>0~5u=l*uyjU@q z0%VeEVryW-eGNHWQ07XoAFmroP^W$4C<;n=9t*ihNb9PdNY7cj^j(A(`FcgoePUj9> zdPOl%N)P2ev>9A5M;(aUkGf68?n!gQC(&)6yz+2s6er7&7QHkT=k@ zAv#aC?b@|MG z^jL1#?NMpCTavL<2H;9ASBLzgA1s8Zeo+NTZ~!b4z#CdL$as|leV<&2rR+s9rnq%f zpiRyOAiutW1s4O*G_ZdG@}d0w_6%fHH1??TDDZjgPM!K2(Q9K{mezFk-yxGhT1sp` z5+&=+J{6R)wQnqC2r(nus{Sv0+nn~AgGglM%nf9-B#N2}e><&ezl8`obK*`0VG>(y zY~Yt=VPZb%07yc|V|lcEi7*>)|V$PmwLm zbN=#(XnC+0j572(e}1m50%Ul<`{n9eN1T9T826Ck_}2{I*~|#(O)p0^#{8It5^pFF zb&?q-ai=CnK|F1gD3hD_k3kJHU7j{SUn^007#_v0QO<9CFrq5_?qTZMRT<7-YHpJ) zT8+_pVKgI(^@+=(wal$=_oA4hN}w+#fJt(Z3=3aOpeXlQEDx_v`w1?9h+wX`m+ivb z+)ZTY>t}c9n-fNE`Jm+73>4iCCV59~xd%SAO!1LyGDEF>;QIW|_rDAl-l`WA1Vg57 z^xneX7r@l_5@Qrcs`J8^7UDb$-WeK4m+XtJ3&KZ7S6+^$NZ>!&X`PU!KTz#P_GAY} zUxpQ}?a#t%EUg?CV!M87FVE7`zVtM28Ws)SA`7AtOq~2ECaAz`qv_?F~C3qSUCUM{7shv@F07{vSkdl;zdrs${-|{2xkbS>yl! literal 0 HcmV?d00001 diff --git a/docs/_images/sphx_glr_engine_caching_example_thumb.png b/docs/_images/sphx_glr_engine_caching_example_thumb.png new file mode 100644 index 0000000000000000000000000000000000000000..8a5fed589d17fe6ee62d6a9788519dba471fbd65 GIT binary patch literal 26794 zcmd3N^;cV6&~zU%f(#fBBGMqQ6{mHpumQ z^~%~w0VJvEnR)nlooKF`yb|H}xA|C<^cUE&K&!B{#ar*%RrUI{fOnuSCL$03!--#| zXbLuXBIIR=G00EgrNBb5!{Wg0L?-(}Cu?Haf+0F7%3dlH!I7q$mL$Yecj9<`jSzfT z^D5Zgt0)_Z#Ecth&8aB^q#^tt+jp*g;+l3Kuta%Af{32Sylxr^V1hg&jmnR}mkXfv zAbI2;{|#XPvE=`E0PeA)C<&;HFH;EcKm_ss4Tt~~BwnThPbCg5(+AS2SjXlQXBFG6FYU#Fiz_d=HGDi{5xv>!;RVHaotlES;Y{+4GrXjSnu(wiwX$)Rm{msq2XKT(PDE-xEgi z_(VivIE9l($F=w;5%X502cyd9S;o)?@*@IUt%-#RKi)is3GxWf#e9B1a0p}-K^GkH zE#RcLH-hEomo}wtS-`xST`geg^-vL1i@Xk;hL(*ihFVL8OfGro z7;=ZwLXX%~LDV&-I)%f^Zc@iBUPr)DOdL&SY8Fe4jRr0M^?OY#0}B$nzJ&L1In9Al zQOK9T4G@xxBY?1&=&}M<$uJ+xfTKBW>PRAN(VR;B>we1G_F=pv`)g>`Z^uYRx_jEF zLgunT}hWZL=J)o}-tSc-H43?kvu z2VGc#wT9ESOGB&rE@vy9Zz3|&#)3Ae*}ua?Sht~cbom|GLpPlTSm&X z8g90~9$!qLFeLh!xc`P0$(R$Ja&h&6EezoO__ACMkpyr*-p^zc6Sxuga|EOJ~>i$53|i zk3uPR?4WZH-H$~ya%J$=ipQkS-owRa4tw(P{%YPDc|taynH>AT=I|^o0}49wU~C{J zauK2=H85m@3Fzliq*hcd5LX*{uUE#$CpBIVU20AE#`cp-cs73fa2V=^W|F4HJO04v zh?+`+`1w}s-C)?|2Q&Bi)%tp}=My{T_L}AL?X!5%*(mp1+HX=ICa0VpGHL53#n?|w zrsVN!1h2aM0663#mGa+wr{K=E&1QL;2u+|ml4!Vcv|H~i8eectf>jNDUkq^angTdW z)#m!O%sOuC%7f^!Qht!qCw#tfe81zkyyL#c-mQq;u_=hU8Y&X0P3ZX%SuC=j%cBp_ zr)IOhDL^juo$ys}!om$%Q`tjo6J*I-hLpYq;KY(EwP$7&><+R6Cs-uf#l&-#?&}Uc z2}fsB6#K32O!O58@ksbMFteP}G2(2lnU(Xx0EQ{|6Fyl`z1ErflfT{pGkLUjTO->; z)k=c&e_nqG{-=f;xxac&YcIobydWM*>-WCPDRj}<`M?J$>U-1W=e@wPjOqn`WB6h3 zK~jXSxO+anw%MV<(bYq5#QL@?&$9}E9mOt!yYPhfLyOPk<($)-DALZ(Z<${Ys2e!I zEf@cqbht`&4b(^c=S@?tyU^-%7nsg>47yyUusdaMN|9?WoX_ry|GMNc@rE*&1ZWvhRi z_S3m-n`bk(@O+$m8Kp&CbTqx#!3Ro^Y+#WD0wD>iyApzndvKsR&ejCNHW0D$L?Y{& z-+!(ex(HR2t+z8ezJ13p%e#_pEzeK4BMasjYdnwJ?s{s>RYi;(GFI&ez1i>3C&Ca}f(#daZAeX;DyOc))i?10TL3cyU9coSu% z)362xbDc4*fSbB%;E}Qhj(t3Y;h0y44|>PJ{Wnh8tcJHxpgfb*cWZ`3JyR%I1eX@D zhJN6ES7k-g^nBRA6+Lz{hr;f=v9U%{mDsRr?D`_~Bl#>!q2hGpQrdVLN8GXyT@?Oz z+7-##Pu6db%j-<<)!r^-I*w+`UXXo=8fYr1;~kdxDdbuG7#=k3gnwVoy*Pse*8?CP{EM4h zidh=Lh5n)*Bd=X5LAxl_{94L7yp5$zMl^#3{EGwW`CdAsJZjClo-A{sUr5=r94uSOqPi$Bx2XqwJ0Xj>oqW`wvg3e07J z)B$YP_^e2>9eu-`H$Ilmv=_G3H|}OW)AT=Y3o&4{OAN>c@maT9IRH=16i5c13XIbq z?_;?AkC)0lxBghxk!|c14fTiYk`u>n>GBPA8g4GyLuofED1?CZ_ogTPNTn=*7Rz>` ztYLiEcuec^&_!PYFq(!2r=O|z#Ft@!biey30#@bucj&Ux#$iwD(4dukXbV1&1KhNt zqjE7f0?G2qOI$YC4rS^$^olhr+rM>-n6e_o=%}lSws$wIa?!a=?|&NpC0eU;zsWc{ zKWScWoG(_FP6kwvxFXl0Buq*6*?NA&t1N%CNhYQxAHE>%%=x@2fFKMkfr^ZzsGxgw zHOYm3bg4i(w(q;Sv}7U`6Pvh6Lm7Ptpx?d8)x>^1RQ1|vuG zD|M?Wx8nh#9<4K3^p^ z+tzr;24w)=wRU5Hn4~bhvixzHRwBbhqo*XVUm->z)SC1la7x*{$-z(_ak+M#+ZP-g zeJQP$k`X`41*d?oW^%4JZH z`2Ld2K6XjjT~MILbr{_xL>OfMd9;zSl*--#DwbWDM4Mp%J~N_%)nv|~hmH9@ z)4zT59opehPLJ^KJ6=VVO6Ho)jzT3gR98(NX*61GGVm56{>rj2M18)d^HhPG&iNrA zw3~jo&)!0v{GJE>iXI;ylLdla?z?dM+T(hzjWMg#Y7nngVIY8XR3%u9|L|1IzJP2% ztaEj&!l5_dWLvR#((N&`TQ%Z_aD4=0dIY1oPwr0rTMp@~krDj&qrOo>N7s>&+3e2# zj!pO%Q}P>}{K5VvO(@IX)|C%9o-~d!1M>GPa$u9}~ znp>y9i{nXjS1MjmdxDO!=)mKBA}-}mBp?edtY>D#G4~`p;ncc6ZCuu^9162{V`}LD z1V6Z{i6Rj#_U_ZbKJm)mqBY>vgDo5E5hb5`g#UiRW{W{;Yv3BoGE+jQunVS36Np2enN*G1C~E`p6Rz z{j*O=Hr16UEyfQtZjl0gr^mQuLw~P`!;^`G!)B$eL?HKwuJHg*Av%bEfp+&rI#S6h3Du-ZF@PZN*czynUs7CxrIzi<4CKW$DVUDZp4t@iHd(8;)$gN|7tcwXZ*}bVfvjelr52^Z_wbOI56$NP(NgU%iVHE zF4%~Q^N)t*gCUrPDQc6o=jm9zod2D$s=kAl9(4KKJVtQbESUEoxkjYn)gw9Hf$ zxkomp*C}$oszKqQn1bcxZ?n$jcvSPa-%+!QHfnAo#hP0@4Eu7#;_sn^YAFm}a5$kj zR>ve*17@Gze8vcL$w+E@`m5(=Qy*Q~>OI;1*l|C1v1PMWOw3eL9{)(8=9b%IM}6!U zMcHwGU&b-gd%E#WWQ1W&us{G=3v*Gczbd>nU~XA+RYmPvAK8xKIjItpke-|6DHLM? zV^m(Trt~H3n8Djvl-rxNw5%%KO97+CBR|c6-1xk!_MSCj{y-=Tl)@CO+~o2vmxWhz zY)tX$O0oEU#QKS|IP#9wssiTL8{ zJD#*5d^it313T$)w|h}CIq*lseFB}(#~)6PjENhpg?^8Xm3F>+oFgAQ0$j`746Z8! z4T=vRkY=AQ@V*?NvGxTdbw)Lbpeb#Hwn$jEkRl=kfxy_mFdnIZZ|v?Bk47e(MA6iu z=n4|Yw5OvgD)#Z`yIX8TL)kRf#j_X&X!Xre9^FbrGeT&ua~Ez z;L~rWd6P4b4BJHLCtjp5IQbK@Yn5nt(u(A%E*th>PWT9zmg#$i405 z1u)h5>|%j;2S6YL!d+3OTS>AcGo{_X_PRbtC^cZ~OZKy980j48DF6mKe0`ShOy)%)$Gpx&xYS;yMa=?0Luop+4>{HHg<+3oxDAZ>#%N{Z}PF z*y2V1MAwfF%&zQdI+R~N;zar?o(NK19$nQ&-1jXxMS(~iDLANz^Z7LflHO0Vw(S<{ z-&6fR1v zwYJ=7+&AQJkPZmGr9o>Y;3Dk51T1!Wny3eOhpSrVHP)r8j)n->0clR9O!IoX`MucX z^Zwiy9^o(E6WZ-Mwy{?YbRia8Hzm{^kdB^~ULyM{?&zV&+;rnvab{}4=iyZAU08O5 z@MV3_6vpe|{fU%r_^adQ-fn4Eqe6TqJ&VNr&47I)qv~~M_?mkg8<^CFz(S?YiNc&J zF8sSlt6VZ89ZVI=m8J2(?|e_Ih#V2_^euv=$5qy@pDZ&CU`?d!{!Xjfsm*EywtUo{ z7M`bz>Lm*;&#qzeUCS&pM&I|_VXXi>@ekIJ%6OP-srY6;NEm+rQ*Vm2SP=dpg`9v=8of0 zhOkW8VtL8ieYY31ECO7=oqO;@Z;1Kzm!dv@X`5}nJ?l8#I_`)#C(=D2%+*Mp=89va z5suRsE0+)2#jpPQZ^r2U4e8Tn0Artnx(_PB{~E_^f3atL)FN~J89#@h6>YEJ)SkOU zkVI-6t3OWeJfD)pik~F|nR+pV9ikPTcqm+4>0E|mxI>`U=jt6gulFW~I2NBxSomP5IND~h{YMxG>1blL@4eF^{6 zk#UK;tX9dPQL#jH$V)FC!y~DdwEKH`o^W&`SzLFoKY%L61$Od-y6mNtoA!97 zb$UD;2B^}_laB+luIKy^Uxf-;n)u=5i3a`beLsWhyI-)zg6gb&zCf*CjB)ffyJji% zQ#_YU?1e#Jdx!n7Hh(Ev_|i!PMx4H_`%&raZn?l8`iT!XKN5a0UHxpS!xGFhTNf+u zxpL3g?fDyTpaY9;=4WvgLgnhu2&s^lz|K2HyQC6^rpuT~EyYofcy)zpZBfmg;sx&O zRGn#3W%*3Vl+k#uA^Ks&Nc5APO_6`soi1jM35BHy;tl(LIhwN&a^a`% zJuR6vsJ;uQlY9qsuOj!Xa9BQ;(SH{l%NmW8PkWZ6_$5iF;!BonGJlTSxyBnademO^ z=BG6}B=bz`T7Vp#E}gK#fA=$pv=-a?p8toJvm3oLf8KYg@E~;oi^ZJM)2> z@1e%vtsSw=PX_(a0OjVZhHr*B{g*Z;W~=k%vO2VPhKypEAzx81pO<(tom_IjrLqcK z@zif`H$2E)oZV2FE-3#8Buc{2P?839y z#~{t0*VWy@2eH+)q@uti{*C6u-6YLDb<8RHk){z#wL@!tX{(Ia1#tU zO(0i6^t)Ueq*UJ0(u~F1x&}Xc`0a@7^&Z~y#d7ER`@Xo6>+bhoC<@-^DTsWJz$8O@ zE6q6A|0WUD9M91pzApN8$^s8ZUZ_$??^mFi= z)P@l=?Y7Lm!mt}6Km%zR(K8vzVwM*Cph$R+Xh-GcA2iO3omIER78s5{b=pAhf9(?R z=jTLCaD?TmqKr8OI!4VC|q!DCP=dfL9nazg#oyW^g{NeK5d*yo|ECkDoA zntQ_k_T^y$F%++7ZZFRcq6{D9EFXT)`{k$vq&&@tPha9Z|9HA1$islU8DzR5%*4N_ z8DoCrp?%1ld;gLWA)zb^F16gFV+?4T=W$cQrr1dXp6_NT3tcXsDig~m+Mf7k+r5{4 zaX`hIACv4>RGTPbo0)OUixoJ^mShV!I=(iQq{0#r8*6Kg8efTj75)X$kJM+F@I}UK zUT6J1kRe_tI*P!@&+7A20@t?a1+@I_mNWPu$Azzu(L~LJHNe5}vc>A0`ZhB2cvGWA zAP_fmED`+(u^vqY)Gl~k>@lkPE7s{N6Gk9+#3TqD@!RmIUBVFrL?mpY%{jeTz>Ul# z^|pRz2~Co5>*5Vwzu~?e8!^P_RU&lWU@N*MZCSemvk~1wkWh+dwBQK1rWMVqOV=%v zyWG(np@yt5?UDMjtFamAdo2)+Vqc~!LSR~pYD@3!kZumq_zz#BBQJsnet2!E?A{p`ONN@Od?Dfmf(G#Pe1^hkYDJ<6WdPOs0PO-=ay0Yv^)oC(`N1#}Xo#K?6jVUH7&Y zt>ZCkQuHIG`z{jqel&L+J&lkI(N??MwcMzEWwuRE z#tsWN%OvjLGsAyYLoMu0+RDt)d<{d{D_#C6oX)J(P(ujhKG$ds$ObrB+81&6HEUZK z6)Ny?G2BiOc@i>AX>0zY@kpJT=631_c&}0lHb_-=K+_#y@r3`Ov!QI~A~_f2I%EKS zJ7emxmhe7MD5eWcXU=kDb2?}pJ=Z2ixgDeUC)Cr0Dgk+Xip@@Z?fW>sIrSfsd3iKl zA^E2N1so#Rx-^RoLChMgcI*of<7Fc`lfSl9eUto$^I#>M(!@{8#<(NM6gWSTnNcQu z#F(n@U$X%u^XVuNW%;9#eG{+J3*S0eDbzOK*JU^!PFZiL;WG-ghu)JglFs-`?gx{0 zg^83973w0~@wfomO4tTCaGoAw5ySDvK+M&*B-%3F6upN6405V@Led1c)Ncz<5n^(< zPiF~QP0i%xt-8+Ans2bE6p^{j(LYwh4VMN%1BUMdM~=lxfRktQVj7Tbz4GRfJ5l**g7#AP{Ll zg26O4>0syPBWE01LOh{Vto8R|+hzwxjMgaa+4w4znI|@APC}4r- z_j-3kt^W!$v8L)O)(h**BQ`G7xTp5~*cIxL$Cbd!4BDqq1iG#K=~9AS9TqsXP_zV8f-mKeyL`)N8UjxE6ah@%CHBRjiYY zIAN$UE$g3x=1lf{!xt1ZRgVcz_({~_;ON4pN2KG(x>7`@%V6tAaX}iC*iEyNuh{Z< zP4>B7WT23ha(YzwOadZ+-><%ECjXix zzBw)Eg|m*luIPw7+s=KKGbMdlsp~TSU^^P?aC>~6uI9b%gV=r|dD=H2?R4lgxOouS z4y$=lE(0JSD=3f=cism*huq^ph>fl#eruAl&Qgel9&OAcjRk)uVWD_8t)s7{ATQEaE!nr^~>g=%(4pS8+Bu z5=G_(yS2XmS6Y<<>th8PL}|H=vh_+`rF++VgzfHwF?)POOi$iLaTu-X)KFZf9*u?T z?)}Bqv}%JriAp(#fdiklAyZx}iL7=i?Yhzm53*70oyBuUiOR%j>5$SfU;Qg4%7qnxL^X zxI~FR3wjB?`nEIEh!feQ?EB5fxl_Xbb9SubMyRuRRP*Q$EOTP&drs zY?gCRJmNK?Yq1V9fe#~p=K6_U2~;O55mu0P&seC38CLy+tv~Wp-x`q#x&>ZM7-Q3m zW4eD3)V})uA3+zGj}#kgL19Tb|B00v_HN*#OGe|6oD8pH=^tCuMly2~^7SO4!t)?g zy6@oxDL?L3C~3=371z7zURp$Z`m{P~37?4aXWtd&fX+CBi)^dH8Pk?Ah@fkZ()!`) z1X0mZu{ka}gkauYDfc*`#Qvobr##rtiB{U4PY809iCvoe)JrGm!V8g_?{9}sjEN0= zJ7$GABFPts7amvF?Wxg3FuLN{L~fQO2Ji_C{H?Zr zfx1Mn>%xiT^c%1j?7^kj&;Ai(5oLtxTKUAKq*cME;>S!-zWFU`$Q5Qp+Y6LFt-d-* zI|Yi&1D5sozyVt6Io*3Y(QbMUH6Zt&!WxBbM#)2%rL4sK&J$q7R-gKw`9Ij&^fvO1 zmLCl&t)>E;}-txS})rj3hexI>&3B9Lw&HJr;em;dp7)6hFSx)BpE zDm*>V;wiw)K?Qc**$GE$ELU6tzKjmkL)s zu`UM#;VDh!tL`#Wf4#+=lYGrdZfsUx$JbO|r4oTwXfXkrg4`=3V%Izv8Kj(cTDGhf z58eQ=^xq<-^r*F|u>!{;_Em2^J)jAy{&s94JOl#v*YHux@HFrJNuq_KMj{cd4B!GI zyQy%L+U#GslYhy*L&uRMJLVUA4D-9ylSREQjKgNu|8-2>jqtiW+McHya%Ket_zkuj zm3$1|$Ks(pmrP4EkMi}jCyvG+`NH5bpZ;|q0Y&){coHggK%;uf=ygA9Gjg@-z<06? z?Ix4=xt4ntSo@+9oy6cjU5elzD4ubWSy9ggSr6AII%4s~kygw*$EfUf zCY$#?+H2MPU7Dk3nKJ$X=vT#P&+45%*v>a)aJH8>pmXiNa08~ z#uSsyGl@a`R z?7#3vi8E>bKryMKl1od9W2}*q@jjedoHB;K~H? z;h{>mbHAOQ?yR})5R|r$vm|$Ig$ZnnL#t*8KK5}E+@@Kv8^(9yG?EfT5PU1;lin{4 zLSUjS%c8>N;~$k2tlzG=zm*BM6jW_WNrM*T)#Gz|3{UBi_BFO`gWIeuu7`n7qJ7!+ zoB49j8dB$kFYl%CkKkQAyfMq60G`QsWATY0(n^Qq0MqRiwat%l{@qJH_O!Wp|2ASV zp~0AbnKEINW-+Zod_NkA)8QUU3OqKkv6{WUv7umiWmN6w4y`-SB|;BV;-wz#^2nX) z8^1f-0CkHspK8Awr2)AoVa;dY`-(cwmj-#a^qWnY@txS1{hgb=G>2m7{0c%!)H zbs=y!>P=MRpDu#0WZhGe0hmuS$N;@)mS|*4E~72mu{YNq`RZI2l38VG; z&9cBZbIPwX$TSBPKN?k&W?cZIRHxaR)9Ch}tUXnw4yeb9w9Y0HH=?rzg6qU0_kXX& zikwQyBcGjclOqS5tHEx2(=0Eyz4ua<7wa614t|PXbnX@1^gG5j4}b>F?#wG?KKQUL z>D3%@JC#ak1x48sS1KpB=kC_yJ!7{-np@4coMInIt48ux_LJd7=zo%8N)@-)@Mqn? zCB{TDMLie03HEs~sahW1-369Q7k;$b8PLlKc|e60hl^hIV7!27_os@JOqQ{&^m@It zOU+BY&Kd1I<4B8XNEOGQ1fxpfLKZXxc`gJZVp(4YLUrlS#O3!F;mrX!*VfOSO6<2k839IE#ij3B6`Z33p3V1pDGW)z||*)hzf4| z#5U$te~Ik~F3J|qKe`0sb25<>Dt`sQv!b)&q;?!Q=CJ{$~>URp+ihe!Y>>%EQdotYO8c58TuoTLmkppGqebiz(tbA&68SL)Ym2PBh`wyUU1P7^o-V&xxQ%&I z1`AGU3bH%$YNuf^IXQ&0!Tdb$I7(3)vaau0-GLq!gUc*E{8a*|5H8aRMZy`Da}U;Q zk0R&2Y^ra(MQ4s*7$;THElm5vNno7%&0ev8lfW&rjt8_-k;O9IROA2H67|DhXL*}G z{w0f8xhpks8E$F1j1p#xIWiXZjBfJ5^IaMxafGqW`53v4sVk*MQwPMFnyS<{K}hw@ zd!wvxfN`sRTTVb-x;RGfyoA$6%2mq}{0gGHfQRm0M9P^IUVT{HFSh1`bO^0`(macn znI;iM@Bp<>s_2kCwF=JY z&PSrL)N>3YjsOL_gI3lipNJ2XrOul1Wkw+f?&ac$$3qPNi@V_?<6r^H*bU;p$q=>> z1HvZjJge#_N_vj=kpT^tx;Xz>wFlov3AYhs1rZwWA82QuRk~@988N#nN1&YOzQHa3 zdf&p06k#ycU~rClDyg|%s8er0F|}lCP@)Z{v-VzuK5T>5<=`+i$=M^zg~GidW{KUn8}&u^k7nksXyJs5{jRl`tHwI4aOs5*?ya1z}BK`rSbW4Osuwk zzP!7-44zPQ)BpsgLeCxhiNT!Za_>?f*CXJ;7qQ8#<^K=Qzj13VkbE1||LKI?N#Dgz zIi7fmv}!*%-MEm8-?F^f%-T%3{*>l+mq_<~HYo0Jb>M|QdT}F~I?YSYm@dv}bQ}pW zFz@r0I@tf#7ANC?Q{`j!@F@?F3h&0m;LY#5vfPc>xE`I4U`@X&vyP-}ERVw+47+d* z8XbB2sGh5Hd+{W*%UR)T5f#vG^F8`sWtR@~8)B#C?QQLfjP@T}EoZ+c4rjM1ZjGme zhKp}sp}+!o1JpBW<=)Fa2TtF_-3=QXVIC2VH8rTfxH&^ZfK)EC-4LZ7vc)gt$j;&B zE7W-upbGnrXAHR~IRB-K{$r25*hy70%I93yIb>rw9p~$V7vkmchY;zmv47Gi0=ZcH zG-@I9F(-ru^j;GAl76&@L6#uC9|fE$kv7K>zz5BI7}0z>OgtK##ulGp-w47^V_Q0# zHuCjpJ4->8o$~7oo7i0@Y29LCCw36}NAQbYir!Nzr3tMo>~3U^4m{^j8Rl_h?YAiE zyhGp#8%ummkQ_Q>=-qTpeH?NTNV6{LaTjc#b_*38J~BuhbgN0&^W0t-1FuH~#hhC1 zMn{OGvG&h*`xt>KB~YjZM>i)`Nsmwxb)F$$rXEa__@Kz+APMRe*11R#?(&0!_hWIB<((OfR`TY$wQ>BK1O&x$6u zr8b-zG`lss-&MNJ83H{XmrzrxLCXP3zx616s_fK!t~=CSlew@PbyMR61D!a| zsLmv7%{S${Gp_IQYMXWX0v;SV?H^Pci3At^G0=Y~46@dK00jSvyO&zMzs`y%Yr2nYu6#9Y z*YTghKFZCC2=D}K1OtgE7My;a;3 zsO4r0%Bicd7q4#W5 zm+Xvgm|hj6LwVVX7t5zeMTOVcpsaPSw)$>PxMkvdb`kyB?G;aGW z#kZjiTUtq)U|p}b)@9@V#PZn`{gD1Td0g|}}0wF!xU=e0hy zHyG}uiLnRu=sn7Bh;pjoG`}Fdv!-#0vnYegTi{ni^fepVyOqj1R+r>Tla9W>4-K)UAhR`x5(nWp0izSdW$kHF1DHoWrOr2;ljn~@&VXv z!;to}?n^_JyK~TjZwrA8#KSX8dc!BRrSBlPOK7p;OZB1NX&jQUu>3+zhyr%DF>>dQ*Vj_VXb;zX( zIo_JSRUrlqS!=a;rg_c(E}I>bRiUcdPKf);b;qbq#k^^jYshOCbOAWZUqIUnQ9fUKBEK$&)hXv1&qdgv{1eugEQ%@G{a2OG^*wXuG3b~$Zv(2H3pToh28)cr zYev{el;D$^5k5A9V}#~b=H}vTS^YGz0)O@*21Z`O0#~2 zm-{BE`O_B3P}xZhi&+@oFrY>(h+qjmQno+gN&jLJZ+iW2Lp5>J!M6k+iEQybApKr0+#up4UogAeoM;JMHEB#{VVG0Br(~t z;o<{@RWdi)*lYX)(Ellg;`^QWQ~uKqQPcgOc8HPgMHiIdN%#N6MK-MwOPxW=U$AsA z6hD+|C?>Sb{SdlDXTqSEA{YPm#r(u&eVOc1!~;E`A@Ii?TO5jMV6eX7U#3^=(KbJw z)X$50!WpsTP5Ivi;dG`1(Q1Mn-iO$?#VW{dUm|NOB$ja@QE>~Ph#-mNJ6c1`0#bL8 zb^;qmJmr$E@2Yme^?T-@OSlKhpe$oGY@`wSL5|#*;yAw8FZzoULOkIY8Pjn%67i)} z{l}6W>Jt9?%jS@lTQ|d>jTNm#PJF*w6sHB{j<*k-*5Fdj_2a$#24~eH=jC%yHg2!H z2%3x{t~SIdKK>vjhfqvt_1%^QjO=(dj%>Ar?0bGs(A!D$zPa)z8Xsv49exN|7Z+O6 zQRUPUKl+X}fXYb+dctaPfkRo?R%^7-e7>~5?UUu~$KcRYt<>l?;Xf7hVW&f)4q zKa7KyF0KN+?^GbTpU)om1euxNw5TZ^rg5oVeTlQIZ`^tJb~Uv$0s|b;q%9c&;>51e z`c$)xs>`*eJ`x%qvrd4^9+drc4!I@NZrA_@HNnBsw{F1nVkF8z@ec`yiSEx14LDqy z(CcM#uu}3QSi0JRtqHAV&yzQlJ+KM6c}f*i3M;DrmHQUjI^~nLX$wx^x;4UaN$FVt z&Pmm5(XnW;jlm;3k}hAJAQFW@KVom}CA#H78GhaYQ}k-~GuPz3;-Eyv;kR5brwy=B zfR_?_yPB`{N!wjgu_j5w7@JOz2OZGu;OL@xz{Q;*pimlR!y=pCB{#+t5AL z32+inpvA+yi-KRXMGWNP^XPZAp}vroYP!sQG4E;+EVla%vHfn| z#|EaX2j^1(GZJzEND26^KBVP~l=+=c^IXE0_g9ivk#n7S?_xastD0^LpRc#-hdV_0 z{hMS2L%wspFLfHM`AppRfmS04pWmTbkzSqy*eo0l65!CmA7^`)_Abq90Nw2}UpR@b zjs1LC*2a%m*M;*QVbo&1a^QSx7#|CBpliq@NG4%{O(l;%KbSXgE&km>;pUK}O#LY7 z=hKdHY$ZPGW{US;_!1p?0+d#mw!Ifo*>HlFQT4NYKK3Xx=%8w}&z^)%lHyL5sH{iV zQlF5F9w7Upp0rr!ed-P?%I&>cfx#jb4Y6>zdKeJ+1*oh}oS%=)jNbSgBl(?h{7G(h z3mkpT=d=<%WB8h)flTzOV9O;(hB8mL9kMwruOktKJSwJobi9HGFIkOasl_Se(vrzI z%21W_{aWwfkgZ?Y-lndle*~}utqX+WJJEhlYJk+i+X|(+ms~(05=ByP)FJUvJ{*@t zA0UuB_vEcydLk(s6&v~4(`h))Cc`&Un_zxPiYeT<0))7e6|JO%vqVKb!nHdzRPQar^O`$jL}9z zcyBLNT;7G_UWPng87D7{$!@~X$jtVVDcw-^!O--QX8KHDsde0RJpzB=Dl8CGTWVX; z@z}4;-=5e^MKO4ZMW5ZX8QNt)v*|Dr=8Jk=B(Ie^;aW$=do+aqxqxHb|3JavGWra5 zNAiX(4t5fh7us?>RPI#)^Zbo>R|As>Q{=BSGd7fKNt-!1Ld6HSL-vHePI$O@TM*k}5G zq^A`yH&o-#vlckF*uR?APTJx>E?Zt)SPk|+!z72N4GVeicTyS;X^jnzEuc?(yP3P> zlN08Op$ef~naLYyYId-keedgm&K9%#VT>(izue;kL43eqt%)>!v%o88*#7LwDY|8_^$tgu9%gq<`i_mHPkDUVII;Qtw_RB+^7)5mC zCO)~}!GYsE*TkmRB3s_G>oXw8BYs`AY5sg;!geCmo1pHIex)aefr2!DvA#Du%ja7g zDT(m(jrA81Ja0NuFR4?npk#SW=Tgxj>f>|w4$m(>3*rxA$Nt%CR|a@Y`UTlmzk=`{ zlJa_LK&$aMpOzv$X4@i{)S2$*P^*VjVyi`d>`~!zR2>ac!W2?GYJ3BA%brp-`1T-|#o!YMKf7{>+eEb0 z!+5L{^jYlE?$bB4;fJkXuVVT8R&evJ-b|1^3j?y}_DNB@Qmhb~hIpqim9aAH5+>c)Y#7rpKts zZ(aG`MU5}{u>`^kuxaA4s219GVYtPiXZkO2mN)&g2(3*85 zekL#c#|vwqsPkYpbh%Qbr(NbNcExY?J_YDh8Z^w7#{*G;43yA?l`OC2K^dy!{v6SH zps#MZGnK;JmJV?nmvaY@bw2t4~(UEcWEs7l%MB_4?CHKX#Q!hWaXl6K6-TZ5uZ=I^?gOjBz|5}r&69B4}z_Tzld}wGy z?b%1(T+-Xdvq{HW@~S}iyt|ILpxD5FIltHoy%AD(m<*-9OrjE#=`&No~h z&a5kP)IFE8-cJKe!KPQ3n0;~FRcZ(VX?#sEJ~^5!nlJ&J+T5a$c0~u#e_WY-)+gKH z{YZ+(zrN=FbSpBh_D&uHpO)o?oyPXOrEcIeidyH1wR_`PpE?ueekOtlSq zD?*Um=U0_Q!WP0HmHr;0IDH($;EznIm)1y;*Uw?!R{{+X&=}A!?<5WRid`iRdZFm& zs+(j3E{U~_ZlOKp%p!IEGLh?yA26aZ^Jr+5KW(&=>FBfk$ha?qWhR^MGpKL!pMKr$ z7H_L>>D%j8#_jS#U#seThtMgb!S$HNFps~D>K6`Rg~$(tc6ZusJL_?`wc1}j9}iJbfaut)2sEk--+ z=k|Oi_(_-E>o54eBAm;1UF`l>4h<9QdG{*Q0M!SAvrL_28w3V~MH0;eJIaZKp0v29 z&8L=>zYp8bg30W+;VutS>>Hof2%kjM&0lXT{-?6DY-;oS+IEp5#jUsnEAH+Nh2lYq z6{p2DxVr_~QrtbbyA>}E#e;i+;BI$*|C#3lJTLPmbIoM!wXeO`ah$Pjy|bz4_Md;~ z1pC@Ph0t7v#4(dWv8uN^bN0A5tHRbFqD*Fu_d=(_{>f6i@-qE|xqU7m}YZ4OB1(@)vk7l|OabbzUsXrP!g8cW9xZUgV;}&7FRAIGETMG4Qrc z@Rcqr_DE;dhCVJEZw@J!9ZD-9!<^q9?2uEH5}_nyZc=qL*#3$=-TZa@m4Q!GdR6EU zY4dEtdO*_8qNu$V?|#l`$KVS~E7m$SZ(prQt+X8QNkep)GiGJi&2G26{L^daEZycT z0URxTOx4*qHtXF&(amaJTQ(5U6R32nkJcNKyGF-aYJi~Z%rhUGfjt8)G*Wh_n4ye| zZ5=c1H%zCvqvVahp$pm$Mk{QSf!7M&wdCw*V-Q<*DX?w&RYHw_PTRtKRD@@Cq(zii zt?u$wxSHhJF;K4=a%UZOB&L8KibDYJ`ufhIDyMtYH6BaRStSfkIWkwNG4l&t(SxYJ zQLQF`xN_vv{wyCaZ3~tpf^SRmTuiH&-0V_x6!%xw6rX|$3ILYPtUHWr;FSzYX&~@n6+zWdPH@-*=_8Us~~T#}hK z@N*_gzQ2b^ zx|WPHq$4pL3fGtHr0?`l)0fV5ivu;yzAajO`U^J&U;4RYQ!5P{U%oY7%`9|Wu8HJR z(-Gzn&XqZwD{+HyJ~^i-OJi@j#631hjddE()`yHmt57~FEo>vS_;W7Y*7Lb0Gi9*k zn6ht=38plebV1|SSqhBbd_+mA=y(3l&F!asQT$W6(CKy$(Z6pXFhQ1N`5w<2w)15j z9odK{jC;(c*JjE&C%`l3cN!Le0fv>-!1OqU$tH*A(E5nr5y;}!?O8NyN;W%q`bp56 zU+>u<+@g=_-?kcDzZ^l_9i(>K7B!WJfCp&eEp4?|?8Okn+GS zT>@*2p4(bjyL&ZTzw2()&PmDGE;^QvR2&Le553=b5P>jB87lho$Vtn?l+lg~B_KxR z%i|BP(Tzs6D+sv*OVrpQ)s;aY6pdCX&P7H}O~-L`b=Z(tWHAO; z0SjPieNi@v_gnd&TD;`@&{MWwPoa7#qokn6N#O3+-VD&~dsROYw)rA29lkd|5CR?R z(!8zwUPV>1syn!rUoQL%UwMJqj8@iI;X5^}PCrMF*a9p@%>$|C_N5b0vDn?GY5L>x`NI|S=4kN8UhC-(u8?-#qWPk>GvY!hTsQ3>$elCyZ1wiFoM=6^zt53mj_xLUqaDdBI*2;&ndxKcbao6CnTY5RGuXUdiv!~|-l`MzV}lS+#5;zJ-IVr*U%&TC*u|K0^ zf}G~(aVqCOX;9?~8S=Gm%SmpysOc;p#n2FFL+_-qi6T)j0v^vhY97Ye|3{p8cC3X>-o}cd&TX9WRcHtwxh2 ze~ZRP9M;=crn5a~#rp#|M{`;hwOYO^_%dziUVV)v8hoIco9tODo1YCDXl76J4dkHs zJH|uIgFy6&y2vc=8okc|G~xbf@a2A>Qf0xX(+!2Zk80(AxF-L5UPvX<_+d%5TpcgA zqP!(O(--EWZ=7YKE7&jl`82e-2$#Mk=OIx_xb!V}a)k5)RX6E|4p?EoHZNl`{meSf zElFi(<2+}sSK$)C+7E*{(ZZzzZS(WOzBm*IT!+MBAvn<76h#vfH7~=|ALqjKjTES@TEU0!YgbU~Rgb@O-&BENqDu=QN24{s@~={eReNU6E4=aQBoJ zzI1Eo9rrzlHmvAQ_pT8W)21an2KAa{)?y|A9Zk|x!u5+hSi>x1juF3}FZ0^t#hMmC zTk?1Rl`0FN@ZeBs%}*q{pq``Yw4`h#`F;oqqW#>abcnmkW{@bmgRNqTdHaKWnv?Tp zi<)ShZ8BM%r9)_kR^mH7aE)~=mmKIomZ5(N1(F&ceUC9UU;u$?qz?Ai4JImr~!Cdmrn(% zu2>#`Z36A`a6A`pZ(gfeo5zzY_B*8_>1=Lqr=v;|31BZsE;nqdPa^0o;xzjW-b08E z47SBlhPGEj5|*c#eE35?#=v_aeN<$j4l4(6aT zV&CYNpDouTGUyjCi1~8B)6iSIvL3A4GX(q={PrkFOw>mrUyHOm+EBK(61DJlxa}Y= zvpU3;AD4pDx3XhMir#gllFyHPVBc%u< zIh0{{Q4)H6;%620zIBt{BW8Mru zG0#P>RK_T=)G^~KMfa>G_|GE;A7Ur_*1uo1drGF38p_<^QW3E@9}!zoGx1zVLFb-K zHy13$*N*>Fr3O2qj}Vp>9|2}$wy{d$r|=3x*aCzS9Hh&~kS6BdE*riDGG1ni$r0zd ze<-U=Y|AifYm)6-Eo)bqks@)j6Y`?jS8~tbvd~zfuO)y#p99oYEMYrOrVxuTKXkq% zdQ(gA)Yawm!!m77y7@&vM%W&6u2*-YH5aIZ)MSx_SwZiSXxb0S_J@S_V~BL@<_2(% zkVMalhm5@ruCOY%ErBRG&ASCDhgov5@S83%31k&=CA#qzFb$Y8gbK&gCJP?ur+n6j zScb?LO@*dr>ImjWXMO>#pm{~#CnB(8MdF)1=^*bI3CS#>z+%>Byr~@v=zrpf04!XI zpK=_lh{0lECh@VP*o-AD@r)(1tWldjgD#k zK}+*BMMMV%5!m_}!(R44G`0JHo9DMNDUT~4yOG|iC?^WL^W<wRcpCpbZ&tw1EYJ#W=mWT|tW?TCwVq%jHJ zZ5KS~_XlicZ_m(Ff<)Eoc0DP3&ouUna3KCA;!{(JC%^V#j@Ed2+jk$ox}cAWLT!99 zV{S6`Sf5E!pbkhSHnkJeY92(9pB&KAeQ|F$V7FrHKUySb>|r7t@3da`;o2QjYJ2-E zl0eP3xq|9^GmNV)WVzfw{1vw%PvvYEccJ>Gs@sPY-40sv7BDqlLtb5#lERvjv2ldvj|V@y1W4oI5q?6foBgkH zDYz9~77WRA3H2zcWY}o{$s6S{99wPto*QrJpn(T6L<$P;6w-mrj^)^~3HEc<$@Ar# zu1BGaCecg=ii9d_l>Il^PFOmd`>1Qb~ z3pRU>UF}qc89%|)49*uDH*dNz!cq)(u{KS#VmiE! z#z$qe#Kt&K+kHG1Um>#0Piq}wKO=`$y!#L z5HC`{Dt5pvJN{Q(&OJYw_BM@;#1qD%k!V==&6QG%_nQBu7E_mv2C|uqtcIY9)u)?u z5sMY0Mfw`Z@GsxhOV(Q6|9WU)zdO;8+(8lH8ZB*5YiQ-W8*FdZMDP=0dd^~WiJm*= zt~Pt=;N_Vj$9W{M!4vHQ`Or;(Pp*_=$=ODDBkM23%0^kMV@2tJ)=W7|KG%?Vh5QM~ zBj=eVy1zJvbxCgG*l|2zr?j|uzMwrE+R?2008;tk7UB*)p8qn#u z=9A~A=hKKPEt8%IZ3FM0FhuPIo5z8pGu~ccK>W+co^5~{Mp86<2`P0( zkvNV<88MPGlZyLr@fBNZEM3R{2{76PN$I^S*wI1vKvJB^%mra3@5vK!Vs5)Qz zKxPI(x%&x-3=2WCh;6hn0N5$<-9|5mWU`)a+#XO$g{YP!XE_8PCpi{~AdTh5f5M_O z<*Dk791XVh6f68NJVB?GLtFtlxQp?<{#8(I>vpkiIj+uXRt@t;heRS#P;ovb%XpWa zdJ30Pu{=JH(2$iHQWjNI824mVzTe)$?T-J1fKX*KY)`TS+`t;+iqol z;mIsxo+5kX)byt1*akU;m|zZjd(20OcPS>rCAaW{cUnf6CRwPR2=+ewrIo2y7W-VF zFS7;ROoKQKJNuVuOz-*04Uk0oNVFL3Gs&2L8Gh=7oE5ir@4{k%K!#b1l0~(2u0FP# zRfF1t6rL-;Vpogd0$)qGX)Z3?W*d+_errbl^3cWsSyyFJ2TWMw9<}NIV~* ziQ$ZD{ys+0Bst9nhuif+JFy#Wo^0?g5Y3&j?IsouNsM;Z`kKNO1HtFxceUu1Kd~|^xTRhcc#Vb;5k~SsfFAZZxddhLo0G} z3XQ~ZBoc-(SrF)~O4ph4UCZZ*ti_wXrY-J2jmOlhiis*74U^XGDF#o|;_$zr)y@`6 zw8?qBzn>RP5rBm6&UEAVOJscQxOX$9`!*+qW8l9X zl5jG0%>Q@h5l6c*)llUKpMT03e3@C%=U3Jh3r`HSr_QCkk)ozTX`djmh^}FCtR8ZG zR+$m09T<-8{%9MBmt0E=zEEQ%PkR|m(%sfdJkBxhXuIFF-ULoL!_6H$65z-;kZK!_ z1a9-*N96h_x>)h%1)TQ=r(d)K=^~jl86(6e6q79lPbq~DD9ExIoj{%4*L6b|9@16X zgV|}qEQj`+s8Tw1M}1<0ERG>reUIxh_04_#vknTAnY0br@_k|l!!<*`sA<3Lsk}U! z@UkCvjE;sa=?oLjDp?*_9iegDH>;F$iE8D6Z{!H^dd(}?mLaV<>&K5c>C$U%^qoD!jRaumQ)@)(XA$Dkz zT(^4T&z~#)WDnRDwi%qp4jcf1$$LhSF%#wxlY8}q2Bw6@OQdw8t9G^Y4RJ+9nT-Y6 z%O1u2;bw>3cGDkqMzqvv&h@`@i&zkP*uSDfu@mtLPZfv!NzXCNJ2|zh{CQRV-Lr3M z;Wdz@&=+h5z76>Bb5e|ZO-N`7ZT+cNYk$-(xkE(rpgMI)(I^#%)(-tkDB*fTcPXhp zWvU_bZ*%x?;7BvzwN;Zn+nzX3P)`);;cv~*ed&MKE?5$&W8?2sBO-h#U z`lPO986`baZ#KaN@%x(30G!*1gOlu~1wG&M#uUBg056Bpl^@=B7zSsPAADrwOX|h? z%==Mq`?o=+f0Y>lA=aeN0Z!%Q&y!LnBGVHP0*irT+o@UU`+OPh<7o+Ou`%iv#h|kW z>rLj=$8^D8fw7SK=;7J-+X&I~J{6!$SfdwNi{UOzQ2vT2(#g=~yxi>SjQ?|_q{{q5 z`(FbAoJULXX$^7Sl_^O-myl|I3$tfb>NGU(2xK?Z=KSN7f(@{Lx>MY=MO;G7F>K!z z-Dcl#iD@YGCE^3Y!`V6p4%P_b5{`a{7g`dB{IWl$iz2o|JuJUm%mzTSZ+2dV>L zQEWu8EL9JzW@1$jlXLDdjKH{?Wk)I-;(;o8Q~5Pi!XcEv5Ii zLotVmMB92J%aVoMyogWLL7dnwcfP$mi?8L{hk_~P7fPHD^@5_@hHP|J4WZt}On=i9 zQ%CstnpM(7#dutJlU61^2=elpvTXWQxmp2+<3lTr_E$E4ROWJGg9^c%@OMmNBHc<{ zU%|z;JF8#oEw%{2c-Eoc7O|@&k5f)`zriTO=`BbF2evcx?_>0iIKAimc}4GL zC#NJXniI{}M4=wMb|iedDvu{l?4W@C2L*l}13n4ks6cvT6Er}i ztbj9P%Qkq}1%O40*57v0tMK*wI{r_$dkv)k{N=2RD>8_LDw5s?hpI>D|x;Z$>2{`J`HFhlE-+GF!3m0x2SB_RP1;euJ+`AzN7xNlP#=50gBf~U_7@K{1mI2 z%l-*CsG7lZe?0Vc%nxxy7ubgq_LWae`ohJ+c_|(vIpgQvEO`-UXFKqDq-cE2M|)>m zzL*-%68p9`aaS7UeRc*GKG#jPo!trmsL6+%57dvN1a;4YFHWv`UX~7PJMQ~!&v`9c zGe<(GfF#s6E!3dzZ|fvA`9y;)nQSTiLg$wuS5BF4t$Uh#)CWugKj!wJ13Z+|pW#7e z^IqR_I?SjjS)ji8b#ONdmf}3o-ZU98-y1XE%b4fkvuLpn{y-rujITk4K-N1!D&I^F z2X}VEH>}(xtMXATDh5|O(Fs|QK{`!DR^1N#C|^e$_Z0Zdfx`AJE6KSJY7yNoHOZ#p z+*+C0Y`^QJwY_hiW#DU7{6@s;)ZvIdz*H~r&gZx~PI>-;B+u*Jh?*!*bw2wDO8D&< zis|h{r$JM8TD5)7v*=M{=h%6>m8Skx;#@my@P{oN zedyIcBCmG;^A9>O{?BgRP`{gm>^`q)nAc!Rc55tlI8Sw_CQ9AfAR%hYen|OCA`+(a zgl=SidR`qiAW#GpGdr~@r(DyH_FVyW_>VS+bq!R?fz%R7suiHW@xz{ z8CIvW51-`xL?0(B#qw+|BTTtYii*ml&&Ra*-%ZV$i@9Kl?{#I1i$fnj1<|;AOjut6 zg}a$irou3fbij@gEUR6u`zQr;Mlk=ATDe`mlpPbqRV{iLog(BP2|V{2AFmSLeT!x% z;Z{TE*7G zKtgA&5zq=-5h;Mal$0cS zR=;iPz&p#?@PR*1kXBO`ZKU*9etEeQ~p5yf^s1rwyc+V@2!>n4x#3^on?vQ1R;{;!(WHM2(v8>E>0~5u=l*uyjU@q z0%VeEVryW-eGNHWQ07XoAFmroP^W$4C<;n=9t*ihNb9PdNY7cj^j(A(`FcgoePUj9> zdPOl%N)P2ev>9A5M;(aUkGf68?n!gQC(&)6yz+2s6er7&7QHkT=k@ zAv#aC?b@|MG z^jL1#?NMpCTavL<2H;9ASBLzgA1s8Zeo+NTZ~!b4z#CdL$as|leV<&2rR+s9rnq%f zpiRyOAiutW1s4O*G_ZdG@}d0w_6%fHH1??TDDZjgPM!K2(Q9K{mezFk-yxGhT1sp` z5+&=+J{6R)wQnqC2r(nus{Sv0+nn~AgGglM%nf9-B#N2}e><&ezl8`obK*`0VG>(y zY~Yt=VPZb%07yc|V|lcEi7*>)|V$PmwLm zbN=#(XnC+0j572(e}1m50%Ul<`{n9eN1T9T826Ck_}2{I*~|#(O)p0^#{8It5^pFF zb&?q-ai=CnK|F1gD3hD_k3kJHU7j{SUn^007#_v0QO<9CFrq5_?qTZMRT<7-YHpJ) zT8+_pVKgI(^@+=(wal$=_oA4hN}w+#fJt(Z3=3aOpeXlQEDx_v`w1?9h+wX`m+ivb z+)ZTY>t}c9n-fNE`Jm+73>4iCCV59~xd%SAO!1LyGDEF>;QIW|_rDAl-l`WA1Vg57 z^xneX7r@l_5@Qrcs`J8^7UDb$-WeK4m+XtJ3&KZ7S6+^$NZ>!&X`PU!KTz#P_GAY} zUxpQ}?a#t%EUg?CV!M87FVE7`zVtM28Ws)SA`7AtOq~2ECaAz`qv_?F~C3qSUCUM{7shv@F07{vSkdl;zdrs${-|{2xkbS>yl! literal 0 HcmV?d00001 diff --git a/docs/_images/sphx_glr_torch_export_gpt2_thumb.png b/docs/_images/sphx_glr_torch_export_gpt2_thumb.png new file mode 100644 index 0000000000000000000000000000000000000000..8a5fed589d17fe6ee62d6a9788519dba471fbd65 GIT binary patch literal 26794 zcmd3N^;cV6&~zU%f(#fBBGMqQ6{mHpumQ z^~%~w0VJvEnR)nlooKF`yb|H}xA|C<^cUE&K&!B{#ar*%RrUI{fOnuSCL$03!--#| zXbLuXBIIR=G00EgrNBb5!{Wg0L?-(}Cu?Haf+0F7%3dlH!I7q$mL$Yecj9<`jSzfT z^D5Zgt0)_Z#Ecth&8aB^q#^tt+jp*g;+l3Kuta%Af{32Sylxr^V1hg&jmnR}mkXfv zAbI2;{|#XPvE=`E0PeA)C<&;HFH;EcKm_ss4Tt~~BwnThPbCg5(+AS2SjXlQXBFG6FYU#Fiz_d=HGDi{5xv>!;RVHaotlES;Y{+4GrXjSnu(wiwX$)Rm{msq2XKT(PDE-xEgi z_(VivIE9l($F=w;5%X502cyd9S;o)?@*@IUt%-#RKi)is3GxWf#e9B1a0p}-K^GkH zE#RcLH-hEomo}wtS-`xST`geg^-vL1i@Xk;hL(*ihFVL8OfGro z7;=ZwLXX%~LDV&-I)%f^Zc@iBUPr)DOdL&SY8Fe4jRr0M^?OY#0}B$nzJ&L1In9Al zQOK9T4G@xxBY?1&=&}M<$uJ+xfTKBW>PRAN(VR;B>we1G_F=pv`)g>`Z^uYRx_jEF zLgunT}hWZL=J)o}-tSc-H43?kvu z2VGc#wT9ESOGB&rE@vy9Zz3|&#)3Ae*}ua?Sht~cbom|GLpPlTSm&X z8g90~9$!qLFeLh!xc`P0$(R$Ja&h&6EezoO__ACMkpyr*-p^zc6Sxuga|EOJ~>i$53|i zk3uPR?4WZH-H$~ya%J$=ipQkS-owRa4tw(P{%YPDc|taynH>AT=I|^o0}49wU~C{J zauK2=H85m@3Fzliq*hcd5LX*{uUE#$CpBIVU20AE#`cp-cs73fa2V=^W|F4HJO04v zh?+`+`1w}s-C)?|2Q&Bi)%tp}=My{T_L}AL?X!5%*(mp1+HX=ICa0VpGHL53#n?|w zrsVN!1h2aM0663#mGa+wr{K=E&1QL;2u+|ml4!Vcv|H~i8eectf>jNDUkq^angTdW z)#m!O%sOuC%7f^!Qht!qCw#tfe81zkyyL#c-mQq;u_=hU8Y&X0P3ZX%SuC=j%cBp_ zr)IOhDL^juo$ys}!om$%Q`tjo6J*I-hLpYq;KY(EwP$7&><+R6Cs-uf#l&-#?&}Uc z2}fsB6#K32O!O58@ksbMFteP}G2(2lnU(Xx0EQ{|6Fyl`z1ErflfT{pGkLUjTO->; z)k=c&e_nqG{-=f;xxac&YcIobydWM*>-WCPDRj}<`M?J$>U-1W=e@wPjOqn`WB6h3 zK~jXSxO+anw%MV<(bYq5#QL@?&$9}E9mOt!yYPhfLyOPk<($)-DALZ(Z<${Ys2e!I zEf@cqbht`&4b(^c=S@?tyU^-%7nsg>47yyUusdaMN|9?WoX_ry|GMNc@rE*&1ZWvhRi z_S3m-n`bk(@O+$m8Kp&CbTqx#!3Ro^Y+#WD0wD>iyApzndvKsR&ejCNHW0D$L?Y{& z-+!(ex(HR2t+z8ezJ13p%e#_pEzeK4BMasjYdnwJ?s{s>RYi;(GFI&ez1i>3C&Ca}f(#daZAeX;DyOc))i?10TL3cyU9coSu% z)362xbDc4*fSbB%;E}Qhj(t3Y;h0y44|>PJ{Wnh8tcJHxpgfb*cWZ`3JyR%I1eX@D zhJN6ES7k-g^nBRA6+Lz{hr;f=v9U%{mDsRr?D`_~Bl#>!q2hGpQrdVLN8GXyT@?Oz z+7-##Pu6db%j-<<)!r^-I*w+`UXXo=8fYr1;~kdxDdbuG7#=k3gnwVoy*Pse*8?CP{EM4h zidh=Lh5n)*Bd=X5LAxl_{94L7yp5$zMl^#3{EGwW`CdAsJZjClo-A{sUr5=r94uSOqPi$Bx2XqwJ0Xj>oqW`wvg3e07J z)B$YP_^e2>9eu-`H$Ilmv=_G3H|}OW)AT=Y3o&4{OAN>c@maT9IRH=16i5c13XIbq z?_;?AkC)0lxBghxk!|c14fTiYk`u>n>GBPA8g4GyLuofED1?CZ_ogTPNTn=*7Rz>` ztYLiEcuec^&_!PYFq(!2r=O|z#Ft@!biey30#@bucj&Ux#$iwD(4dukXbV1&1KhNt zqjE7f0?G2qOI$YC4rS^$^olhr+rM>-n6e_o=%}lSws$wIa?!a=?|&NpC0eU;zsWc{ zKWScWoG(_FP6kwvxFXl0Buq*6*?NA&t1N%CNhYQxAHE>%%=x@2fFKMkfr^ZzsGxgw zHOYm3bg4i(w(q;Sv}7U`6Pvh6Lm7Ptpx?d8)x>^1RQ1|vuG zD|M?Wx8nh#9<4K3^p^ z+tzr;24w)=wRU5Hn4~bhvixzHRwBbhqo*XVUm->z)SC1la7x*{$-z(_ak+M#+ZP-g zeJQP$k`X`41*d?oW^%4JZH z`2Ld2K6XjjT~MILbr{_xL>OfMd9;zSl*--#DwbWDM4Mp%J~N_%)nv|~hmH9@ z)4zT59opehPLJ^KJ6=VVO6Ho)jzT3gR98(NX*61GGVm56{>rj2M18)d^HhPG&iNrA zw3~jo&)!0v{GJE>iXI;ylLdla?z?dM+T(hzjWMg#Y7nngVIY8XR3%u9|L|1IzJP2% ztaEj&!l5_dWLvR#((N&`TQ%Z_aD4=0dIY1oPwr0rTMp@~krDj&qrOo>N7s>&+3e2# zj!pO%Q}P>}{K5VvO(@IX)|C%9o-~d!1M>GPa$u9}~ znp>y9i{nXjS1MjmdxDO!=)mKBA}-}mBp?edtY>D#G4~`p;ncc6ZCuu^9162{V`}LD z1V6Z{i6Rj#_U_ZbKJm)mqBY>vgDo5E5hb5`g#UiRW{W{;Yv3BoGE+jQunVS36Np2enN*G1C~E`p6Rz z{j*O=Hr16UEyfQtZjl0gr^mQuLw~P`!;^`G!)B$eL?HKwuJHg*Av%bEfp+&rI#S6h3Du-ZF@PZN*czynUs7CxrIzi<4CKW$DVUDZp4t@iHd(8;)$gN|7tcwXZ*}bVfvjelr52^Z_wbOI56$NP(NgU%iVHE zF4%~Q^N)t*gCUrPDQc6o=jm9zod2D$s=kAl9(4KKJVtQbESUEoxkjYn)gw9Hf$ zxkomp*C}$oszKqQn1bcxZ?n$jcvSPa-%+!QHfnAo#hP0@4Eu7#;_sn^YAFm}a5$kj zR>ve*17@Gze8vcL$w+E@`m5(=Qy*Q~>OI;1*l|C1v1PMWOw3eL9{)(8=9b%IM}6!U zMcHwGU&b-gd%E#WWQ1W&us{G=3v*Gczbd>nU~XA+RYmPvAK8xKIjItpke-|6DHLM? zV^m(Trt~H3n8Djvl-rxNw5%%KO97+CBR|c6-1xk!_MSCj{y-=Tl)@CO+~o2vmxWhz zY)tX$O0oEU#QKS|IP#9wssiTL8{ zJD#*5d^it313T$)w|h}CIq*lseFB}(#~)6PjENhpg?^8Xm3F>+oFgAQ0$j`746Z8! z4T=vRkY=AQ@V*?NvGxTdbw)Lbpeb#Hwn$jEkRl=kfxy_mFdnIZZ|v?Bk47e(MA6iu z=n4|Yw5OvgD)#Z`yIX8TL)kRf#j_X&X!Xre9^FbrGeT&ua~Ez z;L~rWd6P4b4BJHLCtjp5IQbK@Yn5nt(u(A%E*th>PWT9zmg#$i405 z1u)h5>|%j;2S6YL!d+3OTS>AcGo{_X_PRbtC^cZ~OZKy980j48DF6mKe0`ShOy)%)$Gpx&xYS;yMa=?0Luop+4>{HHg<+3oxDAZ>#%N{Z}PF z*y2V1MAwfF%&zQdI+R~N;zar?o(NK19$nQ&-1jXxMS(~iDLANz^Z7LflHO0Vw(S<{ z-&6fR1v zwYJ=7+&AQJkPZmGr9o>Y;3Dk51T1!Wny3eOhpSrVHP)r8j)n->0clR9O!IoX`MucX z^Zwiy9^o(E6WZ-Mwy{?YbRia8Hzm{^kdB^~ULyM{?&zV&+;rnvab{}4=iyZAU08O5 z@MV3_6vpe|{fU%r_^adQ-fn4Eqe6TqJ&VNr&47I)qv~~M_?mkg8<^CFz(S?YiNc&J zF8sSlt6VZ89ZVI=m8J2(?|e_Ih#V2_^euv=$5qy@pDZ&CU`?d!{!Xjfsm*EywtUo{ z7M`bz>Lm*;&#qzeUCS&pM&I|_VXXi>@ekIJ%6OP-srY6;NEm+rQ*Vm2SP=dpg`9v=8of0 zhOkW8VtL8ieYY31ECO7=oqO;@Z;1Kzm!dv@X`5}nJ?l8#I_`)#C(=D2%+*Mp=89va z5suRsE0+)2#jpPQZ^r2U4e8Tn0Artnx(_PB{~E_^f3atL)FN~J89#@h6>YEJ)SkOU zkVI-6t3OWeJfD)pik~F|nR+pV9ikPTcqm+4>0E|mxI>`U=jt6gulFW~I2NBxSomP5IND~h{YMxG>1blL@4eF^{6 zk#UK;tX9dPQL#jH$V)FC!y~DdwEKH`o^W&`SzLFoKY%L61$Od-y6mNtoA!97 zb$UD;2B^}_laB+luIKy^Uxf-;n)u=5i3a`beLsWhyI-)zg6gb&zCf*CjB)ffyJji% zQ#_YU?1e#Jdx!n7Hh(Ev_|i!PMx4H_`%&raZn?l8`iT!XKN5a0UHxpS!xGFhTNf+u zxpL3g?fDyTpaY9;=4WvgLgnhu2&s^lz|K2HyQC6^rpuT~EyYofcy)zpZBfmg;sx&O zRGn#3W%*3Vl+k#uA^Ks&Nc5APO_6`soi1jM35BHy;tl(LIhwN&a^a`% zJuR6vsJ;uQlY9qsuOj!Xa9BQ;(SH{l%NmW8PkWZ6_$5iF;!BonGJlTSxyBnademO^ z=BG6}B=bz`T7Vp#E}gK#fA=$pv=-a?p8toJvm3oLf8KYg@E~;oi^ZJM)2> z@1e%vtsSw=PX_(a0OjVZhHr*B{g*Z;W~=k%vO2VPhKypEAzx81pO<(tom_IjrLqcK z@zif`H$2E)oZV2FE-3#8Buc{2P?839y z#~{t0*VWy@2eH+)q@uti{*C6u-6YLDb<8RHk){z#wL@!tX{(Ia1#tU zO(0i6^t)Ueq*UJ0(u~F1x&}Xc`0a@7^&Z~y#d7ER`@Xo6>+bhoC<@-^DTsWJz$8O@ zE6q6A|0WUD9M91pzApN8$^s8ZUZ_$??^mFi= z)P@l=?Y7Lm!mt}6Km%zR(K8vzVwM*Cph$R+Xh-GcA2iO3omIER78s5{b=pAhf9(?R z=jTLCaD?TmqKr8OI!4VC|q!DCP=dfL9nazg#oyW^g{NeK5d*yo|ECkDoA zntQ_k_T^y$F%++7ZZFRcq6{D9EFXT)`{k$vq&&@tPha9Z|9HA1$islU8DzR5%*4N_ z8DoCrp?%1ld;gLWA)zb^F16gFV+?4T=W$cQrr1dXp6_NT3tcXsDig~m+Mf7k+r5{4 zaX`hIACv4>RGTPbo0)OUixoJ^mShV!I=(iQq{0#r8*6Kg8efTj75)X$kJM+F@I}UK zUT6J1kRe_tI*P!@&+7A20@t?a1+@I_mNWPu$Azzu(L~LJHNe5}vc>A0`ZhB2cvGWA zAP_fmED`+(u^vqY)Gl~k>@lkPE7s{N6Gk9+#3TqD@!RmIUBVFrL?mpY%{jeTz>Ul# z^|pRz2~Co5>*5Vwzu~?e8!^P_RU&lWU@N*MZCSemvk~1wkWh+dwBQK1rWMVqOV=%v zyWG(np@yt5?UDMjtFamAdo2)+Vqc~!LSR~pYD@3!kZumq_zz#BBQJsnet2!E?A{p`ONN@Od?Dfmf(G#Pe1^hkYDJ<6WdPOs0PO-=ay0Yv^)oC(`N1#}Xo#K?6jVUH7&Y zt>ZCkQuHIG`z{jqel&L+J&lkI(N??MwcMzEWwuRE z#tsWN%OvjLGsAyYLoMu0+RDt)d<{d{D_#C6oX)J(P(ujhKG$ds$ObrB+81&6HEUZK z6)Ny?G2BiOc@i>AX>0zY@kpJT=631_c&}0lHb_-=K+_#y@r3`Ov!QI~A~_f2I%EKS zJ7emxmhe7MD5eWcXU=kDb2?}pJ=Z2ixgDeUC)Cr0Dgk+Xip@@Z?fW>sIrSfsd3iKl zA^E2N1so#Rx-^RoLChMgcI*of<7Fc`lfSl9eUto$^I#>M(!@{8#<(NM6gWSTnNcQu z#F(n@U$X%u^XVuNW%;9#eG{+J3*S0eDbzOK*JU^!PFZiL;WG-ghu)JglFs-`?gx{0 zg^83973w0~@wfomO4tTCaGoAw5ySDvK+M&*B-%3F6upN6405V@Led1c)Ncz<5n^(< zPiF~QP0i%xt-8+Ans2bE6p^{j(LYwh4VMN%1BUMdM~=lxfRktQVj7Tbz4GRfJ5l**g7#AP{Ll zg26O4>0syPBWE01LOh{Vto8R|+hzwxjMgaa+4w4znI|@APC}4r- z_j-3kt^W!$v8L)O)(h**BQ`G7xTp5~*cIxL$Cbd!4BDqq1iG#K=~9AS9TqsXP_zV8f-mKeyL`)N8UjxE6ah@%CHBRjiYY zIAN$UE$g3x=1lf{!xt1ZRgVcz_({~_;ON4pN2KG(x>7`@%V6tAaX}iC*iEyNuh{Z< zP4>B7WT23ha(YzwOadZ+-><%ECjXix zzBw)Eg|m*luIPw7+s=KKGbMdlsp~TSU^^P?aC>~6uI9b%gV=r|dD=H2?R4lgxOouS z4y$=lE(0JSD=3f=cism*huq^ph>fl#eruAl&Qgel9&OAcjRk)uVWD_8t)s7{ATQEaE!nr^~>g=%(4pS8+Bu z5=G_(yS2XmS6Y<<>th8PL}|H=vh_+`rF++VgzfHwF?)POOi$iLaTu-X)KFZf9*u?T z?)}Bqv}%JriAp(#fdiklAyZx}iL7=i?Yhzm53*70oyBuUiOR%j>5$SfU;Qg4%7qnxL^X zxI~FR3wjB?`nEIEh!feQ?EB5fxl_Xbb9SubMyRuRRP*Q$EOTP&drs zY?gCRJmNK?Yq1V9fe#~p=K6_U2~;O55mu0P&seC38CLy+tv~Wp-x`q#x&>ZM7-Q3m zW4eD3)V})uA3+zGj}#kgL19Tb|B00v_HN*#OGe|6oD8pH=^tCuMly2~^7SO4!t)?g zy6@oxDL?L3C~3=371z7zURp$Z`m{P~37?4aXWtd&fX+CBi)^dH8Pk?Ah@fkZ()!`) z1X0mZu{ka}gkauYDfc*`#Qvobr##rtiB{U4PY809iCvoe)JrGm!V8g_?{9}sjEN0= zJ7$GABFPts7amvF?Wxg3FuLN{L~fQO2Ji_C{H?Zr zfx1Mn>%xiT^c%1j?7^kj&;Ai(5oLtxTKUAKq*cME;>S!-zWFU`$Q5Qp+Y6LFt-d-* zI|Yi&1D5sozyVt6Io*3Y(QbMUH6Zt&!WxBbM#)2%rL4sK&J$q7R-gKw`9Ij&^fvO1 zmLCl&t)>E;}-txS})rj3hexI>&3B9Lw&HJr;em;dp7)6hFSx)BpE zDm*>V;wiw)K?Qc**$GE$ELU6tzKjmkL)s zu`UM#;VDh!tL`#Wf4#+=lYGrdZfsUx$JbO|r4oTwXfXkrg4`=3V%Izv8Kj(cTDGhf z58eQ=^xq<-^r*F|u>!{;_Em2^J)jAy{&s94JOl#v*YHux@HFrJNuq_KMj{cd4B!GI zyQy%L+U#GslYhy*L&uRMJLVUA4D-9ylSREQjKgNu|8-2>jqtiW+McHya%Ket_zkuj zm3$1|$Ks(pmrP4EkMi}jCyvG+`NH5bpZ;|q0Y&){coHggK%;uf=ygA9Gjg@-z<06? z?Ix4=xt4ntSo@+9oy6cjU5elzD4ubWSy9ggSr6AII%4s~kygw*$EfUf zCY$#?+H2MPU7Dk3nKJ$X=vT#P&+45%*v>a)aJH8>pmXiNa08~ z#uSsyGl@a`R z?7#3vi8E>bKryMKl1od9W2}*q@jjedoHB;K~H? z;h{>mbHAOQ?yR})5R|r$vm|$Ig$ZnnL#t*8KK5}E+@@Kv8^(9yG?EfT5PU1;lin{4 zLSUjS%c8>N;~$k2tlzG=zm*BM6jW_WNrM*T)#Gz|3{UBi_BFO`gWIeuu7`n7qJ7!+ zoB49j8dB$kFYl%CkKkQAyfMq60G`QsWATY0(n^Qq0MqRiwat%l{@qJH_O!Wp|2ASV zp~0AbnKEINW-+Zod_NkA)8QUU3OqKkv6{WUv7umiWmN6w4y`-SB|;BV;-wz#^2nX) z8^1f-0CkHspK8Awr2)AoVa;dY`-(cwmj-#a^qWnY@txS1{hgb=G>2m7{0c%!)H zbs=y!>P=MRpDu#0WZhGe0hmuS$N;@)mS|*4E~72mu{YNq`RZI2l38VG; z&9cBZbIPwX$TSBPKN?k&W?cZIRHxaR)9Ch}tUXnw4yeb9w9Y0HH=?rzg6qU0_kXX& zikwQyBcGjclOqS5tHEx2(=0Eyz4ua<7wa614t|PXbnX@1^gG5j4}b>F?#wG?KKQUL z>D3%@JC#ak1x48sS1KpB=kC_yJ!7{-np@4coMInIt48ux_LJd7=zo%8N)@-)@Mqn? zCB{TDMLie03HEs~sahW1-369Q7k;$b8PLlKc|e60hl^hIV7!27_os@JOqQ{&^m@It zOU+BY&Kd1I<4B8XNEOGQ1fxpfLKZXxc`gJZVp(4YLUrlS#O3!F;mrX!*VfOSO6<2k839IE#ij3B6`Z33p3V1pDGW)z||*)hzf4| z#5U$te~Ik~F3J|qKe`0sb25<>Dt`sQv!b)&q;?!Q=CJ{$~>URp+ihe!Y>>%EQdotYO8c58TuoTLmkppGqebiz(tbA&68SL)Ym2PBh`wyUU1P7^o-V&xxQ%&I z1`AGU3bH%$YNuf^IXQ&0!Tdb$I7(3)vaau0-GLq!gUc*E{8a*|5H8aRMZy`Da}U;Q zk0R&2Y^ra(MQ4s*7$;THElm5vNno7%&0ev8lfW&rjt8_-k;O9IROA2H67|DhXL*}G z{w0f8xhpks8E$F1j1p#xIWiXZjBfJ5^IaMxafGqW`53v4sVk*MQwPMFnyS<{K}hw@ zd!wvxfN`sRTTVb-x;RGfyoA$6%2mq}{0gGHfQRm0M9P^IUVT{HFSh1`bO^0`(macn znI;iM@Bp<>s_2kCwF=JY z&PSrL)N>3YjsOL_gI3lipNJ2XrOul1Wkw+f?&ac$$3qPNi@V_?<6r^H*bU;p$q=>> z1HvZjJge#_N_vj=kpT^tx;Xz>wFlov3AYhs1rZwWA82QuRk~@988N#nN1&YOzQHa3 zdf&p06k#ycU~rClDyg|%s8er0F|}lCP@)Z{v-VzuK5T>5<=`+i$=M^zg~GidW{KUn8}&u^k7nksXyJs5{jRl`tHwI4aOs5*?ya1z}BK`rSbW4Osuwk zzP!7-44zPQ)BpsgLeCxhiNT!Za_>?f*CXJ;7qQ8#<^K=Qzj13VkbE1||LKI?N#Dgz zIi7fmv}!*%-MEm8-?F^f%-T%3{*>l+mq_<~HYo0Jb>M|QdT}F~I?YSYm@dv}bQ}pW zFz@r0I@tf#7ANC?Q{`j!@F@?F3h&0m;LY#5vfPc>xE`I4U`@X&vyP-}ERVw+47+d* z8XbB2sGh5Hd+{W*%UR)T5f#vG^F8`sWtR@~8)B#C?QQLfjP@T}EoZ+c4rjM1ZjGme zhKp}sp}+!o1JpBW<=)Fa2TtF_-3=QXVIC2VH8rTfxH&^ZfK)EC-4LZ7vc)gt$j;&B zE7W-upbGnrXAHR~IRB-K{$r25*hy70%I93yIb>rw9p~$V7vkmchY;zmv47Gi0=ZcH zG-@I9F(-ru^j;GAl76&@L6#uC9|fE$kv7K>zz5BI7}0z>OgtK##ulGp-w47^V_Q0# zHuCjpJ4->8o$~7oo7i0@Y29LCCw36}NAQbYir!Nzr3tMo>~3U^4m{^j8Rl_h?YAiE zyhGp#8%ummkQ_Q>=-qTpeH?NTNV6{LaTjc#b_*38J~BuhbgN0&^W0t-1FuH~#hhC1 zMn{OGvG&h*`xt>KB~YjZM>i)`Nsmwxb)F$$rXEa__@Kz+APMRe*11R#?(&0!_hWIB<((OfR`TY$wQ>BK1O&x$6u zr8b-zG`lss-&MNJ83H{XmrzrxLCXP3zx616s_fK!t~=CSlew@PbyMR61D!a| zsLmv7%{S${Gp_IQYMXWX0v;SV?H^Pci3At^G0=Y~46@dK00jSvyO&zMzs`y%Yr2nYu6#9Y z*YTghKFZCC2=D}K1OtgE7My;a;3 zsO4r0%Bicd7q4#W5 zm+Xvgm|hj6LwVVX7t5zeMTOVcpsaPSw)$>PxMkvdb`kyB?G;aGW z#kZjiTUtq)U|p}b)@9@V#PZn`{gD1Td0g|}}0wF!xU=e0hy zHyG}uiLnRu=sn7Bh;pjoG`}Fdv!-#0vnYegTi{ni^fepVyOqj1R+r>Tla9W>4-K)UAhR`x5(nWp0izSdW$kHF1DHoWrOr2;ljn~@&VXv z!;to}?n^_JyK~TjZwrA8#KSX8dc!BRrSBlPOK7p;OZB1NX&jQUu>3+zhyr%DF>>dQ*Vj_VXb;zX( zIo_JSRUrlqS!=a;rg_c(E}I>bRiUcdPKf);b;qbq#k^^jYshOCbOAWZUqIUnQ9fUKBEK$&)hXv1&qdgv{1eugEQ%@G{a2OG^*wXuG3b~$Zv(2H3pToh28)cr zYev{el;D$^5k5A9V}#~b=H}vTS^YGz0)O@*21Z`O0#~2 zm-{BE`O_B3P}xZhi&+@oFrY>(h+qjmQno+gN&jLJZ+iW2Lp5>J!M6k+iEQybApKr0+#up4UogAeoM;JMHEB#{VVG0Br(~t z;o<{@RWdi)*lYX)(Ellg;`^QWQ~uKqQPcgOc8HPgMHiIdN%#N6MK-MwOPxW=U$AsA z6hD+|C?>Sb{SdlDXTqSEA{YPm#r(u&eVOc1!~;E`A@Ii?TO5jMV6eX7U#3^=(KbJw z)X$50!WpsTP5Ivi;dG`1(Q1Mn-iO$?#VW{dUm|NOB$ja@QE>~Ph#-mNJ6c1`0#bL8 zb^;qmJmr$E@2Yme^?T-@OSlKhpe$oGY@`wSL5|#*;yAw8FZzoULOkIY8Pjn%67i)} z{l}6W>Jt9?%jS@lTQ|d>jTNm#PJF*w6sHB{j<*k-*5Fdj_2a$#24~eH=jC%yHg2!H z2%3x{t~SIdKK>vjhfqvt_1%^QjO=(dj%>Ar?0bGs(A!D$zPa)z8Xsv49exN|7Z+O6 zQRUPUKl+X}fXYb+dctaPfkRo?R%^7-e7>~5?UUu~$KcRYt<>l?;Xf7hVW&f)4q zKa7KyF0KN+?^GbTpU)om1euxNw5TZ^rg5oVeTlQIZ`^tJb~Uv$0s|b;q%9c&;>51e z`c$)xs>`*eJ`x%qvrd4^9+drc4!I@NZrA_@HNnBsw{F1nVkF8z@ec`yiSEx14LDqy z(CcM#uu}3QSi0JRtqHAV&yzQlJ+KM6c}f*i3M;DrmHQUjI^~nLX$wx^x;4UaN$FVt z&Pmm5(XnW;jlm;3k}hAJAQFW@KVom}CA#H78GhaYQ}k-~GuPz3;-Eyv;kR5brwy=B zfR_?_yPB`{N!wjgu_j5w7@JOz2OZGu;OL@xz{Q;*pimlR!y=pCB{#+t5AL z32+inpvA+yi-KRXMGWNP^XPZAp}vroYP!sQG4E;+EVla%vHfn| z#|EaX2j^1(GZJzEND26^KBVP~l=+=c^IXE0_g9ivk#n7S?_xastD0^LpRc#-hdV_0 z{hMS2L%wspFLfHM`AppRfmS04pWmTbkzSqy*eo0l65!CmA7^`)_Abq90Nw2}UpR@b zjs1LC*2a%m*M;*QVbo&1a^QSx7#|CBpliq@NG4%{O(l;%KbSXgE&km>;pUK}O#LY7 z=hKdHY$ZPGW{US;_!1p?0+d#mw!Ifo*>HlFQT4NYKK3Xx=%8w}&z^)%lHyL5sH{iV zQlF5F9w7Upp0rr!ed-P?%I&>cfx#jb4Y6>zdKeJ+1*oh}oS%=)jNbSgBl(?h{7G(h z3mkpT=d=<%WB8h)flTzOV9O;(hB8mL9kMwruOktKJSwJobi9HGFIkOasl_Se(vrzI z%21W_{aWwfkgZ?Y-lndle*~}utqX+WJJEhlYJk+i+X|(+ms~(05=ByP)FJUvJ{*@t zA0UuB_vEcydLk(s6&v~4(`h))Cc`&Un_zxPiYeT<0))7e6|JO%vqVKb!nHdzRPQar^O`$jL}9z zcyBLNT;7G_UWPng87D7{$!@~X$jtVVDcw-^!O--QX8KHDsde0RJpzB=Dl8CGTWVX; z@z}4;-=5e^MKO4ZMW5ZX8QNt)v*|Dr=8Jk=B(Ie^;aW$=do+aqxqxHb|3JavGWra5 zNAiX(4t5fh7us?>RPI#)^Zbo>R|As>Q{=BSGd7fKNt-!1Ld6HSL-vHePI$O@TM*k}5G zq^A`yH&o-#vlckF*uR?APTJx>E?Zt)SPk|+!z72N4GVeicTyS;X^jnzEuc?(yP3P> zlN08Op$ef~naLYyYId-keedgm&K9%#VT>(izue;kL43eqt%)>!v%o88*#7LwDY|8_^$tgu9%gq<`i_mHPkDUVII;Qtw_RB+^7)5mC zCO)~}!GYsE*TkmRB3s_G>oXw8BYs`AY5sg;!geCmo1pHIex)aefr2!DvA#Du%ja7g zDT(m(jrA81Ja0NuFR4?npk#SW=Tgxj>f>|w4$m(>3*rxA$Nt%CR|a@Y`UTlmzk=`{ zlJa_LK&$aMpOzv$X4@i{)S2$*P^*VjVyi`d>`~!zR2>ac!W2?GYJ3BA%brp-`1T-|#o!YMKf7{>+eEb0 z!+5L{^jYlE?$bB4;fJkXuVVT8R&evJ-b|1^3j?y}_DNB@Qmhb~hIpqim9aAH5+>c)Y#7rpKts zZ(aG`MU5}{u>`^kuxaA4s219GVYtPiXZkO2mN)&g2(3*85 zekL#c#|vwqsPkYpbh%Qbr(NbNcExY?J_YDh8Z^w7#{*G;43yA?l`OC2K^dy!{v6SH zps#MZGnK;JmJV?nmvaY@bw2t4~(UEcWEs7l%MB_4?CHKX#Q!hWaXl6K6-TZ5uZ=I^?gOjBz|5}r&69B4}z_Tzld}wGy z?b%1(T+-Xdvq{HW@~S}iyt|ILpxD5FIltHoy%AD(m<*-9OrjE#=`&No~h z&a5kP)IFE8-cJKe!KPQ3n0;~FRcZ(VX?#sEJ~^5!nlJ&J+T5a$c0~u#e_WY-)+gKH z{YZ+(zrN=FbSpBh_D&uHpO)o?oyPXOrEcIeidyH1wR_`PpE?ueekOtlSq zD?*Um=U0_Q!WP0HmHr;0IDH($;EznIm)1y;*Uw?!R{{+X&=}A!?<5WRid`iRdZFm& zs+(j3E{U~_ZlOKp%p!IEGLh?yA26aZ^Jr+5KW(&=>FBfk$ha?qWhR^MGpKL!pMKr$ z7H_L>>D%j8#_jS#U#seThtMgb!S$HNFps~D>K6`Rg~$(tc6ZusJL_?`wc1}j9}iJbfaut)2sEk--+ z=k|Oi_(_-E>o54eBAm;1UF`l>4h<9QdG{*Q0M!SAvrL_28w3V~MH0;eJIaZKp0v29 z&8L=>zYp8bg30W+;VutS>>Hof2%kjM&0lXT{-?6DY-;oS+IEp5#jUsnEAH+Nh2lYq z6{p2DxVr_~QrtbbyA>}E#e;i+;BI$*|C#3lJTLPmbIoM!wXeO`ah$Pjy|bz4_Md;~ z1pC@Ph0t7v#4(dWv8uN^bN0A5tHRbFqD*Fu_d=(_{>f6i@-qE|xqU7m}YZ4OB1(@)vk7l|OabbzUsXrP!g8cW9xZUgV;}&7FRAIGETMG4Qrc z@Rcqr_DE;dhCVJEZw@J!9ZD-9!<^q9?2uEH5}_nyZc=qL*#3$=-TZa@m4Q!GdR6EU zY4dEtdO*_8qNu$V?|#l`$KVS~E7m$SZ(prQt+X8QNkep)GiGJi&2G26{L^daEZycT z0URxTOx4*qHtXF&(amaJTQ(5U6R32nkJcNKyGF-aYJi~Z%rhUGfjt8)G*Wh_n4ye| zZ5=c1H%zCvqvVahp$pm$Mk{QSf!7M&wdCw*V-Q<*DX?w&RYHw_PTRtKRD@@Cq(zii zt?u$wxSHhJF;K4=a%UZOB&L8KibDYJ`ufhIDyMtYH6BaRStSfkIWkwNG4l&t(SxYJ zQLQF`xN_vv{wyCaZ3~tpf^SRmTuiH&-0V_x6!%xw6rX|$3ILYPtUHWr;FSzYX&~@n6+zWdPH@-*=_8Us~~T#}hK z@N*_gzQ2b^ zx|WPHq$4pL3fGtHr0?`l)0fV5ivu;yzAajO`U^J&U;4RYQ!5P{U%oY7%`9|Wu8HJR z(-Gzn&XqZwD{+HyJ~^i-OJi@j#631hjddE()`yHmt57~FEo>vS_;W7Y*7Lb0Gi9*k zn6ht=38plebV1|SSqhBbd_+mA=y(3l&F!asQT$W6(CKy$(Z6pXFhQ1N`5w<2w)15j z9odK{jC;(c*JjE&C%`l3cN!Le0fv>-!1OqU$tH*A(E5nr5y;}!?O8NyN;W%q`bp56 zU+>u<+@g=_-?kcDzZ^l_9i(>K7B!WJfCp&eEp4?|?8Okn+GS zT>@*2p4(bjyL&ZTzw2()&PmDGE;^QvR2&Le553=b5P>jB87lho$Vtn?l+lg~B_KxR z%i|BP(Tzs6D+sv*OVrpQ)s;aY6pdCX&P7H}O~-L`b=Z(tWHAO; z0SjPieNi@v_gnd&TD;`@&{MWwPoa7#qokn6N#O3+-VD&~dsROYw)rA29lkd|5CR?R z(!8zwUPV>1syn!rUoQL%UwMJqj8@iI;X5^}PCrMF*a9p@%>$|C_N5b0vDn?GY5L>x`NI|S=4kN8UhC-(u8?-#qWPk>GvY!hTsQ3>$elCyZ1wiFoM=6^zt53mj_xLUqaDdBI*2;&ndxKcbao6CnTY5RGuXUdiv!~|-l`MzV}lS+#5;zJ-IVr*U%&TC*u|K0^ zf}G~(aVqCOX;9?~8S=Gm%SmpysOc;p#n2FFL+_-qi6T)j0v^vhY97Ye|3{p8cC3X>-o}cd&TX9WRcHtwxh2 ze~ZRP9M;=crn5a~#rp#|M{`;hwOYO^_%dziUVV)v8hoIco9tODo1YCDXl76J4dkHs zJH|uIgFy6&y2vc=8okc|G~xbf@a2A>Qf0xX(+!2Zk80(AxF-L5UPvX<_+d%5TpcgA zqP!(O(--EWZ=7YKE7&jl`82e-2$#Mk=OIx_xb!V}a)k5)RX6E|4p?EoHZNl`{meSf zElFi(<2+}sSK$)C+7E*{(ZZzzZS(WOzBm*IT!+MBAvn<76h#vfH7~=|ALqjKjTES@TEU0!YgbU~Rgb@O-&BENqDu=QN24{s@~={eReNU6E4=aQBoJ zzI1Eo9rrzlHmvAQ_pT8W)21an2KAa{)?y|A9Zk|x!u5+hSi>x1juF3}FZ0^t#hMmC zTk?1Rl`0FN@ZeBs%}*q{pq``Yw4`h#`F;oqqW#>abcnmkW{@bmgRNqTdHaKWnv?Tp zi<)ShZ8BM%r9)_kR^mH7aE)~=mmKIomZ5(N1(F&ceUC9UU;u$?qz?Ai4JImr~!Cdmrn(% zu2>#`Z36A`a6A`pZ(gfeo5zzY_B*8_>1=Lqr=v;|31BZsE;nqdPa^0o;xzjW-b08E z47SBlhPGEj5|*c#eE35?#=v_aeN<$j4l4(6aT zV&CYNpDouTGUyjCi1~8B)6iSIvL3A4GX(q={PrkFOw>mrUyHOm+EBK(61DJlxa}Y= zvpU3;AD4pDx3XhMir#gllFyHPVBc%u< zIh0{{Q4)H6;%620zIBt{BW8Mru zG0#P>RK_T=)G^~KMfa>G_|GE;A7Ur_*1uo1drGF38p_<^QW3E@9}!zoGx1zVLFb-K zHy13$*N*>Fr3O2qj}Vp>9|2}$wy{d$r|=3x*aCzS9Hh&~kS6BdE*riDGG1ni$r0zd ze<-U=Y|AifYm)6-Eo)bqks@)j6Y`?jS8~tbvd~zfuO)y#p99oYEMYrOrVxuTKXkq% zdQ(gA)Yawm!!m77y7@&vM%W&6u2*-YH5aIZ)MSx_SwZiSXxb0S_J@S_V~BL@<_2(% zkVMalhm5@ruCOY%ErBRG&ASCDhgov5@S83%31k&=CA#qzFb$Y8gbK&gCJP?ur+n6j zScb?LO@*dr>ImjWXMO>#pm{~#CnB(8MdF)1=^*bI3CS#>z+%>Byr~@v=zrpf04!XI zpK=_lh{0lECh@VP*o-AD@r)(1tWldjgD#k zK}+*BMMMV%5!m_}!(R44G`0JHo9DMNDUT~4yOG|iC?^WL^W<wRcpCpbZ&tw1EYJ#W=mWT|tW?TCwVq%jHJ zZ5KS~_XlicZ_m(Ff<)Eoc0DP3&ouUna3KCA;!{(JC%^V#j@Ed2+jk$ox}cAWLT!99 zV{S6`Sf5E!pbkhSHnkJeY92(9pB&KAeQ|F$V7FrHKUySb>|r7t@3da`;o2QjYJ2-E zl0eP3xq|9^GmNV)WVzfw{1vw%PvvYEccJ>Gs@sPY-40sv7BDqlLtb5#lERvjv2ldvj|V@y1W4oI5q?6foBgkH zDYz9~77WRA3H2zcWY}o{$s6S{99wPto*QrJpn(T6L<$P;6w-mrj^)^~3HEc<$@Ar# zu1BGaCecg=ii9d_l>Il^PFOmd`>1Qb~ z3pRU>UF}qc89%|)49*uDH*dNz!cq)(u{KS#VmiE! z#z$qe#Kt&K+kHG1Um>#0Piq}wKO=`$y!#L z5HC`{Dt5pvJN{Q(&OJYw_BM@;#1qD%k!V==&6QG%_nQBu7E_mv2C|uqtcIY9)u)?u z5sMY0Mfw`Z@GsxhOV(Q6|9WU)zdO;8+(8lH8ZB*5YiQ-W8*FdZMDP=0dd^~WiJm*= zt~Pt=;N_Vj$9W{M!4vHQ`Or;(Pp*_=$=ODDBkM23%0^kMV@2tJ)=W7|KG%?Vh5QM~ zBj=eVy1zJvbxCgG*l|2zr?j|uzMwrE+R?2008;tk7UB*)p8qn#u z=9A~A=hKKPEt8%IZ3FM0FhuPIo5z8pGu~ccK>W+co^5~{Mp86<2`P0( zkvNV<88MPGlZyLr@fBNZEM3R{2{76PN$I^S*wI1vKvJB^%mra3@5vK!Vs5)Qz zKxPI(x%&x-3=2WCh;6hn0N5$<-9|5mWU`)a+#XO$g{YP!XE_8PCpi{~AdTh5f5M_O z<*Dk791XVh6f68NJVB?GLtFtlxQp?<{#8(I>vpkiIj+uXRt@t;heRS#P;ovb%XpWa zdJ30Pu{=JH(2$iHQWjNI824mVzTe)$?T-J1fKX*KY)`TS+`t;+iqol z;mIsxo+5kX)byt1*akU;m|zZjd(20OcPS>rCAaW{cUnf6CRwPR2=+ewrIo2y7W-VF zFS7;ROoKQKJNuVuOz-*04Uk0oNVFL3Gs&2L8Gh=7oE5ir@4{k%K!#b1l0~(2u0FP# zRfF1t6rL-;Vpogd0$)qGX)Z3?W*d+_errbl^3cWsSyyFJ2TWMw9<}NIV~* ziQ$ZD{ys+0Bst9nhuif+JFy#Wo^0?g5Y3&j?IsouNsM;Z`kKNO1HtFxceUu1Kd~|^xTRhcc#Vb;5k~SsfFAZZxddhLo0G} z3XQ~ZBoc-(SrF)~O4ph4UCZZ*ti_wXrY-J2jmOlhiis*74U^XGDF#o|;_$zr)y@`6 zw8?qBzn>RP5rBm6&UEAVOJscQxOX$9`!*+qW8l9X zl5jG0%>Q@h5l6c*)llUKpMT03e3@C%=U3Jh3r`HSr_QCkk)ozTX`djmh^}FCtR8ZG zR+$m09T<-8{%9MBmt0E=zEEQ%PkR|m(%sfdJkBxhXuIFF-ULoL!_6H$65z-;kZK!_ z1a9-*N96h_x>)h%1)TQ=r(d)K=^~jl86(6e6q79lPbq~DD9ExIoj{%4*L6b|9@16X zgV|}qEQj`+s8Tw1M}1<0ERG>reUIxh_04_#vknTAnY0br@_k|l!!<*`sA<3Lsk}U! z@UkCvjE;sa=?oLjDp?*_9iegDH>;F$iE8D6Z{!H^dd(}?mLaV<>&K5c>C$U%^qoD!jRaumQ)@)(XA$Dkz zT(^4T&z~#)WDnRDwi%qp4jcf1$$LhSF%#wxlY8}q2Bw6@OQdw8t9G^Y4RJ+9nT-Y6 z%O1u2;bw>3cGDkqMzqvv&h@`@i&zkP*uSDfu@mtLPZfv!NzXCNJ2|zh{CQRV-Lr3M z;Wdz@&=+h5z76>Bb5e|ZO-N`7ZT+cNYk$-(xkE(rpgMI)(I^#%)(-tkDB*fTcPXhp zWvU_bZ*%x?;7BvzwN;Zn+nzX3P)`);;cv~*ed&MKE?5$&W8?2sBO-h#U z`lPO986`baZ#KaN@%x(30G!*1gOlu~1wG&M#uUBg056Bpl^@=B7zSsPAADrwOX|h? z%==Mq`?o=+f0Y>lA=aeN0Z!%Q&y!LnBGVHP0*irT+o@UU`+OPh<7o+Ou`%iv#h|kW z>rLj=$8^D8fw7SK=;7J-+X&I~J{6!$SfdwNi{UOzQ2vT2(#g=~yxi>SjQ?|_q{{q5 z`(FbAoJULXX$^7Sl_^O-myl|I3$tfb>NGU(2xK?Z=KSN7f(@{Lx>MY=MO;G7F>K!z z-Dcl#iD@YGCE^3Y!`V6p4%P_b5{`a{7g`dB{IWl$iz2o|JuJUm%mzTSZ+2dV>L zQEWu8EL9JzW@1$jlXLDdjKH{?Wk)I-;(;o8Q~5Pi!XcEv5Ii zLotVmMB92J%aVoMyogWLL7dnwcfP$mi?8L{hk_~P7fPHD^@5_@hHP|J4WZt}On=i9 zQ%CstnpM(7#dutJlU61^2=elpvTXWQxmp2+<3lTr_E$E4ROWJGg9^c%@OMmNBHc<{ zU%|z;JF8#oEw%{2c-Eoc7O|@&k5f)`zriTO=`BbF2evcx?_>0iIKAimc}4GL zC#NJXniI{}M4=wMb|iedDvu{l?4W@C2L*l}13n4ks6cvT6Er}i ztbj9P%Qkq}1%O40*57v0tMK*wI{r_$dkv)k{N=2RD>8_LDw5s?hpI>D|x;Z$>2{`J`HFhlE-+GF!3m0x2SB_RP1;euJ+`AzN7xNlP#=50gBf~U_7@K{1mI2 z%l-*CsG7lZe?0Vc%nxxy7ubgq_LWae`ohJ+c_|(vIpgQvEO`-UXFKqDq-cE2M|)>m zzL*-%68p9`aaS7UeRc*GKG#jPo!trmsL6+%57dvN1a;4YFHWv`UX~7PJMQ~!&v`9c zGe<(GfF#s6E!3dzZ|fvA`9y;)nQSTiLg$wuS5BF4t$Uh#)CWugKj!wJ13Z+|pW#7e z^IqR_I?SjjS)ji8b#ONdmf}3o-ZU98-y1XE%b4fkvuLpn{y-rujITk4K-N1!D&I^F z2X}VEH>}(xtMXATDh5|O(Fs|QK{`!DR^1N#C|^e$_Z0Zdfx`AJE6KSJY7yNoHOZ#p z+*+C0Y`^QJwY_hiW#DU7{6@s;)ZvIdz*H~r&gZx~PI>-;B+u*Jh?*!*bw2wDO8D&< zis|h{r$JM8TD5)7v*=M{=h%6>m8Skx;#@my@P{oN zedyIcBCmG;^A9>O{?BgRP`{gm>^`q)nAc!Rc55tlI8Sw_CQ9AfAR%hYen|OCA`+(a zgl=SidR`qiAW#GpGdr~@r(DyH_FVyW_>VS+bq!R?fz%R7suiHW@xz{ z8CIvW51-`xL?0(B#qw+|BTTtYii*ml&&Ra*-%ZV$i@9Kl?{#I1i$fnj1<|;AOjut6 zg}a$irou3fbij@gEUR6u`zQr;Mlk=ATDe`mlpPbqRV{iLog(BP2|V{2AFmSLeT!x% z;Z{TE*7G zKtgA&5zq=-5h;Mal$0cS zR=;iPz&p#?@PR*1kXBO`ZKU*9etEeQ~p5yf^s1rwyc+V@2!>n4x#3^on?vQ1R;{;!(WHM2(v8>E>0~5u=l*uyjU@q z0%VeEVryW-eGNHWQ07XoAFmroP^W$4C<;n=9t*ihNb9PdNY7cj^j(A(`FcgoePUj9> zdPOl%N)P2ev>9A5M;(aUkGf68?n!gQC(&)6yz+2s6er7&7QHkT=k@ zAv#aC?b@|MG z^jL1#?NMpCTavL<2H;9ASBLzgA1s8Zeo+NTZ~!b4z#CdL$as|leV<&2rR+s9rnq%f zpiRyOAiutW1s4O*G_ZdG@}d0w_6%fHH1??TDDZjgPM!K2(Q9K{mezFk-yxGhT1sp` z5+&=+J{6R)wQnqC2r(nus{Sv0+nn~AgGglM%nf9-B#N2}e><&ezl8`obK*`0VG>(y zY~Yt=VPZb%07yc|V|lcEi7*>)|V$PmwLm zbN=#(XnC+0j572(e}1m50%Ul<`{n9eN1T9T826Ck_}2{I*~|#(O)p0^#{8It5^pFF zb&?q-ai=CnK|F1gD3hD_k3kJHU7j{SUn^007#_v0QO<9CFrq5_?qTZMRT<7-YHpJ) zT8+_pVKgI(^@+=(wal$=_oA4hN}w+#fJt(Z3=3aOpeXlQEDx_v`w1?9h+wX`m+ivb z+)ZTY>t}c9n-fNE`Jm+73>4iCCV59~xd%SAO!1LyGDEF>;QIW|_rDAl-l`WA1Vg57 z^xneX7r@l_5@Qrcs`J8^7UDb$-WeK4m+XtJ3&KZ7S6+^$NZ>!&X`PU!KTz#P_GAY} zUxpQ}?a#t%EUg?CV!M87FVE7`zVtM28Ws)SA`7AtOq~2ECaAz`qv_?F~C3qSUCUM{7shv@F07{vSkdl;zdrs${-|{2xkbS>yl! literal 0 HcmV?d00001 diff --git a/docs/_images/sphx_glr_torch_export_llama2_thumb.png b/docs/_images/sphx_glr_torch_export_llama2_thumb.png new file mode 100644 index 0000000000000000000000000000000000000000..8a5fed589d17fe6ee62d6a9788519dba471fbd65 GIT binary patch literal 26794 zcmd3N^;cV6&~zU%f(#fBBGMqQ6{mHpumQ z^~%~w0VJvEnR)nlooKF`yb|H}xA|C<^cUE&K&!B{#ar*%RrUI{fOnuSCL$03!--#| zXbLuXBIIR=G00EgrNBb5!{Wg0L?-(}Cu?Haf+0F7%3dlH!I7q$mL$Yecj9<`jSzfT z^D5Zgt0)_Z#Ecth&8aB^q#^tt+jp*g;+l3Kuta%Af{32Sylxr^V1hg&jmnR}mkXfv zAbI2;{|#XPvE=`E0PeA)C<&;HFH;EcKm_ss4Tt~~BwnThPbCg5(+AS2SjXlQXBFG6FYU#Fiz_d=HGDi{5xv>!;RVHaotlES;Y{+4GrXjSnu(wiwX$)Rm{msq2XKT(PDE-xEgi z_(VivIE9l($F=w;5%X502cyd9S;o)?@*@IUt%-#RKi)is3GxWf#e9B1a0p}-K^GkH zE#RcLH-hEomo}wtS-`xST`geg^-vL1i@Xk;hL(*ihFVL8OfGro z7;=ZwLXX%~LDV&-I)%f^Zc@iBUPr)DOdL&SY8Fe4jRr0M^?OY#0}B$nzJ&L1In9Al zQOK9T4G@xxBY?1&=&}M<$uJ+xfTKBW>PRAN(VR;B>we1G_F=pv`)g>`Z^uYRx_jEF zLgunT}hWZL=J)o}-tSc-H43?kvu z2VGc#wT9ESOGB&rE@vy9Zz3|&#)3Ae*}ua?Sht~cbom|GLpPlTSm&X z8g90~9$!qLFeLh!xc`P0$(R$Ja&h&6EezoO__ACMkpyr*-p^zc6Sxuga|EOJ~>i$53|i zk3uPR?4WZH-H$~ya%J$=ipQkS-owRa4tw(P{%YPDc|taynH>AT=I|^o0}49wU~C{J zauK2=H85m@3Fzliq*hcd5LX*{uUE#$CpBIVU20AE#`cp-cs73fa2V=^W|F4HJO04v zh?+`+`1w}s-C)?|2Q&Bi)%tp}=My{T_L}AL?X!5%*(mp1+HX=ICa0VpGHL53#n?|w zrsVN!1h2aM0663#mGa+wr{K=E&1QL;2u+|ml4!Vcv|H~i8eectf>jNDUkq^angTdW z)#m!O%sOuC%7f^!Qht!qCw#tfe81zkyyL#c-mQq;u_=hU8Y&X0P3ZX%SuC=j%cBp_ zr)IOhDL^juo$ys}!om$%Q`tjo6J*I-hLpYq;KY(EwP$7&><+R6Cs-uf#l&-#?&}Uc z2}fsB6#K32O!O58@ksbMFteP}G2(2lnU(Xx0EQ{|6Fyl`z1ErflfT{pGkLUjTO->; z)k=c&e_nqG{-=f;xxac&YcIobydWM*>-WCPDRj}<`M?J$>U-1W=e@wPjOqn`WB6h3 zK~jXSxO+anw%MV<(bYq5#QL@?&$9}E9mOt!yYPhfLyOPk<($)-DALZ(Z<${Ys2e!I zEf@cqbht`&4b(^c=S@?tyU^-%7nsg>47yyUusdaMN|9?WoX_ry|GMNc@rE*&1ZWvhRi z_S3m-n`bk(@O+$m8Kp&CbTqx#!3Ro^Y+#WD0wD>iyApzndvKsR&ejCNHW0D$L?Y{& z-+!(ex(HR2t+z8ezJ13p%e#_pEzeK4BMasjYdnwJ?s{s>RYi;(GFI&ez1i>3C&Ca}f(#daZAeX;DyOc))i?10TL3cyU9coSu% z)362xbDc4*fSbB%;E}Qhj(t3Y;h0y44|>PJ{Wnh8tcJHxpgfb*cWZ`3JyR%I1eX@D zhJN6ES7k-g^nBRA6+Lz{hr;f=v9U%{mDsRr?D`_~Bl#>!q2hGpQrdVLN8GXyT@?Oz z+7-##Pu6db%j-<<)!r^-I*w+`UXXo=8fYr1;~kdxDdbuG7#=k3gnwVoy*Pse*8?CP{EM4h zidh=Lh5n)*Bd=X5LAxl_{94L7yp5$zMl^#3{EGwW`CdAsJZjClo-A{sUr5=r94uSOqPi$Bx2XqwJ0Xj>oqW`wvg3e07J z)B$YP_^e2>9eu-`H$Ilmv=_G3H|}OW)AT=Y3o&4{OAN>c@maT9IRH=16i5c13XIbq z?_;?AkC)0lxBghxk!|c14fTiYk`u>n>GBPA8g4GyLuofED1?CZ_ogTPNTn=*7Rz>` ztYLiEcuec^&_!PYFq(!2r=O|z#Ft@!biey30#@bucj&Ux#$iwD(4dukXbV1&1KhNt zqjE7f0?G2qOI$YC4rS^$^olhr+rM>-n6e_o=%}lSws$wIa?!a=?|&NpC0eU;zsWc{ zKWScWoG(_FP6kwvxFXl0Buq*6*?NA&t1N%CNhYQxAHE>%%=x@2fFKMkfr^ZzsGxgw zHOYm3bg4i(w(q;Sv}7U`6Pvh6Lm7Ptpx?d8)x>^1RQ1|vuG zD|M?Wx8nh#9<4K3^p^ z+tzr;24w)=wRU5Hn4~bhvixzHRwBbhqo*XVUm->z)SC1la7x*{$-z(_ak+M#+ZP-g zeJQP$k`X`41*d?oW^%4JZH z`2Ld2K6XjjT~MILbr{_xL>OfMd9;zSl*--#DwbWDM4Mp%J~N_%)nv|~hmH9@ z)4zT59opehPLJ^KJ6=VVO6Ho)jzT3gR98(NX*61GGVm56{>rj2M18)d^HhPG&iNrA zw3~jo&)!0v{GJE>iXI;ylLdla?z?dM+T(hzjWMg#Y7nngVIY8XR3%u9|L|1IzJP2% ztaEj&!l5_dWLvR#((N&`TQ%Z_aD4=0dIY1oPwr0rTMp@~krDj&qrOo>N7s>&+3e2# zj!pO%Q}P>}{K5VvO(@IX)|C%9o-~d!1M>GPa$u9}~ znp>y9i{nXjS1MjmdxDO!=)mKBA}-}mBp?edtY>D#G4~`p;ncc6ZCuu^9162{V`}LD z1V6Z{i6Rj#_U_ZbKJm)mqBY>vgDo5E5hb5`g#UiRW{W{;Yv3BoGE+jQunVS36Np2enN*G1C~E`p6Rz z{j*O=Hr16UEyfQtZjl0gr^mQuLw~P`!;^`G!)B$eL?HKwuJHg*Av%bEfp+&rI#S6h3Du-ZF@PZN*czynUs7CxrIzi<4CKW$DVUDZp4t@iHd(8;)$gN|7tcwXZ*}bVfvjelr52^Z_wbOI56$NP(NgU%iVHE zF4%~Q^N)t*gCUrPDQc6o=jm9zod2D$s=kAl9(4KKJVtQbESUEoxkjYn)gw9Hf$ zxkomp*C}$oszKqQn1bcxZ?n$jcvSPa-%+!QHfnAo#hP0@4Eu7#;_sn^YAFm}a5$kj zR>ve*17@Gze8vcL$w+E@`m5(=Qy*Q~>OI;1*l|C1v1PMWOw3eL9{)(8=9b%IM}6!U zMcHwGU&b-gd%E#WWQ1W&us{G=3v*Gczbd>nU~XA+RYmPvAK8xKIjItpke-|6DHLM? zV^m(Trt~H3n8Djvl-rxNw5%%KO97+CBR|c6-1xk!_MSCj{y-=Tl)@CO+~o2vmxWhz zY)tX$O0oEU#QKS|IP#9wssiTL8{ zJD#*5d^it313T$)w|h}CIq*lseFB}(#~)6PjENhpg?^8Xm3F>+oFgAQ0$j`746Z8! z4T=vRkY=AQ@V*?NvGxTdbw)Lbpeb#Hwn$jEkRl=kfxy_mFdnIZZ|v?Bk47e(MA6iu z=n4|Yw5OvgD)#Z`yIX8TL)kRf#j_X&X!Xre9^FbrGeT&ua~Ez z;L~rWd6P4b4BJHLCtjp5IQbK@Yn5nt(u(A%E*th>PWT9zmg#$i405 z1u)h5>|%j;2S6YL!d+3OTS>AcGo{_X_PRbtC^cZ~OZKy980j48DF6mKe0`ShOy)%)$Gpx&xYS;yMa=?0Luop+4>{HHg<+3oxDAZ>#%N{Z}PF z*y2V1MAwfF%&zQdI+R~N;zar?o(NK19$nQ&-1jXxMS(~iDLANz^Z7LflHO0Vw(S<{ z-&6fR1v zwYJ=7+&AQJkPZmGr9o>Y;3Dk51T1!Wny3eOhpSrVHP)r8j)n->0clR9O!IoX`MucX z^Zwiy9^o(E6WZ-Mwy{?YbRia8Hzm{^kdB^~ULyM{?&zV&+;rnvab{}4=iyZAU08O5 z@MV3_6vpe|{fU%r_^adQ-fn4Eqe6TqJ&VNr&47I)qv~~M_?mkg8<^CFz(S?YiNc&J zF8sSlt6VZ89ZVI=m8J2(?|e_Ih#V2_^euv=$5qy@pDZ&CU`?d!{!Xjfsm*EywtUo{ z7M`bz>Lm*;&#qzeUCS&pM&I|_VXXi>@ekIJ%6OP-srY6;NEm+rQ*Vm2SP=dpg`9v=8of0 zhOkW8VtL8ieYY31ECO7=oqO;@Z;1Kzm!dv@X`5}nJ?l8#I_`)#C(=D2%+*Mp=89va z5suRsE0+)2#jpPQZ^r2U4e8Tn0Artnx(_PB{~E_^f3atL)FN~J89#@h6>YEJ)SkOU zkVI-6t3OWeJfD)pik~F|nR+pV9ikPTcqm+4>0E|mxI>`U=jt6gulFW~I2NBxSomP5IND~h{YMxG>1blL@4eF^{6 zk#UK;tX9dPQL#jH$V)FC!y~DdwEKH`o^W&`SzLFoKY%L61$Od-y6mNtoA!97 zb$UD;2B^}_laB+luIKy^Uxf-;n)u=5i3a`beLsWhyI-)zg6gb&zCf*CjB)ffyJji% zQ#_YU?1e#Jdx!n7Hh(Ev_|i!PMx4H_`%&raZn?l8`iT!XKN5a0UHxpS!xGFhTNf+u zxpL3g?fDyTpaY9;=4WvgLgnhu2&s^lz|K2HyQC6^rpuT~EyYofcy)zpZBfmg;sx&O zRGn#3W%*3Vl+k#uA^Ks&Nc5APO_6`soi1jM35BHy;tl(LIhwN&a^a`% zJuR6vsJ;uQlY9qsuOj!Xa9BQ;(SH{l%NmW8PkWZ6_$5iF;!BonGJlTSxyBnademO^ z=BG6}B=bz`T7Vp#E}gK#fA=$pv=-a?p8toJvm3oLf8KYg@E~;oi^ZJM)2> z@1e%vtsSw=PX_(a0OjVZhHr*B{g*Z;W~=k%vO2VPhKypEAzx81pO<(tom_IjrLqcK z@zif`H$2E)oZV2FE-3#8Buc{2P?839y z#~{t0*VWy@2eH+)q@uti{*C6u-6YLDb<8RHk){z#wL@!tX{(Ia1#tU zO(0i6^t)Ueq*UJ0(u~F1x&}Xc`0a@7^&Z~y#d7ER`@Xo6>+bhoC<@-^DTsWJz$8O@ zE6q6A|0WUD9M91pzApN8$^s8ZUZ_$??^mFi= z)P@l=?Y7Lm!mt}6Km%zR(K8vzVwM*Cph$R+Xh-GcA2iO3omIER78s5{b=pAhf9(?R z=jTLCaD?TmqKr8OI!4VC|q!DCP=dfL9nazg#oyW^g{NeK5d*yo|ECkDoA zntQ_k_T^y$F%++7ZZFRcq6{D9EFXT)`{k$vq&&@tPha9Z|9HA1$islU8DzR5%*4N_ z8DoCrp?%1ld;gLWA)zb^F16gFV+?4T=W$cQrr1dXp6_NT3tcXsDig~m+Mf7k+r5{4 zaX`hIACv4>RGTPbo0)OUixoJ^mShV!I=(iQq{0#r8*6Kg8efTj75)X$kJM+F@I}UK zUT6J1kRe_tI*P!@&+7A20@t?a1+@I_mNWPu$Azzu(L~LJHNe5}vc>A0`ZhB2cvGWA zAP_fmED`+(u^vqY)Gl~k>@lkPE7s{N6Gk9+#3TqD@!RmIUBVFrL?mpY%{jeTz>Ul# z^|pRz2~Co5>*5Vwzu~?e8!^P_RU&lWU@N*MZCSemvk~1wkWh+dwBQK1rWMVqOV=%v zyWG(np@yt5?UDMjtFamAdo2)+Vqc~!LSR~pYD@3!kZumq_zz#BBQJsnet2!E?A{p`ONN@Od?Dfmf(G#Pe1^hkYDJ<6WdPOs0PO-=ay0Yv^)oC(`N1#}Xo#K?6jVUH7&Y zt>ZCkQuHIG`z{jqel&L+J&lkI(N??MwcMzEWwuRE z#tsWN%OvjLGsAyYLoMu0+RDt)d<{d{D_#C6oX)J(P(ujhKG$ds$ObrB+81&6HEUZK z6)Ny?G2BiOc@i>AX>0zY@kpJT=631_c&}0lHb_-=K+_#y@r3`Ov!QI~A~_f2I%EKS zJ7emxmhe7MD5eWcXU=kDb2?}pJ=Z2ixgDeUC)Cr0Dgk+Xip@@Z?fW>sIrSfsd3iKl zA^E2N1so#Rx-^RoLChMgcI*of<7Fc`lfSl9eUto$^I#>M(!@{8#<(NM6gWSTnNcQu z#F(n@U$X%u^XVuNW%;9#eG{+J3*S0eDbzOK*JU^!PFZiL;WG-ghu)JglFs-`?gx{0 zg^83973w0~@wfomO4tTCaGoAw5ySDvK+M&*B-%3F6upN6405V@Led1c)Ncz<5n^(< zPiF~QP0i%xt-8+Ans2bE6p^{j(LYwh4VMN%1BUMdM~=lxfRktQVj7Tbz4GRfJ5l**g7#AP{Ll zg26O4>0syPBWE01LOh{Vto8R|+hzwxjMgaa+4w4znI|@APC}4r- z_j-3kt^W!$v8L)O)(h**BQ`G7xTp5~*cIxL$Cbd!4BDqq1iG#K=~9AS9TqsXP_zV8f-mKeyL`)N8UjxE6ah@%CHBRjiYY zIAN$UE$g3x=1lf{!xt1ZRgVcz_({~_;ON4pN2KG(x>7`@%V6tAaX}iC*iEyNuh{Z< zP4>B7WT23ha(YzwOadZ+-><%ECjXix zzBw)Eg|m*luIPw7+s=KKGbMdlsp~TSU^^P?aC>~6uI9b%gV=r|dD=H2?R4lgxOouS z4y$=lE(0JSD=3f=cism*huq^ph>fl#eruAl&Qgel9&OAcjRk)uVWD_8t)s7{ATQEaE!nr^~>g=%(4pS8+Bu z5=G_(yS2XmS6Y<<>th8PL}|H=vh_+`rF++VgzfHwF?)POOi$iLaTu-X)KFZf9*u?T z?)}Bqv}%JriAp(#fdiklAyZx}iL7=i?Yhzm53*70oyBuUiOR%j>5$SfU;Qg4%7qnxL^X zxI~FR3wjB?`nEIEh!feQ?EB5fxl_Xbb9SubMyRuRRP*Q$EOTP&drs zY?gCRJmNK?Yq1V9fe#~p=K6_U2~;O55mu0P&seC38CLy+tv~Wp-x`q#x&>ZM7-Q3m zW4eD3)V})uA3+zGj}#kgL19Tb|B00v_HN*#OGe|6oD8pH=^tCuMly2~^7SO4!t)?g zy6@oxDL?L3C~3=371z7zURp$Z`m{P~37?4aXWtd&fX+CBi)^dH8Pk?Ah@fkZ()!`) z1X0mZu{ka}gkauYDfc*`#Qvobr##rtiB{U4PY809iCvoe)JrGm!V8g_?{9}sjEN0= zJ7$GABFPts7amvF?Wxg3FuLN{L~fQO2Ji_C{H?Zr zfx1Mn>%xiT^c%1j?7^kj&;Ai(5oLtxTKUAKq*cME;>S!-zWFU`$Q5Qp+Y6LFt-d-* zI|Yi&1D5sozyVt6Io*3Y(QbMUH6Zt&!WxBbM#)2%rL4sK&J$q7R-gKw`9Ij&^fvO1 zmLCl&t)>E;}-txS})rj3hexI>&3B9Lw&HJr;em;dp7)6hFSx)BpE zDm*>V;wiw)K?Qc**$GE$ELU6tzKjmkL)s zu`UM#;VDh!tL`#Wf4#+=lYGrdZfsUx$JbO|r4oTwXfXkrg4`=3V%Izv8Kj(cTDGhf z58eQ=^xq<-^r*F|u>!{;_Em2^J)jAy{&s94JOl#v*YHux@HFrJNuq_KMj{cd4B!GI zyQy%L+U#GslYhy*L&uRMJLVUA4D-9ylSREQjKgNu|8-2>jqtiW+McHya%Ket_zkuj zm3$1|$Ks(pmrP4EkMi}jCyvG+`NH5bpZ;|q0Y&){coHggK%;uf=ygA9Gjg@-z<06? z?Ix4=xt4ntSo@+9oy6cjU5elzD4ubWSy9ggSr6AII%4s~kygw*$EfUf zCY$#?+H2MPU7Dk3nKJ$X=vT#P&+45%*v>a)aJH8>pmXiNa08~ z#uSsyGl@a`R z?7#3vi8E>bKryMKl1od9W2}*q@jjedoHB;K~H? z;h{>mbHAOQ?yR})5R|r$vm|$Ig$ZnnL#t*8KK5}E+@@Kv8^(9yG?EfT5PU1;lin{4 zLSUjS%c8>N;~$k2tlzG=zm*BM6jW_WNrM*T)#Gz|3{UBi_BFO`gWIeuu7`n7qJ7!+ zoB49j8dB$kFYl%CkKkQAyfMq60G`QsWATY0(n^Qq0MqRiwat%l{@qJH_O!Wp|2ASV zp~0AbnKEINW-+Zod_NkA)8QUU3OqKkv6{WUv7umiWmN6w4y`-SB|;BV;-wz#^2nX) z8^1f-0CkHspK8Awr2)AoVa;dY`-(cwmj-#a^qWnY@txS1{hgb=G>2m7{0c%!)H zbs=y!>P=MRpDu#0WZhGe0hmuS$N;@)mS|*4E~72mu{YNq`RZI2l38VG; z&9cBZbIPwX$TSBPKN?k&W?cZIRHxaR)9Ch}tUXnw4yeb9w9Y0HH=?rzg6qU0_kXX& zikwQyBcGjclOqS5tHEx2(=0Eyz4ua<7wa614t|PXbnX@1^gG5j4}b>F?#wG?KKQUL z>D3%@JC#ak1x48sS1KpB=kC_yJ!7{-np@4coMInIt48ux_LJd7=zo%8N)@-)@Mqn? zCB{TDMLie03HEs~sahW1-369Q7k;$b8PLlKc|e60hl^hIV7!27_os@JOqQ{&^m@It zOU+BY&Kd1I<4B8XNEOGQ1fxpfLKZXxc`gJZVp(4YLUrlS#O3!F;mrX!*VfOSO6<2k839IE#ij3B6`Z33p3V1pDGW)z||*)hzf4| z#5U$te~Ik~F3J|qKe`0sb25<>Dt`sQv!b)&q;?!Q=CJ{$~>URp+ihe!Y>>%EQdotYO8c58TuoTLmkppGqebiz(tbA&68SL)Ym2PBh`wyUU1P7^o-V&xxQ%&I z1`AGU3bH%$YNuf^IXQ&0!Tdb$I7(3)vaau0-GLq!gUc*E{8a*|5H8aRMZy`Da}U;Q zk0R&2Y^ra(MQ4s*7$;THElm5vNno7%&0ev8lfW&rjt8_-k;O9IROA2H67|DhXL*}G z{w0f8xhpks8E$F1j1p#xIWiXZjBfJ5^IaMxafGqW`53v4sVk*MQwPMFnyS<{K}hw@ zd!wvxfN`sRTTVb-x;RGfyoA$6%2mq}{0gGHfQRm0M9P^IUVT{HFSh1`bO^0`(macn znI;iM@Bp<>s_2kCwF=JY z&PSrL)N>3YjsOL_gI3lipNJ2XrOul1Wkw+f?&ac$$3qPNi@V_?<6r^H*bU;p$q=>> z1HvZjJge#_N_vj=kpT^tx;Xz>wFlov3AYhs1rZwWA82QuRk~@988N#nN1&YOzQHa3 zdf&p06k#ycU~rClDyg|%s8er0F|}lCP@)Z{v-VzuK5T>5<=`+i$=M^zg~GidW{KUn8}&u^k7nksXyJs5{jRl`tHwI4aOs5*?ya1z}BK`rSbW4Osuwk zzP!7-44zPQ)BpsgLeCxhiNT!Za_>?f*CXJ;7qQ8#<^K=Qzj13VkbE1||LKI?N#Dgz zIi7fmv}!*%-MEm8-?F^f%-T%3{*>l+mq_<~HYo0Jb>M|QdT}F~I?YSYm@dv}bQ}pW zFz@r0I@tf#7ANC?Q{`j!@F@?F3h&0m;LY#5vfPc>xE`I4U`@X&vyP-}ERVw+47+d* z8XbB2sGh5Hd+{W*%UR)T5f#vG^F8`sWtR@~8)B#C?QQLfjP@T}EoZ+c4rjM1ZjGme zhKp}sp}+!o1JpBW<=)Fa2TtF_-3=QXVIC2VH8rTfxH&^ZfK)EC-4LZ7vc)gt$j;&B zE7W-upbGnrXAHR~IRB-K{$r25*hy70%I93yIb>rw9p~$V7vkmchY;zmv47Gi0=ZcH zG-@I9F(-ru^j;GAl76&@L6#uC9|fE$kv7K>zz5BI7}0z>OgtK##ulGp-w47^V_Q0# zHuCjpJ4->8o$~7oo7i0@Y29LCCw36}NAQbYir!Nzr3tMo>~3U^4m{^j8Rl_h?YAiE zyhGp#8%ummkQ_Q>=-qTpeH?NTNV6{LaTjc#b_*38J~BuhbgN0&^W0t-1FuH~#hhC1 zMn{OGvG&h*`xt>KB~YjZM>i)`Nsmwxb)F$$rXEa__@Kz+APMRe*11R#?(&0!_hWIB<((OfR`TY$wQ>BK1O&x$6u zr8b-zG`lss-&MNJ83H{XmrzrxLCXP3zx616s_fK!t~=CSlew@PbyMR61D!a| zsLmv7%{S${Gp_IQYMXWX0v;SV?H^Pci3At^G0=Y~46@dK00jSvyO&zMzs`y%Yr2nYu6#9Y z*YTghKFZCC2=D}K1OtgE7My;a;3 zsO4r0%Bicd7q4#W5 zm+Xvgm|hj6LwVVX7t5zeMTOVcpsaPSw)$>PxMkvdb`kyB?G;aGW z#kZjiTUtq)U|p}b)@9@V#PZn`{gD1Td0g|}}0wF!xU=e0hy zHyG}uiLnRu=sn7Bh;pjoG`}Fdv!-#0vnYegTi{ni^fepVyOqj1R+r>Tla9W>4-K)UAhR`x5(nWp0izSdW$kHF1DHoWrOr2;ljn~@&VXv z!;to}?n^_JyK~TjZwrA8#KSX8dc!BRrSBlPOK7p;OZB1NX&jQUu>3+zhyr%DF>>dQ*Vj_VXb;zX( zIo_JSRUrlqS!=a;rg_c(E}I>bRiUcdPKf);b;qbq#k^^jYshOCbOAWZUqIUnQ9fUKBEK$&)hXv1&qdgv{1eugEQ%@G{a2OG^*wXuG3b~$Zv(2H3pToh28)cr zYev{el;D$^5k5A9V}#~b=H}vTS^YGz0)O@*21Z`O0#~2 zm-{BE`O_B3P}xZhi&+@oFrY>(h+qjmQno+gN&jLJZ+iW2Lp5>J!M6k+iEQybApKr0+#up4UogAeoM;JMHEB#{VVG0Br(~t z;o<{@RWdi)*lYX)(Ellg;`^QWQ~uKqQPcgOc8HPgMHiIdN%#N6MK-MwOPxW=U$AsA z6hD+|C?>Sb{SdlDXTqSEA{YPm#r(u&eVOc1!~;E`A@Ii?TO5jMV6eX7U#3^=(KbJw z)X$50!WpsTP5Ivi;dG`1(Q1Mn-iO$?#VW{dUm|NOB$ja@QE>~Ph#-mNJ6c1`0#bL8 zb^;qmJmr$E@2Yme^?T-@OSlKhpe$oGY@`wSL5|#*;yAw8FZzoULOkIY8Pjn%67i)} z{l}6W>Jt9?%jS@lTQ|d>jTNm#PJF*w6sHB{j<*k-*5Fdj_2a$#24~eH=jC%yHg2!H z2%3x{t~SIdKK>vjhfqvt_1%^QjO=(dj%>Ar?0bGs(A!D$zPa)z8Xsv49exN|7Z+O6 zQRUPUKl+X}fXYb+dctaPfkRo?R%^7-e7>~5?UUu~$KcRYt<>l?;Xf7hVW&f)4q zKa7KyF0KN+?^GbTpU)om1euxNw5TZ^rg5oVeTlQIZ`^tJb~Uv$0s|b;q%9c&;>51e z`c$)xs>`*eJ`x%qvrd4^9+drc4!I@NZrA_@HNnBsw{F1nVkF8z@ec`yiSEx14LDqy z(CcM#uu}3QSi0JRtqHAV&yzQlJ+KM6c}f*i3M;DrmHQUjI^~nLX$wx^x;4UaN$FVt z&Pmm5(XnW;jlm;3k}hAJAQFW@KVom}CA#H78GhaYQ}k-~GuPz3;-Eyv;kR5brwy=B zfR_?_yPB`{N!wjgu_j5w7@JOz2OZGu;OL@xz{Q;*pimlR!y=pCB{#+t5AL z32+inpvA+yi-KRXMGWNP^XPZAp}vroYP!sQG4E;+EVla%vHfn| z#|EaX2j^1(GZJzEND26^KBVP~l=+=c^IXE0_g9ivk#n7S?_xastD0^LpRc#-hdV_0 z{hMS2L%wspFLfHM`AppRfmS04pWmTbkzSqy*eo0l65!CmA7^`)_Abq90Nw2}UpR@b zjs1LC*2a%m*M;*QVbo&1a^QSx7#|CBpliq@NG4%{O(l;%KbSXgE&km>;pUK}O#LY7 z=hKdHY$ZPGW{US;_!1p?0+d#mw!Ifo*>HlFQT4NYKK3Xx=%8w}&z^)%lHyL5sH{iV zQlF5F9w7Upp0rr!ed-P?%I&>cfx#jb4Y6>zdKeJ+1*oh}oS%=)jNbSgBl(?h{7G(h z3mkpT=d=<%WB8h)flTzOV9O;(hB8mL9kMwruOktKJSwJobi9HGFIkOasl_Se(vrzI z%21W_{aWwfkgZ?Y-lndle*~}utqX+WJJEhlYJk+i+X|(+ms~(05=ByP)FJUvJ{*@t zA0UuB_vEcydLk(s6&v~4(`h))Cc`&Un_zxPiYeT<0))7e6|JO%vqVKb!nHdzRPQar^O`$jL}9z zcyBLNT;7G_UWPng87D7{$!@~X$jtVVDcw-^!O--QX8KHDsde0RJpzB=Dl8CGTWVX; z@z}4;-=5e^MKO4ZMW5ZX8QNt)v*|Dr=8Jk=B(Ie^;aW$=do+aqxqxHb|3JavGWra5 zNAiX(4t5fh7us?>RPI#)^Zbo>R|As>Q{=BSGd7fKNt-!1Ld6HSL-vHePI$O@TM*k}5G zq^A`yH&o-#vlckF*uR?APTJx>E?Zt)SPk|+!z72N4GVeicTyS;X^jnzEuc?(yP3P> zlN08Op$ef~naLYyYId-keedgm&K9%#VT>(izue;kL43eqt%)>!v%o88*#7LwDY|8_^$tgu9%gq<`i_mHPkDUVII;Qtw_RB+^7)5mC zCO)~}!GYsE*TkmRB3s_G>oXw8BYs`AY5sg;!geCmo1pHIex)aefr2!DvA#Du%ja7g zDT(m(jrA81Ja0NuFR4?npk#SW=Tgxj>f>|w4$m(>3*rxA$Nt%CR|a@Y`UTlmzk=`{ zlJa_LK&$aMpOzv$X4@i{)S2$*P^*VjVyi`d>`~!zR2>ac!W2?GYJ3BA%brp-`1T-|#o!YMKf7{>+eEb0 z!+5L{^jYlE?$bB4;fJkXuVVT8R&evJ-b|1^3j?y}_DNB@Qmhb~hIpqim9aAH5+>c)Y#7rpKts zZ(aG`MU5}{u>`^kuxaA4s219GVYtPiXZkO2mN)&g2(3*85 zekL#c#|vwqsPkYpbh%Qbr(NbNcExY?J_YDh8Z^w7#{*G;43yA?l`OC2K^dy!{v6SH zps#MZGnK;JmJV?nmvaY@bw2t4~(UEcWEs7l%MB_4?CHKX#Q!hWaXl6K6-TZ5uZ=I^?gOjBz|5}r&69B4}z_Tzld}wGy z?b%1(T+-Xdvq{HW@~S}iyt|ILpxD5FIltHoy%AD(m<*-9OrjE#=`&No~h z&a5kP)IFE8-cJKe!KPQ3n0;~FRcZ(VX?#sEJ~^5!nlJ&J+T5a$c0~u#e_WY-)+gKH z{YZ+(zrN=FbSpBh_D&uHpO)o?oyPXOrEcIeidyH1wR_`PpE?ueekOtlSq zD?*Um=U0_Q!WP0HmHr;0IDH($;EznIm)1y;*Uw?!R{{+X&=}A!?<5WRid`iRdZFm& zs+(j3E{U~_ZlOKp%p!IEGLh?yA26aZ^Jr+5KW(&=>FBfk$ha?qWhR^MGpKL!pMKr$ z7H_L>>D%j8#_jS#U#seThtMgb!S$HNFps~D>K6`Rg~$(tc6ZusJL_?`wc1}j9}iJbfaut)2sEk--+ z=k|Oi_(_-E>o54eBAm;1UF`l>4h<9QdG{*Q0M!SAvrL_28w3V~MH0;eJIaZKp0v29 z&8L=>zYp8bg30W+;VutS>>Hof2%kjM&0lXT{-?6DY-;oS+IEp5#jUsnEAH+Nh2lYq z6{p2DxVr_~QrtbbyA>}E#e;i+;BI$*|C#3lJTLPmbIoM!wXeO`ah$Pjy|bz4_Md;~ z1pC@Ph0t7v#4(dWv8uN^bN0A5tHRbFqD*Fu_d=(_{>f6i@-qE|xqU7m}YZ4OB1(@)vk7l|OabbzUsXrP!g8cW9xZUgV;}&7FRAIGETMG4Qrc z@Rcqr_DE;dhCVJEZw@J!9ZD-9!<^q9?2uEH5}_nyZc=qL*#3$=-TZa@m4Q!GdR6EU zY4dEtdO*_8qNu$V?|#l`$KVS~E7m$SZ(prQt+X8QNkep)GiGJi&2G26{L^daEZycT z0URxTOx4*qHtXF&(amaJTQ(5U6R32nkJcNKyGF-aYJi~Z%rhUGfjt8)G*Wh_n4ye| zZ5=c1H%zCvqvVahp$pm$Mk{QSf!7M&wdCw*V-Q<*DX?w&RYHw_PTRtKRD@@Cq(zii zt?u$wxSHhJF;K4=a%UZOB&L8KibDYJ`ufhIDyMtYH6BaRStSfkIWkwNG4l&t(SxYJ zQLQF`xN_vv{wyCaZ3~tpf^SRmTuiH&-0V_x6!%xw6rX|$3ILYPtUHWr;FSzYX&~@n6+zWdPH@-*=_8Us~~T#}hK z@N*_gzQ2b^ zx|WPHq$4pL3fGtHr0?`l)0fV5ivu;yzAajO`U^J&U;4RYQ!5P{U%oY7%`9|Wu8HJR z(-Gzn&XqZwD{+HyJ~^i-OJi@j#631hjddE()`yHmt57~FEo>vS_;W7Y*7Lb0Gi9*k zn6ht=38plebV1|SSqhBbd_+mA=y(3l&F!asQT$W6(CKy$(Z6pXFhQ1N`5w<2w)15j z9odK{jC;(c*JjE&C%`l3cN!Le0fv>-!1OqU$tH*A(E5nr5y;}!?O8NyN;W%q`bp56 zU+>u<+@g=_-?kcDzZ^l_9i(>K7B!WJfCp&eEp4?|?8Okn+GS zT>@*2p4(bjyL&ZTzw2()&PmDGE;^QvR2&Le553=b5P>jB87lho$Vtn?l+lg~B_KxR z%i|BP(Tzs6D+sv*OVrpQ)s;aY6pdCX&P7H}O~-L`b=Z(tWHAO; z0SjPieNi@v_gnd&TD;`@&{MWwPoa7#qokn6N#O3+-VD&~dsROYw)rA29lkd|5CR?R z(!8zwUPV>1syn!rUoQL%UwMJqj8@iI;X5^}PCrMF*a9p@%>$|C_N5b0vDn?GY5L>x`NI|S=4kN8UhC-(u8?-#qWPk>GvY!hTsQ3>$elCyZ1wiFoM=6^zt53mj_xLUqaDdBI*2;&ndxKcbao6CnTY5RGuXUdiv!~|-l`MzV}lS+#5;zJ-IVr*U%&TC*u|K0^ zf}G~(aVqCOX;9?~8S=Gm%SmpysOc;p#n2FFL+_-qi6T)j0v^vhY97Ye|3{p8cC3X>-o}cd&TX9WRcHtwxh2 ze~ZRP9M;=crn5a~#rp#|M{`;hwOYO^_%dziUVV)v8hoIco9tODo1YCDXl76J4dkHs zJH|uIgFy6&y2vc=8okc|G~xbf@a2A>Qf0xX(+!2Zk80(AxF-L5UPvX<_+d%5TpcgA zqP!(O(--EWZ=7YKE7&jl`82e-2$#Mk=OIx_xb!V}a)k5)RX6E|4p?EoHZNl`{meSf zElFi(<2+}sSK$)C+7E*{(ZZzzZS(WOzBm*IT!+MBAvn<76h#vfH7~=|ALqjKjTES@TEU0!YgbU~Rgb@O-&BENqDu=QN24{s@~={eReNU6E4=aQBoJ zzI1Eo9rrzlHmvAQ_pT8W)21an2KAa{)?y|A9Zk|x!u5+hSi>x1juF3}FZ0^t#hMmC zTk?1Rl`0FN@ZeBs%}*q{pq``Yw4`h#`F;oqqW#>abcnmkW{@bmgRNqTdHaKWnv?Tp zi<)ShZ8BM%r9)_kR^mH7aE)~=mmKIomZ5(N1(F&ceUC9UU;u$?qz?Ai4JImr~!Cdmrn(% zu2>#`Z36A`a6A`pZ(gfeo5zzY_B*8_>1=Lqr=v;|31BZsE;nqdPa^0o;xzjW-b08E z47SBlhPGEj5|*c#eE35?#=v_aeN<$j4l4(6aT zV&CYNpDouTGUyjCi1~8B)6iSIvL3A4GX(q={PrkFOw>mrUyHOm+EBK(61DJlxa}Y= zvpU3;AD4pDx3XhMir#gllFyHPVBc%u< zIh0{{Q4)H6;%620zIBt{BW8Mru zG0#P>RK_T=)G^~KMfa>G_|GE;A7Ur_*1uo1drGF38p_<^QW3E@9}!zoGx1zVLFb-K zHy13$*N*>Fr3O2qj}Vp>9|2}$wy{d$r|=3x*aCzS9Hh&~kS6BdE*riDGG1ni$r0zd ze<-U=Y|AifYm)6-Eo)bqks@)j6Y`?jS8~tbvd~zfuO)y#p99oYEMYrOrVxuTKXkq% zdQ(gA)Yawm!!m77y7@&vM%W&6u2*-YH5aIZ)MSx_SwZiSXxb0S_J@S_V~BL@<_2(% zkVMalhm5@ruCOY%ErBRG&ASCDhgov5@S83%31k&=CA#qzFb$Y8gbK&gCJP?ur+n6j zScb?LO@*dr>ImjWXMO>#pm{~#CnB(8MdF)1=^*bI3CS#>z+%>Byr~@v=zrpf04!XI zpK=_lh{0lECh@VP*o-AD@r)(1tWldjgD#k zK}+*BMMMV%5!m_}!(R44G`0JHo9DMNDUT~4yOG|iC?^WL^W<wRcpCpbZ&tw1EYJ#W=mWT|tW?TCwVq%jHJ zZ5KS~_XlicZ_m(Ff<)Eoc0DP3&ouUna3KCA;!{(JC%^V#j@Ed2+jk$ox}cAWLT!99 zV{S6`Sf5E!pbkhSHnkJeY92(9pB&KAeQ|F$V7FrHKUySb>|r7t@3da`;o2QjYJ2-E zl0eP3xq|9^GmNV)WVzfw{1vw%PvvYEccJ>Gs@sPY-40sv7BDqlLtb5#lERvjv2ldvj|V@y1W4oI5q?6foBgkH zDYz9~77WRA3H2zcWY}o{$s6S{99wPto*QrJpn(T6L<$P;6w-mrj^)^~3HEc<$@Ar# zu1BGaCecg=ii9d_l>Il^PFOmd`>1Qb~ z3pRU>UF}qc89%|)49*uDH*dNz!cq)(u{KS#VmiE! z#z$qe#Kt&K+kHG1Um>#0Piq}wKO=`$y!#L z5HC`{Dt5pvJN{Q(&OJYw_BM@;#1qD%k!V==&6QG%_nQBu7E_mv2C|uqtcIY9)u)?u z5sMY0Mfw`Z@GsxhOV(Q6|9WU)zdO;8+(8lH8ZB*5YiQ-W8*FdZMDP=0dd^~WiJm*= zt~Pt=;N_Vj$9W{M!4vHQ`Or;(Pp*_=$=ODDBkM23%0^kMV@2tJ)=W7|KG%?Vh5QM~ zBj=eVy1zJvbxCgG*l|2zr?j|uzMwrE+R?2008;tk7UB*)p8qn#u z=9A~A=hKKPEt8%IZ3FM0FhuPIo5z8pGu~ccK>W+co^5~{Mp86<2`P0( zkvNV<88MPGlZyLr@fBNZEM3R{2{76PN$I^S*wI1vKvJB^%mra3@5vK!Vs5)Qz zKxPI(x%&x-3=2WCh;6hn0N5$<-9|5mWU`)a+#XO$g{YP!XE_8PCpi{~AdTh5f5M_O z<*Dk791XVh6f68NJVB?GLtFtlxQp?<{#8(I>vpkiIj+uXRt@t;heRS#P;ovb%XpWa zdJ30Pu{=JH(2$iHQWjNI824mVzTe)$?T-J1fKX*KY)`TS+`t;+iqol z;mIsxo+5kX)byt1*akU;m|zZjd(20OcPS>rCAaW{cUnf6CRwPR2=+ewrIo2y7W-VF zFS7;ROoKQKJNuVuOz-*04Uk0oNVFL3Gs&2L8Gh=7oE5ir@4{k%K!#b1l0~(2u0FP# zRfF1t6rL-;Vpogd0$)qGX)Z3?W*d+_errbl^3cWsSyyFJ2TWMw9<}NIV~* ziQ$ZD{ys+0Bst9nhuif+JFy#Wo^0?g5Y3&j?IsouNsM;Z`kKNO1HtFxceUu1Kd~|^xTRhcc#Vb;5k~SsfFAZZxddhLo0G} z3XQ~ZBoc-(SrF)~O4ph4UCZZ*ti_wXrY-J2jmOlhiis*74U^XGDF#o|;_$zr)y@`6 zw8?qBzn>RP5rBm6&UEAVOJscQxOX$9`!*+qW8l9X zl5jG0%>Q@h5l6c*)llUKpMT03e3@C%=U3Jh3r`HSr_QCkk)ozTX`djmh^}FCtR8ZG zR+$m09T<-8{%9MBmt0E=zEEQ%PkR|m(%sfdJkBxhXuIFF-ULoL!_6H$65z-;kZK!_ z1a9-*N96h_x>)h%1)TQ=r(d)K=^~jl86(6e6q79lPbq~DD9ExIoj{%4*L6b|9@16X zgV|}qEQj`+s8Tw1M}1<0ERG>reUIxh_04_#vknTAnY0br@_k|l!!<*`sA<3Lsk}U! z@UkCvjE;sa=?oLjDp?*_9iegDH>;F$iE8D6Z{!H^dd(}?mLaV<>&K5c>C$U%^qoD!jRaumQ)@)(XA$Dkz zT(^4T&z~#)WDnRDwi%qp4jcf1$$LhSF%#wxlY8}q2Bw6@OQdw8t9G^Y4RJ+9nT-Y6 z%O1u2;bw>3cGDkqMzqvv&h@`@i&zkP*uSDfu@mtLPZfv!NzXCNJ2|zh{CQRV-Lr3M z;Wdz@&=+h5z76>Bb5e|ZO-N`7ZT+cNYk$-(xkE(rpgMI)(I^#%)(-tkDB*fTcPXhp zWvU_bZ*%x?;7BvzwN;Zn+nzX3P)`);;cv~*ed&MKE?5$&W8?2sBO-h#U z`lPO986`baZ#KaN@%x(30G!*1gOlu~1wG&M#uUBg056Bpl^@=B7zSsPAADrwOX|h? z%==Mq`?o=+f0Y>lA=aeN0Z!%Q&y!LnBGVHP0*irT+o@UU`+OPh<7o+Ou`%iv#h|kW z>rLj=$8^D8fw7SK=;7J-+X&I~J{6!$SfdwNi{UOzQ2vT2(#g=~yxi>SjQ?|_q{{q5 z`(FbAoJULXX$^7Sl_^O-myl|I3$tfb>NGU(2xK?Z=KSN7f(@{Lx>MY=MO;G7F>K!z z-Dcl#iD@YGCE^3Y!`V6p4%P_b5{`a{7g`dB{IWl$iz2o|JuJUm%mzTSZ+2dV>L zQEWu8EL9JzW@1$jlXLDdjKH{?Wk)I-;(;o8Q~5Pi!XcEv5Ii zLotVmMB92J%aVoMyogWLL7dnwcfP$mi?8L{hk_~P7fPHD^@5_@hHP|J4WZt}On=i9 zQ%CstnpM(7#dutJlU61^2=elpvTXWQxmp2+<3lTr_E$E4ROWJGg9^c%@OMmNBHc<{ zU%|z;JF8#oEw%{2c-Eoc7O|@&k5f)`zriTO=`BbF2evcx?_>0iIKAimc}4GL zC#NJXniI{}M4=wMb|iedDvu{l?4W@C2L*l}13n4ks6cvT6Er}i ztbj9P%Qkq}1%O40*57v0tMK*wI{r_$dkv)k{N=2RD>8_LDw5s?hpI>D|x;Z$>2{`J`HFhlE-+GF!3m0x2SB_RP1;euJ+`AzN7xNlP#=50gBf~U_7@K{1mI2 z%l-*CsG7lZe?0Vc%nxxy7ubgq_LWae`ohJ+c_|(vIpgQvEO`-UXFKqDq-cE2M|)>m zzL*-%68p9`aaS7UeRc*GKG#jPo!trmsL6+%57dvN1a;4YFHWv`UX~7PJMQ~!&v`9c zGe<(GfF#s6E!3dzZ|fvA`9y;)nQSTiLg$wuS5BF4t$Uh#)CWugKj!wJ13Z+|pW#7e z^IqR_I?SjjS)ji8b#ONdmf}3o-ZU98-y1XE%b4fkvuLpn{y-rujITk4K-N1!D&I^F z2X}VEH>}(xtMXATDh5|O(Fs|QK{`!DR^1N#C|^e$_Z0Zdfx`AJE6KSJY7yNoHOZ#p z+*+C0Y`^QJwY_hiW#DU7{6@s;)ZvIdz*H~r&gZx~PI>-;B+u*Jh?*!*bw2wDO8D&< zis|h{r$JM8TD5)7v*=M{=h%6>m8Skx;#@my@P{oN zedyIcBCmG;^A9>O{?BgRP`{gm>^`q)nAc!Rc55tlI8Sw_CQ9AfAR%hYen|OCA`+(a zgl=SidR`qiAW#GpGdr~@r(DyH_FVyW_>VS+bq!R?fz%R7suiHW@xz{ z8CIvW51-`xL?0(B#qw+|BTTtYii*ml&&Ra*-%ZV$i@9Kl?{#I1i$fnj1<|;AOjut6 zg}a$irou3fbij@gEUR6u`zQr;Mlk=ATDe`mlpPbqRV{iLog(BP2|V{2AFmSLeT!x% z;Z{TE*7G zKtgA&5zq=-5h;Mal$0cS zR=;iPz&p#?@PR*1kXBO`ZKU*9etEeQ~p5yf^s1rwyc+V@2!>n4x#3^on?vQ1R;{;!(WHM2(v8>E>0~5u=l*uyjU@q z0%VeEVryW-eGNHWQ07XoAFmroP^W$4C<;n=9t*ihNb9PdNY7cj^j(A(`FcgoePUj9> zdPOl%N)P2ev>9A5M;(aUkGf68?n!gQC(&)6yz+2s6er7&7QHkT=k@ zAv#aC?b@|MG z^jL1#?NMpCTavL<2H;9ASBLzgA1s8Zeo+NTZ~!b4z#CdL$as|leV<&2rR+s9rnq%f zpiRyOAiutW1s4O*G_ZdG@}d0w_6%fHH1??TDDZjgPM!K2(Q9K{mezFk-yxGhT1sp` z5+&=+J{6R)wQnqC2r(nus{Sv0+nn~AgGglM%nf9-B#N2}e><&ezl8`obK*`0VG>(y zY~Yt=VPZb%07yc|V|lcEi7*>)|V$PmwLm zbN=#(XnC+0j572(e}1m50%Ul<`{n9eN1T9T826Ck_}2{I*~|#(O)p0^#{8It5^pFF zb&?q-ai=CnK|F1gD3hD_k3kJHU7j{SUn^007#_v0QO<9CFrq5_?qTZMRT<7-YHpJ) zT8+_pVKgI(^@+=(wal$=_oA4hN}w+#fJt(Z3=3aOpeXlQEDx_v`w1?9h+wX`m+ivb z+)ZTY>t}c9n-fNE`Jm+73>4iCCV59~xd%SAO!1LyGDEF>;QIW|_rDAl-l`WA1Vg57 z^xneX7r@l_5@Qrcs`J8^7UDb$-WeK4m+XtJ3&KZ7S6+^$NZ>!&X`PU!KTz#P_GAY} zUxpQ}?a#t%EUg?CV!M87FVE7`zVtM28Ws)SA`7AtOq~2ECaAz`qv_?F~C3qSUCUM{7shv@F07{vSkdl;zdrs${-|{2xkbS>yl! literal 0 HcmV?d00001 diff --git a/docs/_modules/index.html b/docs/_modules/index.html index 2cf3f16ce4..50922a5fd6 100644 --- a/docs/_modules/index.html +++ b/docs/_modules/index.html @@ -9,7 +9,7 @@ - Overview: module code — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Overview: module code — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

      @@ -841,7 +843,7 @@

      Source code for torch_tensorrt._compile

               torchtrt_kwarg_inputs = prepare_inputs(kwarg_inputs)
       
               exp_program = torch_tensorrt.dynamo.trace(
      -            module, torchtrt_arg_inputs, kwarg_inputs=torchtrt_kwarg_inputs**kwargs
      +            module, torchtrt_arg_inputs, kwarg_inputs=torchtrt_kwarg_inputs, **kwargs
               )
       
               return dynamo_convert_exported_program_to_serialized_trt_engine(
      diff --git a/docs/_modules/torch_tensorrt/_enums.html b/docs/_modules/torch_tensorrt/_enums.html
      index 153d62a6f4..476a682a2b 100644
      --- a/docs/_modules/torch_tensorrt/_enums.html
      +++ b/docs/_modules/torch_tensorrt/_enums.html
      @@ -9,7 +9,7 @@
         
         
         
      -  torch_tensorrt._enums — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation
      +  torch_tensorrt._enums — Torch-TensorRT v2.5.0.dev0+8759736 documentation
         
       
         
      @@ -272,7 +272,7 @@
                     
                     
                       
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/dynamo/_compiler.html b/docs/_modules/torch_tensorrt/dynamo/_compiler.html index 76e9f99b56..ed23a6f40b 100644 --- a/docs/_modules/torch_tensorrt/dynamo/_compiler.html +++ b/docs/_modules/torch_tensorrt/dynamo/_compiler.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo._compiler — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.dynamo._compiler — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      @@ -471,6 +473,7 @@

      Source code for torch_tensorrt.dynamo._compiler

      < dryrun_stats_display, parse_non_trt_nodes, ) +from torch_tensorrt.dynamo._engine_cache import BaseEngineCache, DiskEngineCache from torch_tensorrt.dynamo.conversion import ( CompilationSettings, UnsupportedOperatorException, @@ -488,8 +491,7 @@

      Source code for torch_tensorrt.dynamo._compiler

      < ) from torch_tensorrt.dynamo.utils import ( get_flat_args_with_check, - get_torch_inputs, - parse_complex_tensor_structs, + parse_graph_io, prepare_inputs, set_log_level, to_torch_device, @@ -536,6 +538,11 @@

      Source code for torch_tensorrt.dynamo._compiler

      < hardware_compatible: bool = _defaults.HARDWARE_COMPATIBLE, timing_cache_path: str = _defaults.TIMING_CACHE_PATH, lazy_engine_init: bool = _defaults.LAZY_ENGINE_INIT, + cache_built_engines: bool = _defaults.CACHE_BUILT_ENGINES, + reuse_cached_engines: bool = _defaults.REUSE_CACHED_ENGINES, + engine_cache_dir: str = _defaults.ENGINE_CACHE_DIR, + engine_cache_size: int = _defaults.ENGINE_CACHE_SIZE, + custom_engine_cache: Optional[BaseEngineCache] = _defaults.CUSTOM_ENGINE_CACHE, **kwargs: Any, ) -> torch.fx.GraphModule: """Compile an ExportedProgram module for NVIDIA GPUs using TensorRT @@ -601,6 +608,11 @@

      Source code for torch_tensorrt.dynamo._compiler

      < hardware_compatible (bool): Build the TensorRT engines compatible with GPU architectures other than that of the GPU on which the engine was built (currently works for NVIDIA Ampere and newer) timing_cache_path (str): Path to the timing cache if it exists (or) where it will be saved after compilation lazy_engine_init (bool): Defer setting up engines until the compilation of all engines is complete. Can allow larger models with multiple graph breaks to compile but can lead to oversubscription of GPU memory at runtime. + cache_built_engines (bool): Whether to save the compiled TRT engines to storage + reuse_cached_engines (bool): Whether to load the compiled TRT engines from storage + engine_cache_dir (Optional[str]): Directory to store the cached TRT engines + engine_cache_size (Optional[int]): Maximum hard-disk space (bytes) to use for the engine cache, default is 1GB. If the cache exceeds this size, the oldest engines will be removed by default + custom_engine_cache (Optional[BaseEngineCache]): Engine cache instance to use for saving and loading engines. Users can provide their own engine cache by inheriting from BaseEngineCache. If used, engine_cache_dir and engine_cache_size will be ignored. **kwargs: Any, Returns: torch.fx.GraphModule: Compiled FX Module, when run it will execute via TensorRT @@ -673,10 +685,22 @@

      Source code for torch_tensorrt.dynamo._compiler

      < ) gm = exported_program.module() logger.debug("Input graph: " + str(gm.graph)) + # Apply lowering on the graph module gm = post_lowering(gm) logger.debug("Lowered Input graph: " + str(gm.graph)) + engine_cache = None + if cache_built_engines or reuse_cached_engines: + assert ( + make_refitable + ), "Engine caching requires make_refitable to be set to True" + engine_cache = ( + custom_engine_cache + if custom_engine_cache is not None + else DiskEngineCache(engine_cache_dir, engine_cache_size) + ) + compilation_options = { "enabled_precisions": ( enabled_precisions if enabled_precisions else _defaults.ENABLED_PRECISIONS @@ -710,11 +734,15 @@

      Source code for torch_tensorrt.dynamo._compiler

      < "hardware_compatible": hardware_compatible, "timing_cache_path": timing_cache_path, "lazy_engine_init": lazy_engine_init, + "cache_built_engines": cache_built_engines, + "reuse_cached_engines": reuse_cached_engines, } settings = CompilationSettings(**compilation_options) logger.info("Compilation Settings: %s\n", settings) - trt_gm = compile_module(gm, trt_arg_inputs, trt_kwarg_inputs, settings) + trt_gm = compile_module( + gm, trt_arg_inputs, trt_kwarg_inputs, settings, engine_cache + ) return trt_gm
      @@ -723,6 +751,7 @@

      Source code for torch_tensorrt.dynamo._compiler

      < sample_arg_inputs: Sequence[Input], sample_kwarg_inputs: Optional[dict[Any, Any]] = None, settings: CompilationSettings = CompilationSettings(), + engine_cache: Optional[BaseEngineCache] = None, ) -> torch.fx.GraphModule: """Compile a traced FX module @@ -733,6 +762,7 @@

      Source code for torch_tensorrt.dynamo._compiler

      < arg_inputs: Inputs to the module kwarg_inputs: kwargs to the module settings: Compilation settings + engine_cache: Engine cache instance to store/load compiled engines Returns: Compiled FX GraphModule """ @@ -752,14 +782,6 @@

      Source code for torch_tensorrt.dynamo._compiler

      < dryrun_tracker.total_ops_in_graph = total_ops dryrun_tracker.supported_ops_in_graph = num_supported_ops - dryrun_tracker.graph_input_shapes = parse_complex_tensor_structs( - sample_arg_inputs, - "shape", - lambda x: dict(x) if isinstance(x, dict) else tuple(x), - ) - dryrun_tracker.graph_input_dtypes = parse_complex_tensor_structs( - sample_arg_inputs, "dtype", lambda t: t.to(torch.dtype, use_default=True) - ) dryrun_tracker.compilation_settings = settings if settings.dryrun and settings.min_block_size > 1: @@ -846,6 +868,11 @@

      Source code for torch_tensorrt.dynamo._compiler

      < # Criteria for a module to be convertible to TRT if settings.use_fast_partitioner and "_run_on_acc" not in name: dryrun_tracker.to_run_in_torch.extend(parse_non_trt_nodes(submodule)) + logger.debug( + "Submodule in PyTorch: %s\n %s", + str(name), + str(submodule.graph), + ) continue subgraph_data = PerSubgraphData() @@ -880,28 +907,8 @@

      Source code for torch_tensorrt.dynamo._compiler

      < name, ) - subgraph_data.subgraph_input_shapes = parse_complex_tensor_structs( - submodule_inputs, - "shape", - lambda x: dict(x) if isinstance(x, dict) else tuple(x), - ) - subgraph_data.subgraph_input_dtypes = parse_complex_tensor_structs( - submodule_inputs, "dtype", lambda t: t.to(torch.dtype) - ) - - submodule_outputs = submodule( - *get_torch_inputs(submodule_inputs, to_torch_device(settings.device)) - ) - - subgraph_data.subgraph_output_shapes = parse_complex_tensor_structs( - submodule_outputs, - "shape", - lambda x: dict(x) if isinstance(x, dict) else tuple(x), - ) - subgraph_data.subgraph_output_dtypes = parse_complex_tensor_structs( - submodule_outputs, "dtype" - ) - + # Parse the subgraph I/O and store it + parse_graph_io(submodule, subgraph_data) dryrun_tracker.tensorrt_graph_count += 1 dryrun_tracker.per_subgraph_data.append(subgraph_data) @@ -912,27 +919,13 @@

      Source code for torch_tensorrt.dynamo._compiler

      < submodule_inputs, settings=settings, name=name, + engine_cache=engine_cache, ) trt_modules[name] = trt_module - torch_sample_arg_inputs = get_torch_inputs( - sample_arg_inputs, to_torch_device(settings.device) - ) - torch_sample_kwarg_inputs = get_torch_inputs( - sample_kwarg_inputs, to_torch_device(settings.device) - ) - sample_outputs = gm(*torch_sample_arg_inputs, **torch_sample_kwarg_inputs) - - if not isinstance(sample_outputs, (list, tuple)): - sample_outputs = [sample_outputs] - - dryrun_tracker.graph_output_shapes = parse_complex_tensor_structs( - sample_outputs, "shape", lambda x: dict(x) if isinstance(x, dict) else tuple(x) - ) - dryrun_tracker.graph_output_dtypes = parse_complex_tensor_structs( - sample_outputs, "dtype" - ) + # Parse the graph I/O and store it in dryrun tracker + parse_graph_io(gm, dryrun_tracker) # Replace all FX Modules with TRT Modules for name, trt_module in trt_modules.items(): @@ -1066,10 +1059,10 @@

      Source code for torch_tensorrt.dynamo._compiler

      < DeprecationWarning, stacklevel=2, ) - if not arg_inputs and not inputs: + if arg_inputs is None and inputs is None: raise AssertionError("'arg_inputs' and 'inputs' should not both be None.") - elif arg_inputs and inputs: + elif arg_inputs is not None and inputs is not None: raise AssertionError( "'arg_inputs' and 'inputs' should not be used at the same time." ) diff --git a/docs/_modules/torch_tensorrt/dynamo/_exporter.html b/docs/_modules/torch_tensorrt/dynamo/_exporter.html index 7beef6382d..12eb2ecd1d 100644 --- a/docs/_modules/torch_tensorrt/dynamo/_exporter.html +++ b/docs/_modules/torch_tensorrt/dynamo/_exporter.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo._exporter — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.dynamo._exporter — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/dynamo/_refit.html b/docs/_modules/torch_tensorrt/dynamo/_refit.html index 154ae0777e..3c8da6aa0c 100644 --- a/docs/_modules/torch_tensorrt/dynamo/_refit.html +++ b/docs/_modules/torch_tensorrt/dynamo/_refit.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo._refit — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.dynamo._refit — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      @@ -459,7 +461,6 @@

      Source code for torch_tensorrt.dynamo._refit

      from typing import Any, List, Optional, Sequence, Tuple
       
       import numpy as np
      -import tensorrt as trt
       import torch
       from torch.export import ExportedProgram
       from torch_tensorrt._enums import dtype
      @@ -488,6 +489,7 @@ 

      Source code for torch_tensorrt.dynamo._refit

      )
       from torch_tensorrt.dynamo.utils import (
           check_module_output,
      +    get_model_device,
           get_torch_inputs,
           set_log_level,
           to_torch_device,
      @@ -495,6 +497,8 @@ 

      Source code for torch_tensorrt.dynamo._refit

      )
       from torch_tensorrt.logging import TRT_LOGGER
       
      +import tensorrt as trt
      +
       logger = logging.getLogger(__name__)
       
       
      @@ -599,7 +603,7 @@ 

      Source code for torch_tensorrt.dynamo._refit

          """
       
           refitted = set()
      -    torch_device = list(new_gm.state_dict().values())[0].device.type
      +    torch_device = get_model_device(new_gm)
           refitter = trt.Refitter(old_engine, TRT_LOGGER)
           weight_list = refitter.get_all_weights()
       
      diff --git a/docs/_modules/torch_tensorrt/dynamo/_settings.html b/docs/_modules/torch_tensorrt/dynamo/_settings.html
      index 62a8669834..73e1e6c6bd 100644
      --- a/docs/_modules/torch_tensorrt/dynamo/_settings.html
      +++ b/docs/_modules/torch_tensorrt/dynamo/_settings.html
      @@ -9,7 +9,7 @@
         
         
         
      -  torch_tensorrt.dynamo._settings — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation
      +  torch_tensorrt.dynamo._settings — Torch-TensorRT v2.5.0.dev0+8759736 documentation
         
       
         
      @@ -272,7 +272,7 @@
                     
                     
                       
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      @@ -459,6 +461,7 @@

      Source code for torch_tensorrt.dynamo._settings

      < from torch_tensorrt._enums import EngineCapability, dtype from torch_tensorrt.dynamo._defaults import ( ASSUME_DYNAMIC_SHAPE_SUPPORT, + CACHE_BUILT_ENGINES, DEBUG, DISABLE_TF32, DLA_GLOBAL_DRAM_SIZE, @@ -477,6 +480,7 @@

      Source code for torch_tensorrt.dynamo._settings

      < OPTIMIZATION_LEVEL, PASS_THROUGH_BUILD_FAILURES, REQUIRE_FULL_COMPILATION, + REUSE_CACHED_ENGINES, SPARSE_WEIGHTS, TIMING_CACHE_PATH, TRUNCATE_DOUBLE, @@ -527,6 +531,8 @@

      Source code for torch_tensorrt.dynamo._settings

      < output to a file if a string path is specified hardware_compatible (bool): Build the TensorRT engines compatible with GPU architectures other than that of the GPU on which the engine was built (currently works for NVIDIA Ampere and newer) timing_cache_path (str): Path to the timing cache if it exists (or) where it will be saved after compilation + cache_built_engines (bool): Whether to save the compiled TRT engines to storage + reuse_cached_engines (bool): Whether to load the compiled TRT engines from storage """ enabled_precisions: Set[dtype] = field(default_factory=lambda: ENABLED_PRECISIONS) @@ -558,7 +564,9 @@

      Source code for torch_tensorrt.dynamo._settings

      < dryrun: Union[bool, str] = DRYRUN hardware_compatible: bool = HARDWARE_COMPATIBLE timing_cache_path: str = TIMING_CACHE_PATH - lazy_engine_init: bool = LAZY_ENGINE_INIT
      + lazy_engine_init: bool = LAZY_ENGINE_INIT + cache_built_engines: bool = CACHE_BUILT_ENGINES + reuse_cached_engines: bool = REUSE_CACHED_ENGINES
      diff --git a/docs/_modules/torch_tensorrt/dynamo/_tracer.html b/docs/_modules/torch_tensorrt/dynamo/_tracer.html index aeda025161..86583fa3a4 100644 --- a/docs/_modules/torch_tensorrt/dynamo/_tracer.html +++ b/docs/_modules/torch_tensorrt/dynamo/_tracer.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo._tracer — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.dynamo._tracer — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/dynamo/runtime/_MutableTorchTensorRTModule.html b/docs/_modules/torch_tensorrt/dynamo/runtime/_MutableTorchTensorRTModule.html index 77ebd98cd9..aedbdef751 100644 --- a/docs/_modules/torch_tensorrt/dynamo/runtime/_MutableTorchTensorRTModule.html +++ b/docs/_modules/torch_tensorrt/dynamo/runtime/_MutableTorchTensorRTModule.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo.runtime._MutableTorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.dynamo.runtime._MutableTorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.html b/docs/_modules/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.html index 6b2afa2855..8ff32a67ab 100644 --- a/docs/_modules/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.html +++ b/docs/_modules/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo.runtime._PythonTorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.dynamo.runtime._PythonTorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.html b/docs/_modules/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.html index 4023689859..e74f349040 100644 --- a/docs/_modules/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.html +++ b/docs/_modules/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo.runtime._TorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.dynamo.runtime._TorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/fx/input_tensor_spec.html b/docs/_modules/torch_tensorrt/fx/input_tensor_spec.html index b7e4537697..35cdcecf84 100644 --- a/docs/_modules/torch_tensorrt/fx/input_tensor_spec.html +++ b/docs/_modules/torch_tensorrt/fx/input_tensor_spec.html @@ -9,7 +9,7 @@ - torch_tensorrt.fx.input_tensor_spec — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.fx.input_tensor_spec — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/fx/trt_module.html b/docs/_modules/torch_tensorrt/fx/trt_module.html index 2e084c0d4a..4e478c0e3f 100644 --- a/docs/_modules/torch_tensorrt/fx/trt_module.html +++ b/docs/_modules/torch_tensorrt/fx/trt_module.html @@ -9,7 +9,7 @@ - torch_tensorrt.fx.trt_module — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.fx.trt_module — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/runtime/_multi_device_safe_mode.html b/docs/_modules/torch_tensorrt/runtime/_multi_device_safe_mode.html index 8752249250..d5cd752294 100644 --- a/docs/_modules/torch_tensorrt/runtime/_multi_device_safe_mode.html +++ b/docs/_modules/torch_tensorrt/runtime/_multi_device_safe_mode.html @@ -9,7 +9,7 @@ - torch_tensorrt.runtime._multi_device_safe_mode — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.runtime._multi_device_safe_mode — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/ts/_compile_spec.html b/docs/_modules/torch_tensorrt/ts/_compile_spec.html index ec664db21c..3a4a6994ad 100644 --- a/docs/_modules/torch_tensorrt/ts/_compile_spec.html +++ b/docs/_modules/torch_tensorrt/ts/_compile_spec.html @@ -9,7 +9,7 @@ - torch_tensorrt.ts._compile_spec — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.ts._compile_spec — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/_modules/torch_tensorrt/ts/_compiler.html b/docs/_modules/torch_tensorrt/ts/_compiler.html index e3c9ceaeea..8b84d0f15c 100644 --- a/docs/_modules/torch_tensorrt/ts/_compiler.html +++ b/docs/_modules/torch_tensorrt/ts/_compiler.html @@ -9,7 +9,7 @@ - torch_tensorrt.ts._compiler — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torch_tensorrt.ts._compiler — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -272,7 +272,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -313,6 +313,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/_sources/index.rst.txt b/docs/_sources/index.rst.txt index da5ee3d690..d1a91beabc 100644 --- a/docs/_sources/index.rst.txt +++ b/docs/_sources/index.rst.txt @@ -51,6 +51,8 @@ User Guide user_guide/using_dla tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage tutorials/_rendered_examples/dynamo/vgg16_fp8_ptq + tutorials/_rendered_examples/dynamo/engine_caching_example + tutorials/_rendered_examples/dynamo/refit_engine_example Dynamo Frontend ---------------- diff --git a/docs/_sources/tutorials/_rendered_examples/dynamo/engine_caching_bert_example.rst.txt b/docs/_sources/tutorials/_rendered_examples/dynamo/engine_caching_bert_example.rst.txt new file mode 100644 index 0000000000..e72f42cfb2 --- /dev/null +++ b/docs/_sources/tutorials/_rendered_examples/dynamo/engine_caching_bert_example.rst.txt @@ -0,0 +1,127 @@ + +.. DO NOT EDIT. +.. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY. +.. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE: +.. "tutorials/_rendered_examples/dynamo/engine_caching_bert_example.py" +.. LINE NUMBERS ARE GIVEN BELOW. + +.. only:: html + + .. note:: + :class: sphx-glr-download-link-note + + :ref:`Go to the end ` + to download the full example code + +.. rst-class:: sphx-glr-example-title + +.. _sphx_glr_tutorials__rendered_examples_dynamo_engine_caching_bert_example.py: + + +.. _engine_caching_bert_example: + +Engine Caching (BERT) +======================= + +Small caching example on BERT. + +.. GENERATED FROM PYTHON SOURCE LINES 10-76 + +.. code-block:: python + + + import numpy as np + import torch + import torch_tensorrt + from engine_caching_example import remove_timing_cache + from transformers import BertModel + + np.random.seed(0) + torch.manual_seed(0) + + model = BertModel.from_pretrained("bert-base-uncased", return_dict=False).cuda().eval() + inputs = [ + torch.randint(0, 2, (1, 14), dtype=torch.int32).to("cuda"), + torch.randint(0, 2, (1, 14), dtype=torch.int32).to("cuda"), + ] + + + def compile_bert(iterations=3): + times = [] + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + + # The 1st iteration is to measure the compilation time without engine caching + # The 2nd and 3rd iterations are to measure the compilation time with engine caching. + # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration. + # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine. + for i in range(iterations): + # remove timing cache and reset dynamo for engine caching messurement + remove_timing_cache() + torch._dynamo.reset() + + if i == 0: + cache_built_engines = False + reuse_cached_engines = False + else: + cache_built_engines = True + reuse_cached_engines = True + + start.record() + compilation_kwargs = { + "use_python_runtime": False, + "enabled_precisions": {torch.float}, + "truncate_double": True, + "debug": False, + "min_block_size": 1, + "make_refitable": True, + "cache_built_engines": cache_built_engines, + "reuse_cached_engines": reuse_cached_engines, + "engine_cache_dir": "/tmp/torch_trt_bert_engine_cache", + "engine_cache_size": 1 << 30, # 1GB + } + optimized_model = torch.compile( + model, + backend="torch_tensorrt", + options=compilation_kwargs, + ) + optimized_model(*inputs) + end.record() + torch.cuda.synchronize() + times.append(start.elapsed_time(end)) + + print("-----compile bert-----> compilation time:\n", times, "milliseconds") + + + if __name__ == "__main__": + compile_bert() + + +.. rst-class:: sphx-glr-timing + + **Total running time of the script:** ( 0 minutes 0.000 seconds) + + +.. _sphx_glr_download_tutorials__rendered_examples_dynamo_engine_caching_bert_example.py: + +.. only:: html + + .. container:: sphx-glr-footer sphx-glr-footer-example + + + + + .. container:: sphx-glr-download sphx-glr-download-python + + :download:`Download Python source code: engine_caching_bert_example.py ` + + .. container:: sphx-glr-download sphx-glr-download-jupyter + + :download:`Download Jupyter notebook: engine_caching_bert_example.ipynb ` + + +.. only:: html + + .. rst-class:: sphx-glr-signature + + `Gallery generated by Sphinx-Gallery `_ diff --git a/docs/_sources/tutorials/_rendered_examples/dynamo/engine_caching_example.rst.txt b/docs/_sources/tutorials/_rendered_examples/dynamo/engine_caching_example.rst.txt new file mode 100644 index 0000000000..df61bec65e --- /dev/null +++ b/docs/_sources/tutorials/_rendered_examples/dynamo/engine_caching_example.rst.txt @@ -0,0 +1,361 @@ + +.. DO NOT EDIT. +.. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY. +.. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE: +.. "tutorials/_rendered_examples/dynamo/engine_caching_example.py" +.. LINE NUMBERS ARE GIVEN BELOW. + +.. only:: html + + .. note:: + :class: sphx-glr-download-link-note + + :ref:`Go to the end ` + to download the full example code + +.. rst-class:: sphx-glr-example-title + +.. _sphx_glr_tutorials__rendered_examples_dynamo_engine_caching_example.py: + + +.. _engine_caching_example: + +Engine Caching +======================= + +As model sizes increase, the cost of compilation will as well. With AOT methods +like ``torch.dynamo.compile``, this cost is paid upfront. However if the weights +change, the session ends or you are using JIT methods like ``torch.compile``, as +graphs get invalidated they get re-compiled, this cost will get paid repeatedly. +Engine caching is a way to mitigate this cost by saving constructed engines to disk +and re-using them when possible. This tutorial demonstrates how to use engine caching +with TensorRT in PyTorch. Engine caching can significantly speed up subsequent model +compilations reusing previously built TensorRT engines. + +We'll explore two approaches: + + 1. Using torch_tensorrt.dynamo.compile + 2. Using torch.compile with the TensorRT backend + +The example uses a pre-trained ResNet18 model and shows the +differences between compilation without caching, with caching enabled, +and when reusing cached engines. + +.. GENERATED FROM PYTHON SOURCE LINES 26-52 + +.. code-block:: python + + + import os + from typing import Dict, Optional + + import numpy as np + import torch + import torch_tensorrt as torch_trt + import torchvision.models as models + from torch_tensorrt.dynamo._defaults import TIMING_CACHE_PATH + from torch_tensorrt.dynamo._engine_cache import BaseEngineCache + + np.random.seed(0) + torch.manual_seed(0) + + model = models.resnet18(pretrained=True).eval().to("cuda") + enabled_precisions = {torch.float} + debug = False + min_block_size = 1 + use_python_runtime = False + + + def remove_timing_cache(path=TIMING_CACHE_PATH): + if os.path.exists(path): + os.remove(path) + + + +.. GENERATED FROM PYTHON SOURCE LINES 53-67 + +Engine Caching for JIT Compilation +---------------------------------- + +The primary goal of engine caching is to help speed up JIT workflows. ``torch.compile`` +provides a great deal of flexibility in model construction which makes it a good +first tool to try when looking to speed up your workflow. However, historically +the cost of compilation and in particular recompilation has been a barrier to entry +for many users. If for some reason a subgraph gets invalidated, that graph is reconstructed +scratch prior to the addition of engine caching. Now as engines are constructed, with ``cache_built_engines=True``, +engines are saved to disk tied to a hash of their corresponding PyTorch subgraph. If +in a subsequent compilation, either as part of this session or a new session, the cache will +pull the built engine and **refit** the weights which can reduce compilation times by orders of magnitude. +As such, in order to insert a new engine into the cache (i.e. ``cache_built_engines=True``), +the engine must be refitable (``make_refittable=True``). See :ref:`refit_engine_example` for more details. + +.. GENERATED FROM PYTHON SOURCE LINES 67-118 + +.. code-block:: python + + + + def torch_compile(iterations=3): + times = [] + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + + # The 1st iteration is to measure the compilation time without engine caching + # The 2nd and 3rd iterations are to measure the compilation time with engine caching. + # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration. + # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine. + for i in range(iterations): + inputs = [torch.rand((100, 3, 224, 224)).to("cuda")] + # remove timing cache and reset dynamo just for engine caching messurement + remove_timing_cache() + torch._dynamo.reset() + + if i == 0: + cache_built_engines = False + reuse_cached_engines = False + else: + cache_built_engines = True + reuse_cached_engines = True + + start.record() + compiled_model = torch.compile( + model, + backend="tensorrt", + options={ + "use_python_runtime": True, + "enabled_precisions": enabled_precisions, + "debug": debug, + "min_block_size": min_block_size, + "make_refitable": True, + "cache_built_engines": cache_built_engines, + "reuse_cached_engines": reuse_cached_engines, + }, + ) + compiled_model(*inputs) # trigger the compilation + end.record() + torch.cuda.synchronize() + times.append(start.elapsed_time(end)) + + print("----------------torch_compile----------------") + print("disable engine caching, used:", times[0], "ms") + print("enable engine caching to cache engines, used:", times[1], "ms") + print("enable engine caching to reuse engines, used:", times[2], "ms") + + + torch_compile() + + +.. GENERATED FROM PYTHON SOURCE LINES 119-124 + +Engine Caching for AOT Compilation +---------------------------------- +Similarly to the JIT workflow, AOT workflows can benefit from engine caching. +As the same architecture or common subgraphs get recompiled, the cache will pull +previously built engines and refit the weights. + +.. GENERATED FROM PYTHON SOURCE LINES 124-178 + +.. code-block:: python + + + + def dynamo_compile(iterations=3): + times = [] + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + + example_inputs = (torch.randn((100, 3, 224, 224)).to("cuda"),) + # Mark the dim0 of inputs as dynamic + batch = torch.export.Dim("batch", min=1, max=200) + exp_program = torch.export.export( + model, args=example_inputs, dynamic_shapes={"x": {0: batch}} + ) + + # The 1st iteration is to measure the compilation time without engine caching + # The 2nd and 3rd iterations are to measure the compilation time with engine caching. + # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration. + # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine. + for i in range(iterations): + inputs = [torch.rand((100 + i, 3, 224, 224)).to("cuda")] + remove_timing_cache() # remove timing cache just for engine caching messurement + if i == 0: + cache_built_engines = False + reuse_cached_engines = False + else: + cache_built_engines = True + reuse_cached_engines = True + + start.record() + trt_gm = torch_trt.dynamo.compile( + exp_program, + tuple(inputs), + use_python_runtime=use_python_runtime, + enabled_precisions=enabled_precisions, + debug=debug, + min_block_size=min_block_size, + make_refitable=True, + cache_built_engines=cache_built_engines, + reuse_cached_engines=reuse_cached_engines, + engine_cache_size=1 << 30, # 1GB + ) + # output = trt_gm(*inputs) + end.record() + torch.cuda.synchronize() + times.append(start.elapsed_time(end)) + + print("----------------dynamo_compile----------------") + print("disable engine caching, used:", times[0], "ms") + print("enable engine caching to cache engines, used:", times[1], "ms") + print("enable engine caching to reuse engines, used:", times[2], "ms") + + + dynamo_compile() + + +.. GENERATED FROM PYTHON SOURCE LINES 179-195 + +Custom Engine Cache +---------------------- + +By default, the engine cache is stored in the system's temporary directory. Both the cache directory and +size limit can be customized by passing ``engine_cache_dir`` and ``engine_cache_size``. +Users can also define their own engine cache implementation by extending the ``BaseEngineCache`` class. +This allows for remote or shared caching if so desired. + +The custom engine cache should implement the following methods: + - ``save``: Save the engine blob to the cache. + - ``load``: Load the engine blob from the cache. + +The hash provided by the cache systen is a weight agnostic hash of the originating PyTorch subgraph (post lowering). +The blob contains a serialized engine, calling spec data, and weight map information in the pickle format + +Below is an example of a custom engine cache implementation that implents a ``RAMEngineCache``. + +.. GENERATED FROM PYTHON SOURCE LINES 195-289 + +.. code-block:: python + + + + class RAMEngineCache(BaseEngineCache): + def __init__( + self, + ) -> None: + """ + Constructs a user held engine cache in memory. + """ + self.engine_cache: Dict[str, bytes] = {} + + def save( + self, + hash: str, + blob: bytes, + ): + """ + Insert the engine blob to the cache. + + Args: + hash (str): The hash key to associate with the engine blob. + blob (bytes): The engine blob to be saved. + + Returns: + None + """ + self.engine_cache[hash] = blob + + def load(self, hash: str) -> Optional[bytes]: + """ + Load the engine blob from the cache. + + Args: + hash (str): The hash key of the engine to load. + + Returns: + Optional[bytes]: The engine blob if found, None otherwise. + """ + if hash in self.engine_cache: + return self.engine_cache[hash] + else: + return None + + + def torch_compile_my_cache(iterations=3): + times = [] + engine_cache = RAMEngineCache() + start = torch.cuda.Event(enable_timing=True) + end = torch.cuda.Event(enable_timing=True) + + # The 1st iteration is to measure the compilation time without engine caching + # The 2nd and 3rd iterations are to measure the compilation time with engine caching. + # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration. + # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine. + for i in range(iterations): + inputs = [torch.rand((100, 3, 224, 224)).to("cuda")] + # remove timing cache and reset dynamo just for engine caching messurement + remove_timing_cache() + torch._dynamo.reset() + + if i == 0: + cache_built_engines = False + reuse_cached_engines = False + else: + cache_built_engines = True + reuse_cached_engines = True + + start.record() + compiled_model = torch.compile( + model, + backend="tensorrt", + options={ + "use_python_runtime": True, + "enabled_precisions": enabled_precisions, + "debug": debug, + "min_block_size": min_block_size, + "make_refitable": True, + "cache_built_engines": cache_built_engines, + "reuse_cached_engines": reuse_cached_engines, + "custom_engine_cache": engine_cache, + }, + ) + compiled_model(*inputs) # trigger the compilation + end.record() + torch.cuda.synchronize() + times.append(start.elapsed_time(end)) + + print("----------------torch_compile----------------") + print("disable engine caching, used:", times[0], "ms") + print("enable engine caching to cache engines, used:", times[1], "ms") + print("enable engine caching to reuse engines, used:", times[2], "ms") + + + torch_compile_my_cache() + + +.. rst-class:: sphx-glr-timing + + **Total running time of the script:** ( 0 minutes 0.000 seconds) + + +.. _sphx_glr_download_tutorials__rendered_examples_dynamo_engine_caching_example.py: + +.. only:: html + + .. container:: sphx-glr-footer sphx-glr-footer-example + + + + + .. container:: sphx-glr-download sphx-glr-download-python + + :download:`Download Python source code: engine_caching_example.py ` + + .. container:: sphx-glr-download sphx-glr-download-jupyter + + :download:`Download Jupyter notebook: engine_caching_example.ipynb ` + + +.. only:: html + + .. rst-class:: sphx-glr-signature + + `Gallery generated by Sphinx-Gallery `_ diff --git a/docs/_sources/tutorials/_rendered_examples/dynamo/index.rst.txt b/docs/_sources/tutorials/_rendered_examples/dynamo/index.rst.txt index 6e5917ae7b..64ecdc59aa 100644 --- a/docs/_sources/tutorials/_rendered_examples/dynamo/index.rst.txt +++ b/docs/_sources/tutorials/_rendered_examples/dynamo/index.rst.txt @@ -19,6 +19,8 @@ a number of ways you can leverage this backend to accelerate inference. * :ref:`refit_engine_example`: Refitting a compiled TensorRT Graph Module with updated weights * :ref:`mutable_torchtrt_module_example`: Compile, use, and modify TensorRT Graph Module with MutableTorchTensorRTModule * :ref:`vgg16_fp8_ptq`: Compiling a VGG16 model with FP8 and PTQ using ``torch.compile`` +* :ref:`engine_caching_example`: Utilizing engine caching to speed up compilation times +* :ref:`engine_caching_bert_example`: Demonstrating engine caching on BERT @@ -61,6 +63,23 @@ a number of ways you can leverage this backend to accelerate inference. +.. raw:: html + +
      + +.. only:: html + + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_refit_engine_example_thumb.png + :alt: + + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_refit_engine_example.py` + +.. raw:: html + +
      Refitting Torch-TensorRT Programs with New Weights
      +
      + + .. raw:: html
      @@ -80,18 +99,18 @@ a number of ways you can leverage this backend to accelerate inference. .. raw:: html -
      +
      .. only:: html - .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_refit_engine_example_thumb.png + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_torch_export_gpt2_thumb.png :alt: - :ref:`sphx_glr_tutorials__rendered_examples_dynamo_refit_engine_example.py` + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_torch_export_gpt2.py` .. raw:: html -
      Refit TenorRT Graph Module with Torch-TensorRT
      +
      Compiling GPT2 using the Torch-TensorRT with dynamo backend
      @@ -112,6 +131,40 @@ a number of ways you can leverage this backend to accelerate inference.
      +.. raw:: html + +
      + +.. only:: html + + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_torch_export_llama2_thumb.png + :alt: + + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_torch_export_llama2.py` + +.. raw:: html + +
      Compiling Llama2 using the Torch-TensorRT with dynamo backend
      +
      + + +.. raw:: html + +
      + +.. only:: html + + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_engine_caching_bert_example_thumb.png + :alt: + + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_engine_caching_bert_example.py` + +.. raw:: html + +
      Engine Caching (BERT)
      +
      + + .. raw:: html
      @@ -163,6 +216,23 @@ a number of ways you can leverage this backend to accelerate inference.
      +.. raw:: html + +
      + +.. only:: html + + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_engine_caching_example_thumb.png + :alt: + + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_engine_caching_example.py` + +.. raw:: html + +
      Engine Caching
      +
      + + .. raw:: html
      @@ -190,11 +260,15 @@ a number of ways you can leverage this backend to accelerate inference. /tutorials/_rendered_examples/dynamo/torch_compile_stable_diffusion /tutorials/_rendered_examples/dynamo/torch_export_cudagraphs - /tutorials/_rendered_examples/dynamo/torch_compile_transformers_example /tutorials/_rendered_examples/dynamo/refit_engine_example + /tutorials/_rendered_examples/dynamo/torch_compile_transformers_example + /tutorials/_rendered_examples/dynamo/torch_export_gpt2 /tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage + /tutorials/_rendered_examples/dynamo/torch_export_llama2 + /tutorials/_rendered_examples/dynamo/engine_caching_bert_example /tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example /tutorials/_rendered_examples/dynamo/torch_compile_resnet_example /tutorials/_rendered_examples/dynamo/vgg16_ptq + /tutorials/_rendered_examples/dynamo/engine_caching_example /tutorials/_rendered_examples/dynamo/custom_kernel_plugins diff --git a/docs/_sources/tutorials/_rendered_examples/dynamo/refit_engine_example.rst.txt b/docs/_sources/tutorials/_rendered_examples/dynamo/refit_engine_example.rst.txt index cc0b9fd21e..fb48bc8536 100644 --- a/docs/_sources/tutorials/_rendered_examples/dynamo/refit_engine_example.rst.txt +++ b/docs/_sources/tutorials/_rendered_examples/dynamo/refit_engine_example.rst.txt @@ -20,31 +20,38 @@ .. _refit_engine_example: -Refit TenorRT Graph Module with Torch-TensorRT +Refitting Torch-TensorRT Programs with New Weights =================================================================== -We are going to demonstrate how a compiled TensorRT Graph Module can be refitted with updated weights. - -In many cases, we frequently update the weights of models, such as applying various LoRA to Stable Diffusion or constant A/B testing of AI products. -That poses challenges for TensorRT inference optimizations, as compiling the TensorRT engines takes significant time, making repetitive compilation highly inefficient. -Torch-TensorRT supports refitting TensorRT graph modules without re-compiling the engine, considerably accelerating the workflow. +Compilation is an expensive operation as it involves many graph transformations, translations +and optimizations applied on the model. In cases were the weights of a model might be updated +occasionally (e.g. inserting LoRA adapters), the large cost of recompilation can make it infeasible +to use TensorRT if the compiled program needed to be built from scratch each time. Torch-TensorRT +provides a PyTorch native mechanism to update the weights of a compiled TensorRT program without +recompiling from scratch through weight refitting. In this tutorial, we are going to walk through -1. Compiling a PyTorch model to a TensorRT Graph Module -2. Save and load a graph module -3. Refit the graph module -.. GENERATED FROM PYTHON SOURCE LINES 20-22 + 1. Compiling a PyTorch model to a TensorRT Graph Module + 2. Save and load a graph module + 3. Refit the graph module + +This tutorial focuses mostly on the AOT workflow where it is most likely that a user might need to +manually refit a module. In the JIT workflow, weight changes trigger recompilation. As the engine +has previously been built, with an engine cache enabled, Torch-TensorRT can automatically recognize +a previously built engine, trigger refit and short cut recompilation on behalf of the user (see: :ref:`engine_caching_example`). + +.. GENERATED FROM PYTHON SOURCE LINES 27-29 Standard Workflow ----------------------------- -.. GENERATED FROM PYTHON SOURCE LINES 24-26 +.. GENERATED FROM PYTHON SOURCE LINES 31-33 Imports and model definition ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -.. GENERATED FROM PYTHON SOURCE LINES 26-38 +.. GENERATED FROM PYTHON SOURCE LINES 33-45 .. code-block:: python @@ -61,17 +68,24 @@ Imports and model definition -.. GENERATED FROM PYTHON SOURCE LINES 39-41 +.. GENERATED FROM PYTHON SOURCE LINES 46-55 -Compile the module for the first time and save it. -^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +Make a Refitable Compilation Program +--------------------------------------- -.. GENERATED FROM PYTHON SOURCE LINES 41-66 +The inital step is to compile a module and save it as with a normal. Note that there is an +additional parameter `make_refitable` that is set to `True`. This parameter is used to +indicate that the engine being built should support weight refitting later. Engines built without +these setttings will not be able to be refit. + +In this case we are going to compile a ResNet18 model with randomly initialized weights and save it. + +.. GENERATED FROM PYTHON SOURCE LINES 55-79 .. code-block:: python - model = models.resnet18(pretrained=True).eval().to("cuda") + model = models.resnet18(pretrained=False).eval().to("cuda") exp_program = torch.export.export(model, tuple(inputs)) enabled_precisions = {torch.float} debug = False @@ -91,23 +105,27 @@ Compile the module for the first time and save it. ) # Output is a torch.fx.GraphModule # Save the graph module as an exported program - # This is only supported when use_python_runtime = False torch_trt.save(trt_gm, "./compiled.ep", inputs=inputs) -.. GENERATED FROM PYTHON SOURCE LINES 67-69 +.. GENERATED FROM PYTHON SOURCE LINES 80-87 -Refit the module with update model weights -^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +Refit the Program with Pretrained Weights +------------------------------------------ -.. GENERATED FROM PYTHON SOURCE LINES 69-93 +Random weights are not useful for inference. But now instead of recompiling the model, we can +refit the model with the pretrained weights. This is done by setting up another PyTorch module +with the target weights and exporting it as an ExportedProgram. Then the ``refit_module_weights`` +function is used to update the weights of the compiled module with the new weights. + +.. GENERATED FROM PYTHON SOURCE LINES 87-111 .. code-block:: python # Create and compile the updated model - model2 = models.resnet18(pretrained=False).eval().to("cuda") + model2 = models.resnet18(pretrained=True).eval().to("cuda") exp_program2 = torch.export.export(model2, tuple(inputs)) @@ -130,18 +148,36 @@ Refit the module with update model weights print("Refit successfully!") -.. GENERATED FROM PYTHON SOURCE LINES 94-96 +.. GENERATED FROM PYTHON SOURCE LINES 112-140 -Alternative Workflow using Python Runtime +Advanced Usage ----------------------------- -.. GENERATED FROM PYTHON SOURCE LINES 96-99 +There are a number of settings you can use to control the refit process -.. code-block:: python +Weight Map Cache +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + +Weight refitting works by matching the weights of the compiled module with the new weights from +the user supplied ExportedProgram. Since 1:1 name matching from PyTorch to TensorRT is hard to accomplish, +the only gaurenteed way to match weights at *refit-time* is to pass the new ExportedProgram through the +early phases of the compilation process to generate near identical weight names. This can be expensive +and is not always necessary. +To avoid this, **At initial compile**, Torch-TensorRt will attempt to cache a direct mapping from PyTorch +weights to TensorRT weights. This cache is stored in the compiled module as metadata and can be used +to speed up refit. If the cache is not present, the refit system will fallback to rebuilding the mapping at +refit-time. Use of this cache is controlled by the ``use_weight_map_cache`` parameter. + +Since the cache uses a heuristic based system for matching PyTorch and TensorRT weights, you may want to verify the refitting. This can be done by setting +``verify_output`` to True and providing sample ``arg_inputs`` and ``kwarg_inputs``. When this is done, the refit +system will run the refitted module and the user supplied module on the same inputs and compare the outputs. + +In-Place Refit +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ - # Currently python runtime does not support engine serialization. So the refitting will be done in the same runtime. - # This usecase is more useful when you need to switch different weights in the same runtime, such as using Stable Diffusion. +``in_place`` allows the user to refit the module in place. This is useful when the user wants to update the weights +of the compiled module without creating a new module. .. rst-class:: sphx-glr-timing diff --git a/docs/_sources/tutorials/_rendered_examples/dynamo/torch_export_gpt2.rst.txt b/docs/_sources/tutorials/_rendered_examples/dynamo/torch_export_gpt2.rst.txt new file mode 100644 index 0000000000..be90efc337 --- /dev/null +++ b/docs/_sources/tutorials/_rendered_examples/dynamo/torch_export_gpt2.rst.txt @@ -0,0 +1,168 @@ + +.. DO NOT EDIT. +.. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY. +.. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE: +.. "tutorials/_rendered_examples/dynamo/torch_export_gpt2.py" +.. LINE NUMBERS ARE GIVEN BELOW. + +.. only:: html + + .. note:: + :class: sphx-glr-download-link-note + + :ref:`Go to the end ` + to download the full example code + +.. rst-class:: sphx-glr-example-title + +.. _sphx_glr_tutorials__rendered_examples_dynamo_torch_export_gpt2.py: + + +.. _torch_export_gpt2: + +Compiling GPT2 using the Torch-TensorRT with dynamo backend +========================================================== + +This interactive script is intended as a sample of the Torch-TensorRT workflow with dynamo backend on a GPT2 model. + +.. GENERATED FROM PYTHON SOURCE LINES 10-12 + +Imports and Model Definition +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + +.. GENERATED FROM PYTHON SOURCE LINES 12-17 + +.. code-block:: python + + import torch + import torch_tensorrt + from transformers import AutoModelForCausalLM, AutoTokenizer + from utils import export_llm, generate + + +.. GENERATED FROM PYTHON SOURCE LINES 18-35 + +.. code-block:: python + + + # Define the parameters and initialize the model + MAX_TOKENS = 32 + DEVICE = torch.device("cuda:0") + + # Define the GPT2 model from hugging face + # kv_cache is not supported in Torch-TRT currently. + # CPU is used here so that GPU memory is reserved for TRT compilation. + with torch.no_grad(): + tokenizer = AutoTokenizer.from_pretrained("gpt2") + model = AutoModelForCausalLM.from_pretrained( + "gpt2", + pad_token_id=tokenizer.eos_token_id, + use_cache=False, + attn_implementation="eager", + ).eval() + + +.. GENERATED FROM PYTHON SOURCE LINES 36-37 + +Tokenize a sample input prompt and get pytorch model outputs + +.. GENERATED FROM PYTHON SOURCE LINES 37-46 + +.. code-block:: python + + prompt = "I enjoy walking with my cute dog" + model_inputs = tokenizer(prompt, return_tensors="pt") + input_ids = model_inputs["input_ids"] + + # Auto-regressive generation loop for greedy decoding using PyTorch model + # We use a custom generate function which is very similar to the huggingface one. + pyt_gen_tokens = generate(model, input_ids, MAX_TOKENS, tokenizer.eos_token_id) + + + +.. GENERATED FROM PYTHON SOURCE LINES 47-49 + +Compilation with `Torch-TensorRT` using dynamo backend and generate TensorRT outputs +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + +.. GENERATED FROM PYTHON SOURCE LINES 49-67 + +.. code-block:: python + + + # Export the GPT2 model into an ExportedProgram which is input of TRT compilation + gpt2_ep = export_llm(model, input_ids, max_seq_len=1024) + trt_model = torch_tensorrt.dynamo.compile( + gpt2_ep, + inputs=[input_ids], + enabled_precisions={torch.float32}, + truncate_double=True, + device=DEVICE, + disable_tf32=True, + ) + + # Auto-regressive generation loop for greedy decoding using TensorRT model + # We use a custom generate function which is very similar to the huggingface one. + # Move inputs to GPU + input_ids = input_ids.to(DEVICE) + trt_gen_tokens = generate(trt_model, input_ids, MAX_TOKENS, tokenizer.eos_token_id) + + +.. GENERATED FROM PYTHON SOURCE LINES 68-70 + +Decode the output sentences of PyTorch and TensorRT +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + +.. GENERATED FROM PYTHON SOURCE LINES 70-81 + +.. code-block:: python + + print("=============================") + print( + "Pytorch model generated text: ", + tokenizer.decode(pyt_gen_tokens[0], skip_special_tokens=True), + ) + print("=============================") + print( + "TensorRT model generated text: ", + tokenizer.decode(trt_gen_tokens[0], skip_special_tokens=True), + ) + + +.. GENERATED FROM PYTHON SOURCE LINES 82-87 + +The output sentences should look like +============================= +Pytorch model generated text: I enjoy walking with my cute dog, but I'm not sure if I'll ever be able to walk with my dog. I'm not sure if I'll ever be able to walk with my +============================= +TensorRT model generated text: I enjoy walking with my cute dog, but I'm not sure if I'll ever be able to walk with my dog. I'm not sure if I'll ever be able to walk with my + + +.. rst-class:: sphx-glr-timing + + **Total running time of the script:** ( 0 minutes 0.000 seconds) + + +.. _sphx_glr_download_tutorials__rendered_examples_dynamo_torch_export_gpt2.py: + +.. only:: html + + .. container:: sphx-glr-footer sphx-glr-footer-example + + + + + .. container:: sphx-glr-download sphx-glr-download-python + + :download:`Download Python source code: torch_export_gpt2.py ` + + .. container:: sphx-glr-download sphx-glr-download-jupyter + + :download:`Download Jupyter notebook: torch_export_gpt2.ipynb ` + + +.. only:: html + + .. rst-class:: sphx-glr-signature + + `Gallery generated by Sphinx-Gallery `_ diff --git a/docs/_sources/tutorials/_rendered_examples/dynamo/torch_export_llama2.rst.txt b/docs/_sources/tutorials/_rendered_examples/dynamo/torch_export_llama2.rst.txt new file mode 100644 index 0000000000..5e66a72aab --- /dev/null +++ b/docs/_sources/tutorials/_rendered_examples/dynamo/torch_export_llama2.rst.txt @@ -0,0 +1,175 @@ + +.. DO NOT EDIT. +.. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY. +.. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE: +.. "tutorials/_rendered_examples/dynamo/torch_export_llama2.py" +.. LINE NUMBERS ARE GIVEN BELOW. + +.. only:: html + + .. note:: + :class: sphx-glr-download-link-note + + :ref:`Go to the end ` + to download the full example code + +.. rst-class:: sphx-glr-example-title + +.. _sphx_glr_tutorials__rendered_examples_dynamo_torch_export_llama2.py: + + +.. _torch_export_llama2: + +Compiling Llama2 using the Torch-TensorRT with dynamo backend +========================================================== + +This interactive script is intended as a sample of the Torch-TensorRT workflow with dynamo backend on a Llama2 model. + +.. GENERATED FROM PYTHON SOURCE LINES 10-12 + +Imports and Model Definition +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + +.. GENERATED FROM PYTHON SOURCE LINES 12-17 + +.. code-block:: python + + import torch + import torch_tensorrt + from transformers import AutoModelForCausalLM, AutoTokenizer + from utils import export_llm, generate + + +.. GENERATED FROM PYTHON SOURCE LINES 18-19 + +Define the parameters and initialize the model + +.. GENERATED FROM PYTHON SOURCE LINES 19-33 + +.. code-block:: python + + MAX_TOKENS = 32 + DEVICE = torch.device("cuda:0") + + # Define the Llama2 model from hugging face + # kv_cache is not supported in Torch-TRT currently. + # CPU is used here so that GPU memory is reserved for TRT compilation. + llama_path = "meta-llama/Llama-2-7b-chat-hf" + with torch.no_grad(): + model = AutoModelForCausalLM.from_pretrained( + llama_path, use_cache=False, attn_implementation="eager" + ).eval() + + tokenizer = AutoTokenizer.from_pretrained(llama_path) + + +.. GENERATED FROM PYTHON SOURCE LINES 34-35 + +Tokenize a sample input prompt and get pytorch model outputs + +.. GENERATED FROM PYTHON SOURCE LINES 35-43 + +.. code-block:: python + + prompt = "What is dynamic programming?" + model_inputs = tokenizer(prompt, return_tensors="pt") + input_ids = model_inputs.input_ids + + # Auto-regressive generation loop for greedy decoding using PyTorch model + # We use a custom generate function which is very similar to the huggingface one. + pyt_gen_tokens = generate(model, input_ids, MAX_TOKENS, tokenizer.eos_token_id) + + +.. GENERATED FROM PYTHON SOURCE LINES 44-46 + +Compilation with `Torch-TensorRT` using dynamo backend and generate TensorRT outputs +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + +.. GENERATED FROM PYTHON SOURCE LINES 46-65 + +.. code-block:: python + + + # Export the llama2 model into an ExportedProgram which is input of TRT compilation + llama2_ep = export_llm(model, input_ids, max_seq_len=64) + trt_model = torch_tensorrt.dynamo.compile( + llama2_ep, + inputs=[input_ids], + enabled_precisions={torch.float32}, + min_block_size=1, + truncate_double=True, + device=DEVICE, + disable_tf32=True, + ) + + # Auto-regressive generation loop for greedy decoding using TensorRT model + # We use a custom generate function which is very similar to the huggingface one. + # Move inputs to GPU + input_ids = input_ids.to(DEVICE) + trt_gen_tokens = generate(trt_model, input_ids, MAX_TOKENS, tokenizer.eos_token_id) + + +.. GENERATED FROM PYTHON SOURCE LINES 66-68 + +Decode the output sentences of PyTorch and TensorRT +^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ + +.. GENERATED FROM PYTHON SOURCE LINES 68-85 + +.. code-block:: python + + print("=============================") + print( + "Pytorch model generated text: ", + tokenizer.batch_decode( + pyt_gen_tokens, skip_special_tokens=True, clean_up_tokenization_spaces=False + )[0], + ) + print("=============================") + print( + "TensorRT model generated text: ", + tokenizer.batch_decode( + trt_gen_tokens, + skip_special_tokens=True, + clean_up_tokenization_spaces=False, + )[0], + ) + + +.. GENERATED FROM PYTHON SOURCE LINES 86-91 + +The output sentences should look like +============================= +Pytorch model generated text: I enjoy walking with my cute dog, but I'm not sure if I'll ever be able to walk with my dog. I'm not sure if I'll ever be able to walk with my +============================= +TensorRT model generated text: I enjoy walking with my cute dog, but I'm not sure if I'll ever be able to walk with my dog. I'm not sure if I'll ever be able to walk with my + + +.. rst-class:: sphx-glr-timing + + **Total running time of the script:** ( 0 minutes 0.000 seconds) + + +.. _sphx_glr_download_tutorials__rendered_examples_dynamo_torch_export_llama2.py: + +.. only:: html + + .. container:: sphx-glr-footer sphx-glr-footer-example + + + + + .. container:: sphx-glr-download sphx-glr-download-python + + :download:`Download Python source code: torch_export_llama2.py ` + + .. container:: sphx-glr-download sphx-glr-download-jupyter + + :download:`Download Jupyter notebook: torch_export_llama2.ipynb ` + + +.. only:: html + + .. rst-class:: sphx-glr-signature + + `Gallery generated by Sphinx-Gallery `_ diff --git a/docs/_sources/tutorials/_rendered_examples/index.rst.txt b/docs/_sources/tutorials/_rendered_examples/index.rst.txt index f68c1fb417..c688d7370f 100644 --- a/docs/_sources/tutorials/_rendered_examples/index.rst.txt +++ b/docs/_sources/tutorials/_rendered_examples/index.rst.txt @@ -35,6 +35,8 @@ a number of ways you can leverage this backend to accelerate inference. * :ref:`refit_engine_example`: Refitting a compiled TensorRT Graph Module with updated weights * :ref:`mutable_torchtrt_module_example`: Compile, use, and modify TensorRT Graph Module with MutableTorchTensorRTModule * :ref:`vgg16_fp8_ptq`: Compiling a VGG16 model with FP8 and PTQ using ``torch.compile`` +* :ref:`engine_caching_example`: Utilizing engine caching to speed up compilation times +* :ref:`engine_caching_bert_example`: Demonstrating engine caching on BERT @@ -77,6 +79,23 @@ a number of ways you can leverage this backend to accelerate inference.
      +.. raw:: html + +
      + +.. only:: html + + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_refit_engine_example_thumb.png + :alt: + + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_refit_engine_example.py` + +.. raw:: html + +
      Refitting Torch-TensorRT Programs with New Weights
      +
      + + .. raw:: html
      @@ -96,18 +115,18 @@ a number of ways you can leverage this backend to accelerate inference. .. raw:: html -
      +
      .. only:: html - .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_refit_engine_example_thumb.png + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_torch_export_gpt2_thumb.png :alt: - :ref:`sphx_glr_tutorials__rendered_examples_dynamo_refit_engine_example.py` + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_torch_export_gpt2.py` .. raw:: html -
      Refit TenorRT Graph Module with Torch-TensorRT
      +
      Compiling GPT2 using the Torch-TensorRT with dynamo backend
      @@ -128,6 +147,40 @@ a number of ways you can leverage this backend to accelerate inference.
      +.. raw:: html + +
      + +.. only:: html + + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_torch_export_llama2_thumb.png + :alt: + + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_torch_export_llama2.py` + +.. raw:: html + +
      Compiling Llama2 using the Torch-TensorRT with dynamo backend
      +
      + + +.. raw:: html + +
      + +.. only:: html + + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_engine_caching_bert_example_thumb.png + :alt: + + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_engine_caching_bert_example.py` + +.. raw:: html + +
      Engine Caching (BERT)
      +
      + + .. raw:: html
      @@ -179,6 +232,23 @@ a number of ways you can leverage this backend to accelerate inference.
      +.. raw:: html + +
      + +.. only:: html + + .. image:: /tutorials/_rendered_examples/dynamo/images/thumb/sphx_glr_engine_caching_example_thumb.png + :alt: + + :ref:`sphx_glr_tutorials__rendered_examples_dynamo_engine_caching_example.py` + +.. raw:: html + +
      Engine Caching
      +
      + + .. raw:: html
      diff --git a/docs/_static/documentation_options.js b/docs/_static/documentation_options.js index c01a50945c..1540ac6f48 100644 --- a/docs/_static/documentation_options.js +++ b/docs/_static/documentation_options.js @@ -1,6 +1,6 @@ var DOCUMENTATION_OPTIONS = { URL_ROOT: document.getElementById("documentation_options").getAttribute('data-url_root'), - VERSION: 'v2.5.0.dev0+b3a8cdd', + VERSION: 'v2.5.0.dev0+8759736', LANGUAGE: 'en', COLLAPSE_INDEX: false, BUILDER: 'html', diff --git a/docs/cli/torchtrtc.html b/docs/cli/torchtrtc.html index f6792e2ae9..5309cb19d6 100644 --- a/docs/cli/torchtrtc.html +++ b/docs/cli/torchtrtc.html @@ -10,7 +10,7 @@ - torchtrtc — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + torchtrtc — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/contributors/useful_links.html b/docs/contributors/useful_links.html index 9fba4ea080..874f4968e7 100644 --- a/docs/contributors/useful_links.html +++ b/docs/contributors/useful_links.html @@ -10,7 +10,7 @@ - Useful Links for Torch-TensorRT Development — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Useful Links for Torch-TensorRT Development — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/contributors/writing_dynamo_aten_lowering_passes.html b/docs/contributors/writing_dynamo_aten_lowering_passes.html index a4342f98de..da2a3bb37d 100644 --- a/docs/contributors/writing_dynamo_aten_lowering_passes.html +++ b/docs/contributors/writing_dynamo_aten_lowering_passes.html @@ -10,7 +10,7 @@ - Writing Dynamo ATen Lowering Passes — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Writing Dynamo ATen Lowering Passes — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/dynamo/torch_compile.html b/docs/dynamo/torch_compile.html index a0531ea32e..9715f4e330 100644 --- a/docs/dynamo/torch_compile.html +++ b/docs/dynamo/torch_compile.html @@ -10,7 +10,7 @@ - TensorRT Backend for torch.compile — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + TensorRT Backend for torch.compile — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -40,7 +40,7 @@ - + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
      +
      +
      + + + + + +
      +
      +
      + + + + + + + + + + + +
      +
      +
      + + + + + + + + + + + + + + + + +
      + +
        + +
      • + + + Docs + + > +
      • + + +
      • Engine Caching (BERT)
      • + + +
      • + + + + + +
      • + +
      + + +
      +
      + +
      + Shortcuts +
      +
      + +
      +
      + + + + + + +
      + +
      +
      + + +
      +

      Engine Caching (BERT)

      +

      Small caching example on BERT.

      +
      import numpy as np
      +import torch
      +import torch_tensorrt
      +from engine_caching_example import remove_timing_cache
      +from transformers import BertModel
      +
      +np.random.seed(0)
      +torch.manual_seed(0)
      +
      +model = BertModel.from_pretrained("bert-base-uncased", return_dict=False).cuda().eval()
      +inputs = [
      +    torch.randint(0, 2, (1, 14), dtype=torch.int32).to("cuda"),
      +    torch.randint(0, 2, (1, 14), dtype=torch.int32).to("cuda"),
      +]
      +
      +
      +def compile_bert(iterations=3):
      +    times = []
      +    start = torch.cuda.Event(enable_timing=True)
      +    end = torch.cuda.Event(enable_timing=True)
      +
      +    # The 1st iteration is to measure the compilation time without engine caching
      +    # The 2nd and 3rd iterations are to measure the compilation time with engine caching.
      +    # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration.
      +    # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine.
      +    for i in range(iterations):
      +        # remove timing cache and reset dynamo for engine caching messurement
      +        remove_timing_cache()
      +        torch._dynamo.reset()
      +
      +        if i == 0:
      +            cache_built_engines = False
      +            reuse_cached_engines = False
      +        else:
      +            cache_built_engines = True
      +            reuse_cached_engines = True
      +
      +        start.record()
      +        compilation_kwargs = {
      +            "use_python_runtime": False,
      +            "enabled_precisions": {torch.float},
      +            "truncate_double": True,
      +            "debug": False,
      +            "min_block_size": 1,
      +            "make_refitable": True,
      +            "cache_built_engines": cache_built_engines,
      +            "reuse_cached_engines": reuse_cached_engines,
      +            "engine_cache_dir": "/tmp/torch_trt_bert_engine_cache",
      +            "engine_cache_size": 1 << 30,  # 1GB
      +        }
      +        optimized_model = torch.compile(
      +            model,
      +            backend="torch_tensorrt",
      +            options=compilation_kwargs,
      +        )
      +        optimized_model(*inputs)
      +        end.record()
      +        torch.cuda.synchronize()
      +        times.append(start.elapsed_time(end))
      +
      +    print("-----compile bert-----> compilation time:\n", times, "milliseconds")
      +
      +
      +if __name__ == "__main__":
      +    compile_bert()
      +
      +
      +

      Total running time of the script: ( 0 minutes 0.000 seconds)

      + +

      Gallery generated by Sphinx-Gallery

      +
      + + +
      + +
      +
      + + + + +
      + + + +
      +

      + © Copyright 2024, NVIDIA Corporation. + +

      +
      + +
      + Built with Sphinx using a theme provided by Read the Docs. +
      + + +
      + +
      +
      + +
      +
      + +
      +
      +
      +
      + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
      +
      +
      +
      +

      Docs

      +

      Access comprehensive developer documentation for PyTorch

      + View Docs +
      + +
      +

      Tutorials

      +

      Get in-depth tutorials for beginners and advanced developers

      + View Tutorials +
      + +
      +

      Resources

      +

      Find development resources and get your questions answered

      + View Resources +
      +
      +
      +
      + + + + + + + + + +
      +
      +
      +
      + + +
      +
      +
      + + +
      + + + + + + + + \ No newline at end of file diff --git a/docs/tutorials/_rendered_examples/dynamo/engine_caching_example.html b/docs/tutorials/_rendered_examples/dynamo/engine_caching_example.html new file mode 100644 index 0000000000..0d5f095791 --- /dev/null +++ b/docs/tutorials/_rendered_examples/dynamo/engine_caching_example.html @@ -0,0 +1,1093 @@ + + + + + + + + + + + + + Engine Caching — Torch-TensorRT v2.5.0.dev0+8759736 documentation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
      +
      +
      + + + + + +
      +
      +
      + + + + + + + + + + + +
      +
      +
      + + + + + + + + + + + + + + + + +
      + + + + +
      +
      + +
      + Shortcuts +
      +
      + +
      +
      + + + + + + +
      + +
      +
      + + +
      +

      Engine Caching

      +

      As model sizes increase, the cost of compilation will as well. With AOT methods +like torch.dynamo.compile, this cost is paid upfront. However if the weights +change, the session ends or you are using JIT methods like torch.compile, as +graphs get invalidated they get re-compiled, this cost will get paid repeatedly. +Engine caching is a way to mitigate this cost by saving constructed engines to disk +and re-using them when possible. This tutorial demonstrates how to use engine caching +with TensorRT in PyTorch. Engine caching can significantly speed up subsequent model +compilations reusing previously built TensorRT engines.

      +

      We’ll explore two approaches:

      +
      +
        +
      1. Using torch_tensorrt.dynamo.compile

      2. +
      3. Using torch.compile with the TensorRT backend

      4. +
      +
      +

      The example uses a pre-trained ResNet18 model and shows the +differences between compilation without caching, with caching enabled, +and when reusing cached engines.

      +
      import os
      +from typing import Dict, Optional
      +
      +import numpy as np
      +import torch
      +import torch_tensorrt as torch_trt
      +import torchvision.models as models
      +from torch_tensorrt.dynamo._defaults import TIMING_CACHE_PATH
      +from torch_tensorrt.dynamo._engine_cache import BaseEngineCache
      +
      +np.random.seed(0)
      +torch.manual_seed(0)
      +
      +model = models.resnet18(pretrained=True).eval().to("cuda")
      +enabled_precisions = {torch.float}
      +debug = False
      +min_block_size = 1
      +use_python_runtime = False
      +
      +
      +def remove_timing_cache(path=TIMING_CACHE_PATH):
      +    if os.path.exists(path):
      +        os.remove(path)
      +
      +
      +
      +

      Engine Caching for JIT Compilation

      +

      The primary goal of engine caching is to help speed up JIT workflows. torch.compile +provides a great deal of flexibility in model construction which makes it a good +first tool to try when looking to speed up your workflow. However, historically +the cost of compilation and in particular recompilation has been a barrier to entry +for many users. If for some reason a subgraph gets invalidated, that graph is reconstructed +scratch prior to the addition of engine caching. Now as engines are constructed, with cache_built_engines=True, +engines are saved to disk tied to a hash of their corresponding PyTorch subgraph. If +in a subsequent compilation, either as part of this session or a new session, the cache will +pull the built engine and refit the weights which can reduce compilation times by orders of magnitude. +As such, in order to insert a new engine into the cache (i.e. cache_built_engines=True), +the engine must be refitable (make_refittable=True). See Refitting Torch-TensorRT Programs with New Weights for more details.

      +
      def torch_compile(iterations=3):
      +    times = []
      +    start = torch.cuda.Event(enable_timing=True)
      +    end = torch.cuda.Event(enable_timing=True)
      +
      +    # The 1st iteration is to measure the compilation time without engine caching
      +    # The 2nd and 3rd iterations are to measure the compilation time with engine caching.
      +    # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration.
      +    # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine.
      +    for i in range(iterations):
      +        inputs = [torch.rand((100, 3, 224, 224)).to("cuda")]
      +        # remove timing cache and reset dynamo just for engine caching messurement
      +        remove_timing_cache()
      +        torch._dynamo.reset()
      +
      +        if i == 0:
      +            cache_built_engines = False
      +            reuse_cached_engines = False
      +        else:
      +            cache_built_engines = True
      +            reuse_cached_engines = True
      +
      +        start.record()
      +        compiled_model = torch.compile(
      +            model,
      +            backend="tensorrt",
      +            options={
      +                "use_python_runtime": True,
      +                "enabled_precisions": enabled_precisions,
      +                "debug": debug,
      +                "min_block_size": min_block_size,
      +                "make_refitable": True,
      +                "cache_built_engines": cache_built_engines,
      +                "reuse_cached_engines": reuse_cached_engines,
      +            },
      +        )
      +        compiled_model(*inputs)  # trigger the compilation
      +        end.record()
      +        torch.cuda.synchronize()
      +        times.append(start.elapsed_time(end))
      +
      +    print("----------------torch_compile----------------")
      +    print("disable engine caching, used:", times[0], "ms")
      +    print("enable engine caching to cache engines, used:", times[1], "ms")
      +    print("enable engine caching to reuse engines, used:", times[2], "ms")
      +
      +
      +torch_compile()
      +
      +
      +
      +
      +

      Engine Caching for AOT Compilation

      +

      Similarly to the JIT workflow, AOT workflows can benefit from engine caching. +As the same architecture or common subgraphs get recompiled, the cache will pull +previously built engines and refit the weights.

      +
      def dynamo_compile(iterations=3):
      +    times = []
      +    start = torch.cuda.Event(enable_timing=True)
      +    end = torch.cuda.Event(enable_timing=True)
      +
      +    example_inputs = (torch.randn((100, 3, 224, 224)).to("cuda"),)
      +    # Mark the dim0 of inputs as dynamic
      +    batch = torch.export.Dim("batch", min=1, max=200)
      +    exp_program = torch.export.export(
      +        model, args=example_inputs, dynamic_shapes={"x": {0: batch}}
      +    )
      +
      +    # The 1st iteration is to measure the compilation time without engine caching
      +    # The 2nd and 3rd iterations are to measure the compilation time with engine caching.
      +    # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration.
      +    # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine.
      +    for i in range(iterations):
      +        inputs = [torch.rand((100 + i, 3, 224, 224)).to("cuda")]
      +        remove_timing_cache()  # remove timing cache just for engine caching messurement
      +        if i == 0:
      +            cache_built_engines = False
      +            reuse_cached_engines = False
      +        else:
      +            cache_built_engines = True
      +            reuse_cached_engines = True
      +
      +        start.record()
      +        trt_gm = torch_trt.dynamo.compile(
      +            exp_program,
      +            tuple(inputs),
      +            use_python_runtime=use_python_runtime,
      +            enabled_precisions=enabled_precisions,
      +            debug=debug,
      +            min_block_size=min_block_size,
      +            make_refitable=True,
      +            cache_built_engines=cache_built_engines,
      +            reuse_cached_engines=reuse_cached_engines,
      +            engine_cache_size=1 << 30,  # 1GB
      +        )
      +        # output = trt_gm(*inputs)
      +        end.record()
      +        torch.cuda.synchronize()
      +        times.append(start.elapsed_time(end))
      +
      +    print("----------------dynamo_compile----------------")
      +    print("disable engine caching, used:", times[0], "ms")
      +    print("enable engine caching to cache engines, used:", times[1], "ms")
      +    print("enable engine caching to reuse engines, used:", times[2], "ms")
      +
      +
      +dynamo_compile()
      +
      +
      +
      +
      +

      Custom Engine Cache

      +

      By default, the engine cache is stored in the system’s temporary directory. Both the cache directory and +size limit can be customized by passing engine_cache_dir and engine_cache_size. +Users can also define their own engine cache implementation by extending the BaseEngineCache class. +This allows for remote or shared caching if so desired.

      +
      +
      The custom engine cache should implement the following methods:
        +
      • save: Save the engine blob to the cache.

      • +
      • load: Load the engine blob from the cache.

      • +
      +
      +
      +

      The hash provided by the cache systen is a weight agnostic hash of the originating PyTorch subgraph (post lowering). +The blob contains a serialized engine, calling spec data, and weight map information in the pickle format

      +

      Below is an example of a custom engine cache implementation that implents a RAMEngineCache.

      +
      class RAMEngineCache(BaseEngineCache):
      +    def __init__(
      +        self,
      +    ) -> None:
      +        """
      +        Constructs a user held engine cache in memory.
      +        """
      +        self.engine_cache: Dict[str, bytes] = {}
      +
      +    def save(
      +        self,
      +        hash: str,
      +        blob: bytes,
      +    ):
      +        """
      +        Insert the engine blob to the cache.
      +
      +        Args:
      +            hash (str): The hash key to associate with the engine blob.
      +            blob (bytes): The engine blob to be saved.
      +
      +        Returns:
      +            None
      +        """
      +        self.engine_cache[hash] = blob
      +
      +    def load(self, hash: str) -> Optional[bytes]:
      +        """
      +        Load the engine blob from the cache.
      +
      +        Args:
      +            hash (str): The hash key of the engine to load.
      +
      +        Returns:
      +            Optional[bytes]: The engine blob if found, None otherwise.
      +        """
      +        if hash in self.engine_cache:
      +            return self.engine_cache[hash]
      +        else:
      +            return None
      +
      +
      +def torch_compile_my_cache(iterations=3):
      +    times = []
      +    engine_cache = RAMEngineCache()
      +    start = torch.cuda.Event(enable_timing=True)
      +    end = torch.cuda.Event(enable_timing=True)
      +
      +    # The 1st iteration is to measure the compilation time without engine caching
      +    # The 2nd and 3rd iterations are to measure the compilation time with engine caching.
      +    # Since the 2nd iteration needs to compile and save the engine, it will be slower than the 1st iteration.
      +    # The 3rd iteration should be faster than the 1st iteration because it loads the cached engine.
      +    for i in range(iterations):
      +        inputs = [torch.rand((100, 3, 224, 224)).to("cuda")]
      +        # remove timing cache and reset dynamo just for engine caching messurement
      +        remove_timing_cache()
      +        torch._dynamo.reset()
      +
      +        if i == 0:
      +            cache_built_engines = False
      +            reuse_cached_engines = False
      +        else:
      +            cache_built_engines = True
      +            reuse_cached_engines = True
      +
      +        start.record()
      +        compiled_model = torch.compile(
      +            model,
      +            backend="tensorrt",
      +            options={
      +                "use_python_runtime": True,
      +                "enabled_precisions": enabled_precisions,
      +                "debug": debug,
      +                "min_block_size": min_block_size,
      +                "make_refitable": True,
      +                "cache_built_engines": cache_built_engines,
      +                "reuse_cached_engines": reuse_cached_engines,
      +                "custom_engine_cache": engine_cache,
      +            },
      +        )
      +        compiled_model(*inputs)  # trigger the compilation
      +        end.record()
      +        torch.cuda.synchronize()
      +        times.append(start.elapsed_time(end))
      +
      +    print("----------------torch_compile----------------")
      +    print("disable engine caching, used:", times[0], "ms")
      +    print("enable engine caching to cache engines, used:", times[1], "ms")
      +    print("enable engine caching to reuse engines, used:", times[2], "ms")
      +
      +
      +torch_compile_my_cache()
      +
      +
      +

      Total running time of the script: ( 0 minutes 0.000 seconds)

      + +

      Gallery generated by Sphinx-Gallery

      +
      +
      + + +
      + +
      + + +
      +
      + + +
      +
      + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
      +
      +
      +
      +

      Docs

      +

      Access comprehensive developer documentation for PyTorch

      + View Docs +
      + +
      +

      Tutorials

      +

      Get in-depth tutorials for beginners and advanced developers

      + View Tutorials +
      + +
      +

      Resources

      +

      Find development resources and get your questions answered

      + View Resources +
      +
      +
      +
      + + + + + + + + + +
      +
      +
      +
      + + +
      +
      +
      + + +
      + + + + + + + + \ No newline at end of file diff --git a/docs/tutorials/_rendered_examples/dynamo/index.html b/docs/tutorials/_rendered_examples/dynamo/index.html index be6b1c8ea8..7125ab06d8 100644 --- a/docs/tutorials/_rendered_examples/dynamo/index.html +++ b/docs/tutorials/_rendered_examples/dynamo/index.html @@ -10,7 +10,7 @@ - Dynamo / torch.compile — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Dynamo / torch.compile — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Torch Compile Stable Diffusion

    @@ -474,15 +478,24 @@

    Torch Export with Cudagraphs

    Torch Export with Cudagraphs
    +
    +

    Refitting Torch-TensorRT Programs with New Weights

    +
    Refitting Torch-TensorRT Programs with New Weights

    Compiling a Transformer using torch.compile and TensorRT

    Compiling a Transformer using torch.compile and TensorRT
    -
    -

    Refit TenorRT Graph Module with Torch-TensorRT

    -
    Refit TenorRT Graph Module with Torch-TensorRT
    +
    +

    Compiling GPT2 using the Torch-TensorRT with dynamo backend

    +
    Compiling GPT2 using the Torch-TensorRT with dynamo backend

    Torch Compile Advanced Usage

    Torch Compile Advanced Usage
    +
    +

    Compiling Llama2 using the Torch-TensorRT with dynamo backend

    +
    Compiling Llama2 using the Torch-TensorRT with dynamo backend
    +
    +

    Engine Caching (BERT)

    +
    Engine Caching (BERT)

    Mutable Torch TensorRT Module

    Mutable Torch TensorRT Module
    @@ -492,6 +505,9 @@

    Deploy Quantized Models using Torch-TensorRT

    Deploy Quantized Models using Torch-TensorRT
    +
    +

    Engine Caching

    +
    Engine Caching

    Using Custom Kernels within TensorRT Engines with Torch-TensorRT

    Using Custom Kernels within TensorRT Engines with Torch-TensorRT
    diff --git a/docs/tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example.html b/docs/tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example.html index b6d222ca05..3b787c393d 100644 --- a/docs/tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example.html +++ b/docs/tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example.html @@ -10,7 +10,7 @@ - Mutable Torch TensorRT Module — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Mutable Torch TensorRT Module — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
    @@ -316,6 +316,8 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Engine Caching
  • +
  • Refitting Torch-TensorRT Programs with New Weights
  • Dynamo Frontend

      diff --git a/docs/tutorials/_rendered_examples/dynamo/refit_engine_example.html b/docs/tutorials/_rendered_examples/dynamo/refit_engine_example.html index ad1b46f3bb..226f25ca43 100644 --- a/docs/tutorials/_rendered_examples/dynamo/refit_engine_example.html +++ b/docs/tutorials/_rendered_examples/dynamo/refit_engine_example.html @@ -10,7 +10,7 @@ - Refit TenorRT Graph Module with Torch-TensorRT — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Refitting Torch-TensorRT Programs with New Weights — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -39,6 +39,8 @@ + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
      +
      +
      + + + + + +
      +
      +
      + + + + + + + + + + + +
      +
      +
      + + + + + + + + + + + + + + + + +
      + +
        + +
      • + + + Docs + + > +
      • + + +
      • Compiling GPT2 using the Torch-TensorRT with dynamo backend
      • + + +
      • + + + + + +
      • + +
      + + +
      +
      + +
      + Shortcuts +
      +
      + +
      +
      + + + + + + +
      + +
      +
      + + +
      +

      Compiling GPT2 using the Torch-TensorRT with dynamo backend

      +

      This interactive script is intended as a sample of the Torch-TensorRT workflow with dynamo backend on a GPT2 model.

      +
      +

      Imports and Model Definition

      +
      import torch
      +import torch_tensorrt
      +from transformers import AutoModelForCausalLM, AutoTokenizer
      +from utils import export_llm, generate
      +
      +
      +
      # Define the parameters and initialize the model
      +MAX_TOKENS = 32
      +DEVICE = torch.device("cuda:0")
      +
      +# Define the GPT2 model from hugging face
      +# kv_cache is not supported in Torch-TRT currently.
      +# CPU is used here so that GPU memory is reserved for TRT compilation.
      +with torch.no_grad():
      +    tokenizer = AutoTokenizer.from_pretrained("gpt2")
      +    model = AutoModelForCausalLM.from_pretrained(
      +        "gpt2",
      +        pad_token_id=tokenizer.eos_token_id,
      +        use_cache=False,
      +        attn_implementation="eager",
      +    ).eval()
      +
      +
      +

      Tokenize a sample input prompt and get pytorch model outputs

      +
      prompt = "I enjoy walking with my cute dog"
      +model_inputs = tokenizer(prompt, return_tensors="pt")
      +input_ids = model_inputs["input_ids"]
      +
      +# Auto-regressive generation loop for greedy decoding using PyTorch model
      +# We use a custom generate function which is very similar to the huggingface one.
      +pyt_gen_tokens = generate(model, input_ids, MAX_TOKENS, tokenizer.eos_token_id)
      +
      +
      +
      +
      +

      Compilation with Torch-TensorRT using dynamo backend and generate TensorRT outputs

      +
      # Export the GPT2 model into an ExportedProgram which is input of TRT compilation
      +gpt2_ep = export_llm(model, input_ids, max_seq_len=1024)
      +trt_model = torch_tensorrt.dynamo.compile(
      +    gpt2_ep,
      +    inputs=[input_ids],
      +    enabled_precisions={torch.float32},
      +    truncate_double=True,
      +    device=DEVICE,
      +    disable_tf32=True,
      +)
      +
      +# Auto-regressive generation loop for greedy decoding using TensorRT model
      +# We use a custom generate function which is very similar to the huggingface one.
      +# Move inputs to GPU
      +input_ids = input_ids.to(DEVICE)
      +trt_gen_tokens = generate(trt_model, input_ids, MAX_TOKENS, tokenizer.eos_token_id)
      +
      +
      +
      +
      +

      Decode the output sentences of PyTorch and TensorRT

      +
      print("=============================")
      +print(
      +    "Pytorch model generated text: ",
      +    tokenizer.decode(pyt_gen_tokens[0], skip_special_tokens=True),
      +)
      +print("=============================")
      +print(
      +    "TensorRT model generated text: ",
      +    tokenizer.decode(trt_gen_tokens[0], skip_special_tokens=True),
      +)
      +
      +
      +
      +
      +
      +

      The output sentences should look like

      +
      +
      +

      Pytorch model generated text: I enjoy walking with my cute dog, but I’m not sure if I’ll ever be able to walk with my dog. I’m not sure if I’ll ever be able to walk with my

      +

      TensorRT model generated text: I enjoy walking with my cute dog, but I’m not sure if I’ll ever be able to walk with my dog. I’m not sure if I’ll ever be able to walk with my

      +

      Total running time of the script: ( 0 minutes 0.000 seconds)

      + +

      Gallery generated by Sphinx-Gallery

      +
      + + +
      + +
      +
      + + + + +
      + + + +
      +

      + © Copyright 2024, NVIDIA Corporation. + +

      +
      + +
      + Built with Sphinx using a theme provided by Read the Docs. +
      + + +
      + +
      +
      + + +
      +
      + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
      +
      +
      +
      +

      Docs

      +

      Access comprehensive developer documentation for PyTorch

      + View Docs +
      + +
      +

      Tutorials

      +

      Get in-depth tutorials for beginners and advanced developers

      + View Tutorials +
      + +
      +

      Resources

      +

      Find development resources and get your questions answered

      + View Resources +
      +
      +
      +
      + + + + + + + + + +
      +
      +
      +
      + + +
      +
      +
      + + +
      + + + + + + + + \ No newline at end of file diff --git a/docs/tutorials/_rendered_examples/dynamo/torch_export_llama2.html b/docs/tutorials/_rendered_examples/dynamo/torch_export_llama2.html new file mode 100644 index 0000000000..14f2207626 --- /dev/null +++ b/docs/tutorials/_rendered_examples/dynamo/torch_export_llama2.html @@ -0,0 +1,893 @@ + + + + + + + + + + + + + Compiling Llama2 using the Torch-TensorRT with dynamo backend — Torch-TensorRT v2.5.0.dev0+8759736 documentation + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
      +
      +
      + + + + + +
      +
      +
      + + + + + + + + + + + +
      +
      +
      + + + + + + + + + + + + + + + + +
      + +
        + +
      • + + + Docs + + > +
      • + + +
      • Compiling Llama2 using the Torch-TensorRT with dynamo backend
      • + + +
      • + + + + + +
      • + +
      + + +
      +
      + +
      + Shortcuts +
      +
      + +
      +
      + + + + + + +
      + +
      +
      + + +
      +

      Compiling Llama2 using the Torch-TensorRT with dynamo backend

      +

      This interactive script is intended as a sample of the Torch-TensorRT workflow with dynamo backend on a Llama2 model.

      +
      +

      Imports and Model Definition

      +
      import torch
      +import torch_tensorrt
      +from transformers import AutoModelForCausalLM, AutoTokenizer
      +from utils import export_llm, generate
      +
      +
      +

      Define the parameters and initialize the model

      +
      MAX_TOKENS = 32
      +DEVICE = torch.device("cuda:0")
      +
      +# Define the Llama2 model from hugging face
      +# kv_cache is not supported in Torch-TRT currently.
      +# CPU is used here so that GPU memory is reserved for TRT compilation.
      +llama_path = "meta-llama/Llama-2-7b-chat-hf"
      +with torch.no_grad():
      +    model = AutoModelForCausalLM.from_pretrained(
      +        llama_path, use_cache=False, attn_implementation="eager"
      +    ).eval()
      +
      +tokenizer = AutoTokenizer.from_pretrained(llama_path)
      +
      +
      +

      Tokenize a sample input prompt and get pytorch model outputs

      +
      prompt = "What is dynamic programming?"
      +model_inputs = tokenizer(prompt, return_tensors="pt")
      +input_ids = model_inputs.input_ids
      +
      +# Auto-regressive generation loop for greedy decoding using PyTorch model
      +# We use a custom generate function which is very similar to the huggingface one.
      +pyt_gen_tokens = generate(model, input_ids, MAX_TOKENS, tokenizer.eos_token_id)
      +
      +
      +
      +
      +

      Compilation with Torch-TensorRT using dynamo backend and generate TensorRT outputs

      +
      # Export the llama2 model into an ExportedProgram which is input of TRT compilation
      +llama2_ep = export_llm(model, input_ids, max_seq_len=64)
      +trt_model = torch_tensorrt.dynamo.compile(
      +    llama2_ep,
      +    inputs=[input_ids],
      +    enabled_precisions={torch.float32},
      +    min_block_size=1,
      +    truncate_double=True,
      +    device=DEVICE,
      +    disable_tf32=True,
      +)
      +
      +# Auto-regressive generation loop for greedy decoding using TensorRT model
      +# We use a custom generate function which is very similar to the huggingface one.
      +# Move inputs to GPU
      +input_ids = input_ids.to(DEVICE)
      +trt_gen_tokens = generate(trt_model, input_ids, MAX_TOKENS, tokenizer.eos_token_id)
      +
      +
      +
      +
      +

      Decode the output sentences of PyTorch and TensorRT

      +
      print("=============================")
      +print(
      +    "Pytorch model generated text: ",
      +    tokenizer.batch_decode(
      +        pyt_gen_tokens, skip_special_tokens=True, clean_up_tokenization_spaces=False
      +    )[0],
      +)
      +print("=============================")
      +print(
      +    "TensorRT model generated text: ",
      +    tokenizer.batch_decode(
      +        trt_gen_tokens,
      +        skip_special_tokens=True,
      +        clean_up_tokenization_spaces=False,
      +    )[0],
      +)
      +
      +
      +
      +
      +
      +

      The output sentences should look like

      +
      +
      +

      Pytorch model generated text: I enjoy walking with my cute dog, but I’m not sure if I’ll ever be able to walk with my dog. I’m not sure if I’ll ever be able to walk with my

      +

      TensorRT model generated text: I enjoy walking with my cute dog, but I’m not sure if I’ll ever be able to walk with my dog. I’m not sure if I’ll ever be able to walk with my

      +

      Total running time of the script: ( 0 minutes 0.000 seconds)

      + +

      Gallery generated by Sphinx-Gallery

      +
      + + +
      + +
      +
      + + + + +
      + + + +
      +

      + © Copyright 2024, NVIDIA Corporation. + +

      +
      + +
      + Built with Sphinx using a theme provided by Read the Docs. +
      + + +
      + +
      +
      + + +
      +
      + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
      +
      +
      +
      +

      Docs

      +

      Access comprehensive developer documentation for PyTorch

      + View Docs +
      + +
      +

      Tutorials

      +

      Get in-depth tutorials for beginners and advanced developers

      + View Tutorials +
      + +
      +

      Resources

      +

      Find development resources and get your questions answered

      + View Resources +
      +
      +
      +
      + + + + + + + + + +
      +
      +
      +
      + + +
      +
      +
      + + +
      + + + + + + + + \ No newline at end of file diff --git a/docs/tutorials/_rendered_examples/dynamo/vgg16_ptq.html b/docs/tutorials/_rendered_examples/dynamo/vgg16_ptq.html index c39e454736..fbcc228e7a 100644 --- a/docs/tutorials/_rendered_examples/dynamo/vgg16_ptq.html +++ b/docs/tutorials/_rendered_examples/dynamo/vgg16_ptq.html @@ -10,7 +10,7 @@ - Deploy Quantized Models using Torch-TensorRT — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Deploy Quantized Models using Torch-TensorRT — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

      diff --git a/docs/tutorials/_rendered_examples/index.html b/docs/tutorials/_rendered_examples/index.html index 2df289be4a..c73d007b2b 100644 --- a/docs/tutorials/_rendered_examples/index.html +++ b/docs/tutorials/_rendered_examples/index.html @@ -10,7 +10,7 @@ - Torch-TensorRT Tutorials — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Torch-TensorRT Tutorials — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -273,7 +273,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -314,6 +314,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Torch Compile Stable Diffusion

    @@ -478,15 +482,24 @@

    Dynamo / torch.

    Torch Export with Cudagraphs

    Torch Export with Cudagraphs
    +
    +

    Refitting Torch-TensorRT Programs with New Weights

    +
    Refitting Torch-TensorRT Programs with New Weights

    Compiling a Transformer using torch.compile and TensorRT

    Compiling a Transformer using torch.compile and TensorRT
    -
    -

    Refit TenorRT Graph Module with Torch-TensorRT

    -
    Refit TenorRT Graph Module with Torch-TensorRT
    +
    +

    Compiling GPT2 using the Torch-TensorRT with dynamo backend

    +
    Compiling GPT2 using the Torch-TensorRT with dynamo backend

    Torch Compile Advanced Usage

    Torch Compile Advanced Usage
    +
    +

    Compiling Llama2 using the Torch-TensorRT with dynamo backend

    +
    Compiling Llama2 using the Torch-TensorRT with dynamo backend
    +
    +

    Engine Caching (BERT)

    +
    Engine Caching (BERT)

    Mutable Torch TensorRT Module

    Mutable Torch TensorRT Module
    @@ -496,6 +509,9 @@

    Dynamo / torch.

    Deploy Quantized Models using Torch-TensorRT

    Deploy Quantized Models using Torch-TensorRT
    +
    +

    Engine Caching

    +
    Engine Caching

    Using Custom Kernels within TensorRT Engines with Torch-TensorRT

    Using Custom Kernels within TensorRT Engines with Torch-TensorRT
    diff --git a/docs/tutorials/notebooks.html b/docs/tutorials/notebooks.html index a4edc64f90..794f4231cf 100644 --- a/docs/tutorials/notebooks.html +++ b/docs/tutorials/notebooks.html @@ -10,7 +10,7 @@ - Example notebooks — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Example notebooks — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
    @@ -316,6 +316,8 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Engine Caching
  • +
  • Refitting Torch-TensorRT Programs with New Weights
  • Dynamo Frontend

      diff --git a/docs/tutorials/serving_torch_tensorrt_with_triton.html b/docs/tutorials/serving_torch_tensorrt_with_triton.html index 3f41b7ce44..9b106e7ffb 100644 --- a/docs/tutorials/serving_torch_tensorrt_with_triton.html +++ b/docs/tutorials/serving_torch_tensorrt_with_triton.html @@ -10,7 +10,7 @@ - Serving a Torch-TensorRT model with Triton — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Serving a Torch-TensorRT model with Triton — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

    Dynamo Frontend

      diff --git a/docs/user_guide/torch_tensorrt_explained.html b/docs/user_guide/torch_tensorrt_explained.html index e93a2c6a02..97be66c8b2 100644 --- a/docs/user_guide/torch_tensorrt_explained.html +++ b/docs/user_guide/torch_tensorrt_explained.html @@ -10,7 +10,7 @@ - Torch-TensorRT Explained — Torch-TensorRT v2.5.0.dev0+b3a8cdd documentation + Torch-TensorRT Explained — Torch-TensorRT v2.5.0.dev0+8759736 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+b3a8cdd + v2.5.0.dev0+8759736
      @@ -316,6 +316,8 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Engine Caching
    • +
    • Refitting Torch-TensorRT Programs with New Weights

    Dynamo Frontend

    Dynamo Frontend

      From a4a9419ad68d6440a890927021289fba4a03756b Mon Sep 17 00:00:00 2001 From: HolyWu Date: Thu, 5 Sep 2024 01:28:42 +0800 Subject: [PATCH 07/14] Fix doc index (#3130) Co-authored-by: Naren Dasan <1790613+narendasan@users.noreply.github.com> --- docsrc/index.rst | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/docsrc/index.rst b/docsrc/index.rst index d1a91beabc..757acc2011 100644 --- a/docsrc/index.rst +++ b/docsrc/index.rst @@ -44,13 +44,12 @@ User Guide :hidden: user_guide/torch_tensorrt_explained - user_guide/getting_started user_guide/dynamic_shapes user_guide/saving_models user_guide/runtime user_guide/using_dla tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage - tutorials/_rendered_examples/dynamo/vgg16_fp8_ptq + tutorials/_rendered_examples/dynamo/vgg16_ptq tutorials/_rendered_examples/dynamo/engine_caching_example tutorials/_rendered_examples/dynamo/refit_engine_example @@ -113,13 +112,11 @@ Tutorials tutorials/notebooks tutorials/_rendered_examples/dynamo/torch_compile_resnet_example tutorials/_rendered_examples/dynamo/torch_compile_transformers_example - tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage tutorials/_rendered_examples/dynamo/torch_compile_stable_diffusion tutorials/_rendered_examples/dynamo/torch_export_cudagraphs tutorials/_rendered_examples/dynamo/custom_kernel_plugins tutorials/_rendered_examples/distributed_inference/data_parallel_gpt2 tutorials/_rendered_examples/distributed_inference/data_parallel_stable_diffusion - tutorials/_rendered_examples/dynamo/vgg16_fp8_ptq tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example Python API Documentation From 29b4913b1fb03f55552e4be2f9d79f709b7905d5 Mon Sep 17 00:00:00 2001 From: Torch-TensorRT Github Bot Date: Wed, 4 Sep 2024 17:44:31 +0000 Subject: [PATCH 08/14] docs: [Automated] Regenerating documenation for a4a9419 Signed-off-by: Torch-TensorRT Github Bot --- .../classtorch__tensorrt_1_1DataType.html | 6 +++--- ...rch__tensorrt_1_1Device_1_1DeviceType.html | 6 +++--- .../classtorch__tensorrt_1_1TensorFormat.html | 6 +++--- ...ensorrt_1_1ptq_1_1Int8CacheCalibrator.html | 6 +++--- ...ch__tensorrt_1_1ptq_1_1Int8Calibrator.html | 6 +++--- ...8h_1a18d295a837ac71add5578860b55e5502.html | 6 +++--- ...8h_1a282fd3c0b1c3a215148ae372070e1268.html | 6 +++--- ...8h_1a31398a6d4d27e28817afb0f0139e909e.html | 6 +++--- ...8h_1a35703561b26b1a9d2738ad7d58b27827.html | 6 +++--- ...8h_1abd1465eb38256d3f22cc1426b23d516b.html | 6 +++--- ...8h_1abe87b341f562fd1cf40b7672e4d759da.html | 6 +++--- ...8h_1ad19939408f7be171a74a89928b36eb59.html | 6 +++--- ...8h_1adad592a7b1b7eed529cdf6acd584c883.html | 6 +++--- docs/_cpp_api/dir_cpp.html | 6 +++--- docs/_cpp_api/dir_cpp_include.html | 6 +++--- .../dir_cpp_include_torch_tensorrt.html | 6 +++--- ...ng_1a130f65408ad8cbaee060f05e8db69558.html | 6 +++--- ...rt_1a3fbe5d72e4fc624dbd038853079620eb.html | 6 +++--- ..._cpp_include_torch_tensorrt_logging.h.html | 6 +++--- ...e_cpp_include_torch_tensorrt_macros.h.html | 6 +++--- ...file_cpp_include_torch_tensorrt_ptq.h.html | 6 +++--- ...clude_torch_tensorrt_torch_tensorrt.h.html | 6 +++--- ...ng_1a0593f776f469c20469e2f729fc7861a3.html | 6 +++--- ...ng_1a0c012cb374addd90eb1f42eaec570650.html | 6 +++--- ...ng_1a56e110feaaba2c3fd44bd201fd21a76a.html | 6 +++--- ...ng_1a7cb50492421ea9de4e3db895819df6f2.html | 6 +++--- ...ng_1ac46ac0901cb97e3ae6e93b45f24e90b8.html | 6 +++--- ...ng_1ad2efd47b6c3689e58ccc595680579ae5.html | 6 +++--- ...ng_1af8f3443813315af7901903d25dd495cc.html | 6 +++--- ...tq_1a226e3c83379d1012cde8578c1c86b16c.html | 6 +++--- ...tq_1a6186e305f47c1d94b6130ef6c7f7e178.html | 6 +++--- ...pt_1a5b405fd3bf3c8fc2e2a54cbbab979797.html | 6 +++--- ...pt_1a6e19490a08fb1553c9dd347a5ae79db9.html | 6 +++--- ...pt_1a81f9783517335dda877d8cfcf38987c9.html | 6 +++--- ...pt_1ae8d56472106eeef37fbe51ff7f40c9b2.html | 6 +++--- ...rt_1ac4ab8313ae72c2c899ea31548b528528.html | 6 +++--- ...rt_1ad1acd06eaeaffbbcf6e7ebf426891384.html | 6 +++--- ...rt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html | 6 +++--- docs/_cpp_api/namespace_torch_tensorrt.html | 6 +++--- .../namespace_torch_tensorrt__logging.html | 6 +++--- .../namespace_torch_tensorrt__ptq.html | 6 +++--- ...namespace_torch_tensorrt__torchscript.html | 6 +++--- ..._cpp_include_torch_tensorrt_logging.h.html | 6 +++--- ...e_cpp_include_torch_tensorrt_macros.h.html | 6 +++--- ...file_cpp_include_torch_tensorrt_ptq.h.html | 6 +++--- ...clude_torch_tensorrt_torch_tensorrt.h.html | 6 +++--- .../structtorch__tensorrt_1_1Device.html | 6 +++--- .../structtorch__tensorrt_1_1GraphInputs.html | 6 +++--- .../structtorch__tensorrt_1_1Input.html | 6 +++--- ...ensorrt_1_1torchscript_1_1CompileSpec.html | 6 +++--- docs/_cpp_api/torch_tensort_cpp.html | 8 ++++---- docs/_cpp_api/unabridged_orphan.html | 6 +++--- .../_rendered_examples_jupyter.zip | Bin 103522 -> 103522 bytes .../_rendered_examples_python.zip | Bin 78451 -> 78451 bytes docs/_modules/index.html | 6 +++--- docs/_modules/torch_tensorrt/_Device.html | 6 +++--- docs/_modules/torch_tensorrt/_Input.html | 6 +++--- docs/_modules/torch_tensorrt/_compile.html | 6 +++--- docs/_modules/torch_tensorrt/_enums.html | 6 +++--- .../torch_tensorrt/dynamo/_compiler.html | 6 +++--- .../torch_tensorrt/dynamo/_exporter.html | 6 +++--- .../torch_tensorrt/dynamo/_refit.html | 6 +++--- .../torch_tensorrt/dynamo/_settings.html | 6 +++--- .../torch_tensorrt/dynamo/_tracer.html | 6 +++--- .../runtime/_MutableTorchTensorRTModule.html | 6 +++--- .../runtime/_PythonTorchTensorRTModule.html | 6 +++--- .../dynamo/runtime/_TorchTensorRTModule.html | 6 +++--- docs/_modules/torch_tensorrt/fx/fx2trt.html | 6 +++--- .../torch_tensorrt/fx/input_tensor_spec.html | 6 +++--- docs/_modules/torch_tensorrt/fx/lower.html | 6 +++--- .../torch_tensorrt/fx/trt_module.html | 6 +++--- docs/_modules/torch_tensorrt/logging.html | 6 +++--- .../runtime/_multi_device_safe_mode.html | 6 +++--- .../torch_tensorrt/ts/_compile_spec.html | 6 +++--- .../_modules/torch_tensorrt/ts/_compiler.html | 6 +++--- docs/_modules/torch_tensorrt/ts/ptq.html | 6 +++--- docs/_sources/index.rst.txt | 5 +---- docs/_static/documentation_options.js | 2 +- docs/cli/torchtrtc.html | 6 +++--- docs/contributors/conversion.html | 6 +++--- docs/contributors/dynamo_converters.html | 6 +++--- docs/contributors/lowering.html | 6 +++--- docs/contributors/partitioning.html | 6 +++--- docs/contributors/phases.html | 6 +++--- docs/contributors/runtime.html | 6 +++--- docs/contributors/system_overview.html | 6 +++--- docs/contributors/ts_converters.html | 6 +++--- docs/contributors/useful_links.html | 6 +++--- .../writing_dynamo_aten_lowering_passes.html | 6 +++--- docs/dynamo/dynamo_export.html | 6 +++--- docs/dynamo/torch_compile.html | 6 +++--- docs/fx/getting_started_with_fx_path.html | 6 +++--- docs/genindex.html | 6 +++--- docs/getting_started/installation.html | 6 +++--- docs/getting_started/quick_start.html | 6 +++--- docs/index.html | 6 +++--- docs/indices/supported_ops.html | 6 +++--- docs/objects.inv | Bin 31865 -> 31865 bytes docs/py-modindex.html | 6 +++--- docs/py_api/dynamo.html | 6 +++--- docs/py_api/fx.html | 6 +++--- docs/py_api/logging.html | 6 +++--- docs/py_api/ptq.html | 6 +++--- docs/py_api/runtime.html | 6 +++--- docs/py_api/torch_tensorrt.html | 6 +++--- docs/py_api/ts.html | 8 ++++---- docs/search.html | 6 +++--- docs/searchindex.js | 2 +- docs/sg_execution_times.html | 6 +++--- .../pytorch-sphinx-theme/docs/changelog.html | 6 +++--- .../docs/configuring.html | 6 +++--- .../pytorch-sphinx-theme/docs/demo/api.html | 6 +++--- .../pytorch-sphinx-theme/docs/demo/demo.html | 8 ++++---- .../docs/demo/lists_tables.html | 6 +++--- .../pytorch-sphinx-theme/docs/demo/long.html | 6 +++--- .../docs/demo/structure.html | 6 +++--- docs/src/pytorch-sphinx-theme/docs/index.html | 6 +++--- .../pytorch-sphinx-theme/docs/installing.html | 6 +++--- ...creating_torchscript_module_in_python.html | 6 +++--- docs/ts/getting_started_with_cpp_api.html | 6 +++--- docs/ts/getting_started_with_python_api.html | 6 +++--- docs/ts/ptq.html | 6 +++--- .../ts/torchscript_frontend_from_pytorch.html | 6 +++--- .../dynamo/custom_kernel_plugins.html | 6 +++--- .../dynamo/engine_caching_bert_example.html | 6 +++--- .../dynamo/engine_caching_example.html | 10 ++++----- .../_rendered_examples/dynamo/index.html | 6 +++--- .../mutable_torchtrt_module_example.html | 6 +++--- .../dynamo/refit_engine_example.html | 6 +++--- .../dynamo/torch_compile_advanced_usage.html | 12 +++++------ .../dynamo/torch_compile_resnet_example.html | 6 +++--- .../torch_compile_stable_diffusion.html | 6 +++--- .../torch_compile_transformers_example.html | 6 +++--- .../dynamo/torch_export_cudagraphs.html | 6 +++--- .../dynamo/torch_export_gpt2.html | 6 +++--- .../dynamo/torch_export_llama2.html | 6 +++--- .../_rendered_examples/dynamo/vgg16_ptq.html | 19 ++++++++++++++---- docs/tutorials/_rendered_examples/index.html | 6 +++--- docs/tutorials/notebooks.html | 6 +++--- .../serving_torch_tensorrt_with_triton.html | 6 +++--- docs/user_guide/dynamic_shapes.html | 6 +++--- docs/user_guide/runtime.html | 6 +++--- docs/user_guide/saving_models.html | 6 +++--- docs/user_guide/torch_tensorrt_explained.html | 6 +++--- docs/user_guide/using_dla.html | 6 +++--- 145 files changed, 440 insertions(+), 432 deletions(-) diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1DataType.html b/docs/_cpp_api/classtorch__tensorrt_1_1DataType.html index 5261a19ab4..bf2a98805e 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1DataType.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1DataType.html @@ -10,7 +10,7 @@ - Class DataType — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Class DataType — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
      - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
      @@ -316,6 +316,7 @@
    • Deploying Torch-TensorRT Programs
    • DLA
    • Torch Compile Advanced Usage
    • +
    • Deploy Quantized Models using Torch-TensorRT
    • Engine Caching
    • Refitting Torch-TensorRT Programs with New Weights
    @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.html b/docs/_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.html index db11858bb9..b6e941d3c1 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.html @@ -10,7 +10,7 @@ - Class Device::DeviceType — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Class Device::DeviceType — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1TensorFormat.html b/docs/_cpp_api/classtorch__tensorrt_1_1TensorFormat.html index a886b42cc9..4cfffeea31 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1TensorFormat.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1TensorFormat.html @@ -10,7 +10,7 @@ - Class TensorFormat — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Class TensorFormat — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.html b/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.html index 9f2904306f..86212bd359 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.html @@ -10,7 +10,7 @@ - Template Class Int8CacheCalibrator — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Template Class Int8CacheCalibrator — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.html b/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.html index 524ac6c6e5..5c05cefe88 100644 --- a/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.html +++ b/docs/_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.html @@ -10,7 +10,7 @@ - Template Class Int8Calibrator — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Template Class Int8Calibrator — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.html b/docs/_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.html index d81a35efa7..7479a7bcc5 100644 --- a/docs/_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.html +++ b/docs/_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.html @@ -10,7 +10,7 @@ - Define STR — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Define STR — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.html b/docs/_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.html index 5585b7b9bb..15ea36fa48 100644 --- a/docs/_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.html +++ b/docs/_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.html @@ -10,7 +10,7 @@ - Define TORCH_TENSORRT_PATCH_VERSION — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Define TORCH_TENSORRT_PATCH_VERSION — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.html b/docs/_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.html index be9d363205..95eaefaf9b 100644 --- a/docs/_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.html +++ b/docs/_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.html @@ -10,7 +10,7 @@ - Define TORCH_TENSORRT_MAJOR_VERSION — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Define TORCH_TENSORRT_MAJOR_VERSION — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.html b/docs/_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.html index ac1ae177d2..a6b97324d5 100644 --- a/docs/_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.html +++ b/docs/_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.html @@ -10,7 +10,7 @@ - Define TORCH_TENSORRT_MINOR_VERSION — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Define TORCH_TENSORRT_MINOR_VERSION — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.html b/docs/_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.html index 481e0df5f5..12fda77cb5 100644 --- a/docs/_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.html +++ b/docs/_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.html @@ -10,7 +10,7 @@ - Define TORCHTRT_API — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Define TORCHTRT_API — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.html b/docs/_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.html index 820561ec0a..13de35adbc 100644 --- a/docs/_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.html +++ b/docs/_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.html @@ -10,7 +10,7 @@ - Define XSTR — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Define XSTR — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.html b/docs/_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.html index e4b2984c99..bbf22b806f 100644 --- a/docs/_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.html +++ b/docs/_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.html @@ -10,7 +10,7 @@ - Define TORCHTRT_HIDDEN — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Define TORCHTRT_HIDDEN — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.html b/docs/_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.html index 6a24f61376..5068def6da 100644 --- a/docs/_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.html +++ b/docs/_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.html @@ -10,7 +10,7 @@ - Define TORCH_TENSORRT_VERSION — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Define TORCH_TENSORRT_VERSION — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/dir_cpp.html b/docs/_cpp_api/dir_cpp.html index 10070969be..205143791a 100644 --- a/docs/_cpp_api/dir_cpp.html +++ b/docs/_cpp_api/dir_cpp.html @@ -10,7 +10,7 @@ - Directory cpp — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Directory cpp — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/dir_cpp_include.html b/docs/_cpp_api/dir_cpp_include.html index b2fd476f06..0fc4d1b8d1 100644 --- a/docs/_cpp_api/dir_cpp_include.html +++ b/docs/_cpp_api/dir_cpp_include.html @@ -10,7 +10,7 @@ - Directory include — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Directory include — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/dir_cpp_include_torch_tensorrt.html b/docs/_cpp_api/dir_cpp_include_torch_tensorrt.html index adfae6b652..aeae674d0e 100644 --- a/docs/_cpp_api/dir_cpp_include_torch_tensorrt.html +++ b/docs/_cpp_api/dir_cpp_include_torch_tensorrt.html @@ -10,7 +10,7 @@ - Directory torch_tensorrt — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Directory torch_tensorrt — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.html b/docs/_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.html index b5e17cc9f0..5ca232b20b 100644 --- a/docs/_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.html +++ b/docs/_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.html @@ -10,7 +10,7 @@ - Enum Level — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Enum Level — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.html b/docs/_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.html index 46efbbf20c..294db733f0 100644 --- a/docs/_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.html +++ b/docs/_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.html @@ -10,7 +10,7 @@ - Enum EngineCapability — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Enum EngineCapability — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/file_cpp_include_torch_tensorrt_logging.h.html b/docs/_cpp_api/file_cpp_include_torch_tensorrt_logging.h.html index 26994ee744..002b7b81ac 100644 --- a/docs/_cpp_api/file_cpp_include_torch_tensorrt_logging.h.html +++ b/docs/_cpp_api/file_cpp_include_torch_tensorrt_logging.h.html @@ -10,7 +10,7 @@ - File logging.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation + File logging.h — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/file_cpp_include_torch_tensorrt_macros.h.html b/docs/_cpp_api/file_cpp_include_torch_tensorrt_macros.h.html index ae2e3cad77..d675a45257 100644 --- a/docs/_cpp_api/file_cpp_include_torch_tensorrt_macros.h.html +++ b/docs/_cpp_api/file_cpp_include_torch_tensorrt_macros.h.html @@ -10,7 +10,7 @@ - File macros.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation + File macros.h — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.html b/docs/_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.html index ef1d5cd486..4c857b25a1 100644 --- a/docs/_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.html +++ b/docs/_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.html @@ -10,7 +10,7 @@ - File ptq.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation + File ptq.h — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.html b/docs/_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.html index 3a7dbe4cfa..1bcca64dd7 100644 --- a/docs/_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.html +++ b/docs/_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.html @@ -10,7 +10,7 @@ - File torch_tensorrt.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation + File torch_tensorrt.h — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.html index cf75405f74..19cbec2837 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::get_logging_prefix — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::logging::get_logging_prefix — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.html index ab72cb40b8..e45b666a00 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::get_reportable_log_level — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::logging::get_reportable_log_level — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.html index 060361232f..0f9d251cfc 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::get_is_colored_output_on — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::logging::get_is_colored_output_on — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.html index 6164b6828e..b154d22aa1 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::set_reportable_log_level — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::logging::set_reportable_log_level — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.html index dd6d140e6e..60f579ff66 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::log — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::logging::log — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.html index 15c8fd97ef..66171ecc9b 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::set_is_colored_output_on — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::logging::set_is_colored_output_on — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.html index f0d0f268d8..0dd4ac29da 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::logging::set_logging_prefix — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::logging::set_logging_prefix — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.html index bd2afc3fdd..a226db5cad 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.html @@ -10,7 +10,7 @@ - Template Function torch_tensorrt::ptq::make_int8_cache_calibrator — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Template Function torch_tensorrt::ptq::make_int8_cache_calibrator — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.html index 8dcdf7e780..ec7f2a9203 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.html @@ -10,7 +10,7 @@ - Template Function torch_tensorrt::ptq::make_int8_calibrator — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Template Function torch_tensorrt::ptq::make_int8_calibrator — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.html index 3b763f23e3..08597feb01 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::torchscript::check_method_operator_support — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::torchscript::check_method_operator_support — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.html index af1aa92340..4e2fdde13c 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::torchscript::compile — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::torchscript::compile — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.html index b5bdaacb80..bfe03233e2 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::torchscript::embed_engine_in_new_module — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::torchscript::embed_engine_in_new_module — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.html index 0c01a5133c..ebc880f7b4 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::torchscript::convert_method_to_trt_engine — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::torchscript::convert_method_to_trt_engine — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.html index fd1c397c98..83277c0813 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::get_build_info — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::get_build_info — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.html index 1fc6a140b4..21df901f52 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::set_device — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::set_device — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html b/docs/_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html index 1313a1cae4..fccbf44c67 100644 --- a/docs/_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html +++ b/docs/_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.html @@ -10,7 +10,7 @@ - Function torch_tensorrt::dump_build_info — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Function torch_tensorrt::dump_build_info — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/namespace_torch_tensorrt.html b/docs/_cpp_api/namespace_torch_tensorrt.html index 0a160319b4..698b41c524 100644 --- a/docs/_cpp_api/namespace_torch_tensorrt.html +++ b/docs/_cpp_api/namespace_torch_tensorrt.html @@ -10,7 +10,7 @@ - Namespace torch_tensorrt — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Namespace torch_tensorrt — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/namespace_torch_tensorrt__logging.html b/docs/_cpp_api/namespace_torch_tensorrt__logging.html index a9d7faf7b0..21d0632d2b 100644 --- a/docs/_cpp_api/namespace_torch_tensorrt__logging.html +++ b/docs/_cpp_api/namespace_torch_tensorrt__logging.html @@ -10,7 +10,7 @@ - Namespace torch_tensorrt::logging — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Namespace torch_tensorrt::logging — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/namespace_torch_tensorrt__ptq.html b/docs/_cpp_api/namespace_torch_tensorrt__ptq.html index 751ff237ff..029460073a 100644 --- a/docs/_cpp_api/namespace_torch_tensorrt__ptq.html +++ b/docs/_cpp_api/namespace_torch_tensorrt__ptq.html @@ -10,7 +10,7 @@ - Namespace torch_tensorrt::ptq — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Namespace torch_tensorrt::ptq — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/namespace_torch_tensorrt__torchscript.html b/docs/_cpp_api/namespace_torch_tensorrt__torchscript.html index 9b1540b94c..d5b6745481 100644 --- a/docs/_cpp_api/namespace_torch_tensorrt__torchscript.html +++ b/docs/_cpp_api/namespace_torch_tensorrt__torchscript.html @@ -10,7 +10,7 @@ - Namespace torch_tensorrt::torchscript — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Namespace torch_tensorrt::torchscript — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.html b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.html index c0eaedc8f1..2ab9c757d0 100644 --- a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.html +++ b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.html @@ -10,7 +10,7 @@ - Program Listing for File logging.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Program Listing for File logging.h — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.html b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.html index db77c27520..751efa791a 100644 --- a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.html +++ b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.html @@ -10,7 +10,7 @@ - Program Listing for File macros.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Program Listing for File macros.h — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.html b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.html index 0215df7549..ae17d3ca24 100644 --- a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.html +++ b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.html @@ -10,7 +10,7 @@ - Program Listing for File ptq.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Program Listing for File ptq.h — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.html b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.html index 1062cc16ed..7bd771ebaf 100644 --- a/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.html +++ b/docs/_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.html @@ -10,7 +10,7 @@ - Program Listing for File torch_tensorrt.h — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Program Listing for File torch_tensorrt.h — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/structtorch__tensorrt_1_1Device.html b/docs/_cpp_api/structtorch__tensorrt_1_1Device.html index 9c7609b1aa..d1173d5cb5 100644 --- a/docs/_cpp_api/structtorch__tensorrt_1_1Device.html +++ b/docs/_cpp_api/structtorch__tensorrt_1_1Device.html @@ -10,7 +10,7 @@ - Struct Device — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Struct Device — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/structtorch__tensorrt_1_1GraphInputs.html b/docs/_cpp_api/structtorch__tensorrt_1_1GraphInputs.html index f997950fda..bc6f7ab5d1 100644 --- a/docs/_cpp_api/structtorch__tensorrt_1_1GraphInputs.html +++ b/docs/_cpp_api/structtorch__tensorrt_1_1GraphInputs.html @@ -10,7 +10,7 @@ - Struct GraphInputs — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Struct GraphInputs — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/structtorch__tensorrt_1_1Input.html b/docs/_cpp_api/structtorch__tensorrt_1_1Input.html index 022270bbf0..c3058a8c41 100644 --- a/docs/_cpp_api/structtorch__tensorrt_1_1Input.html +++ b/docs/_cpp_api/structtorch__tensorrt_1_1Input.html @@ -10,7 +10,7 @@ - Struct Input — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Struct Input — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.html b/docs/_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.html index 007335249b..ec3a8e9864 100644 --- a/docs/_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.html +++ b/docs/_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.html @@ -10,7 +10,7 @@ - Struct CompileSpec — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Struct CompileSpec — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_cpp_api/torch_tensort_cpp.html b/docs/_cpp_api/torch_tensort_cpp.html index 00a5d2d8d6..578a191a92 100644 --- a/docs/_cpp_api/torch_tensort_cpp.html +++ b/docs/_cpp_api/torch_tensort_cpp.html @@ -10,7 +10,7 @@ - Torch-TensorRT C++ API — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Torch-TensorRT C++ API — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • @@ -464,7 +464,7 @@

    Class Hierarchy
  • diff --git a/docs/_cpp_api/unabridged_orphan.html b/docs/_cpp_api/unabridged_orphan.html index 78d659f647..8e9003f519 100644 --- a/docs/_cpp_api/unabridged_orphan.html +++ b/docs/_cpp_api/unabridged_orphan.html @@ -10,7 +10,7 @@ - Full API — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Full API — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_downloads/6a6052d9668b2cb8332d349d328e21c1/_rendered_examples_jupyter.zip b/docs/_downloads/6a6052d9668b2cb8332d349d328e21c1/_rendered_examples_jupyter.zip index 34927a4abd82acfb1dd9f2e4811efe98dc046cd3..339f7611aaf76e6abc4b5e5a0ae8e36390598f79 100644 GIT binary patch delta 224 zcmaE~g6+`?Hl6@)W)=|!5NPb($dkg!)Y!YZh_i_gM1K?Kl>yPLCdr~8I?JnAVh5Dh8szDQoe7Xe9#(?PoIy z%zxyu3rx=oE$0Nui^oR`g6NV0Nq!LBU%x{hM0?JEs|BJhPN}2->FIHLj3FR;iyosd zh-TJj^a9cT`iv1EdZj+2JBa?I&lm)v-3%B#LG(le#y}AL#DFmrL|Yp|rmV#SqLmCFw4co+ zF#nOqE-*bWw44(pFCHH)2%<|0B>6#ffBg=55bZhttrm#3IHi&Tq^HN}F@}KXEqaW; zAevd9(F;WT>oZ1x=#~16?jZV;K4TDwb~9k~1kn=>7z07{69dLj5N&M;k?%KTv - Overview: module code — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Overview: module code — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/_Device.html b/docs/_modules/torch_tensorrt/_Device.html index c138050a63..9cb3211cd7 100644 --- a/docs/_modules/torch_tensorrt/_Device.html +++ b/docs/_modules/torch_tensorrt/_Device.html @@ -9,7 +9,7 @@ - torch_tensorrt._Device — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt._Device — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/_Input.html b/docs/_modules/torch_tensorrt/_Input.html index 81ac55ff91..f0f59f2901 100644 --- a/docs/_modules/torch_tensorrt/_Input.html +++ b/docs/_modules/torch_tensorrt/_Input.html @@ -9,7 +9,7 @@ - torch_tensorrt._Input — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt._Input — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/_compile.html b/docs/_modules/torch_tensorrt/_compile.html index 635a35c891..eb02c1886d 100644 --- a/docs/_modules/torch_tensorrt/_compile.html +++ b/docs/_modules/torch_tensorrt/_compile.html @@ -9,7 +9,7 @@ - torch_tensorrt._compile — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt._compile — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/_enums.html b/docs/_modules/torch_tensorrt/_enums.html index 476a682a2b..15ac34ce4a 100644 --- a/docs/_modules/torch_tensorrt/_enums.html +++ b/docs/_modules/torch_tensorrt/_enums.html @@ -9,7 +9,7 @@ - torch_tensorrt._enums — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt._enums — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/dynamo/_compiler.html b/docs/_modules/torch_tensorrt/dynamo/_compiler.html index ed23a6f40b..7d2fd097f5 100644 --- a/docs/_modules/torch_tensorrt/dynamo/_compiler.html +++ b/docs/_modules/torch_tensorrt/dynamo/_compiler.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo._compiler — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.dynamo._compiler — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/dynamo/_exporter.html b/docs/_modules/torch_tensorrt/dynamo/_exporter.html index 12eb2ecd1d..f3d4462113 100644 --- a/docs/_modules/torch_tensorrt/dynamo/_exporter.html +++ b/docs/_modules/torch_tensorrt/dynamo/_exporter.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo._exporter — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.dynamo._exporter — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/dynamo/_refit.html b/docs/_modules/torch_tensorrt/dynamo/_refit.html index 3c8da6aa0c..e76c68fa67 100644 --- a/docs/_modules/torch_tensorrt/dynamo/_refit.html +++ b/docs/_modules/torch_tensorrt/dynamo/_refit.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo._refit — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.dynamo._refit — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/dynamo/_settings.html b/docs/_modules/torch_tensorrt/dynamo/_settings.html index 73e1e6c6bd..37fab4bc97 100644 --- a/docs/_modules/torch_tensorrt/dynamo/_settings.html +++ b/docs/_modules/torch_tensorrt/dynamo/_settings.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo._settings — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.dynamo._settings — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/dynamo/_tracer.html b/docs/_modules/torch_tensorrt/dynamo/_tracer.html index 86583fa3a4..0f0a03d1dc 100644 --- a/docs/_modules/torch_tensorrt/dynamo/_tracer.html +++ b/docs/_modules/torch_tensorrt/dynamo/_tracer.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo._tracer — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.dynamo._tracer — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/dynamo/runtime/_MutableTorchTensorRTModule.html b/docs/_modules/torch_tensorrt/dynamo/runtime/_MutableTorchTensorRTModule.html index aedbdef751..c08ee587b7 100644 --- a/docs/_modules/torch_tensorrt/dynamo/runtime/_MutableTorchTensorRTModule.html +++ b/docs/_modules/torch_tensorrt/dynamo/runtime/_MutableTorchTensorRTModule.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo.runtime._MutableTorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.dynamo.runtime._MutableTorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.html b/docs/_modules/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.html index 8ff32a67ab..bfe3827ee0 100644 --- a/docs/_modules/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.html +++ b/docs/_modules/torch_tensorrt/dynamo/runtime/_PythonTorchTensorRTModule.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo.runtime._PythonTorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.dynamo.runtime._PythonTorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.html b/docs/_modules/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.html index e74f349040..0678f3103b 100644 --- a/docs/_modules/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.html +++ b/docs/_modules/torch_tensorrt/dynamo/runtime/_TorchTensorRTModule.html @@ -9,7 +9,7 @@ - torch_tensorrt.dynamo.runtime._TorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.dynamo.runtime._TorchTensorRTModule — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/fx/fx2trt.html b/docs/_modules/torch_tensorrt/fx/fx2trt.html index 0beaf911c2..4ab21bbaae 100644 --- a/docs/_modules/torch_tensorrt/fx/fx2trt.html +++ b/docs/_modules/torch_tensorrt/fx/fx2trt.html @@ -9,7 +9,7 @@ - torch_tensorrt.fx.fx2trt — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.fx.fx2trt — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/fx/input_tensor_spec.html b/docs/_modules/torch_tensorrt/fx/input_tensor_spec.html index 35cdcecf84..c085ea8e07 100644 --- a/docs/_modules/torch_tensorrt/fx/input_tensor_spec.html +++ b/docs/_modules/torch_tensorrt/fx/input_tensor_spec.html @@ -9,7 +9,7 @@ - torch_tensorrt.fx.input_tensor_spec — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.fx.input_tensor_spec — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/fx/lower.html b/docs/_modules/torch_tensorrt/fx/lower.html index 998a1988b2..4db3a7f204 100644 --- a/docs/_modules/torch_tensorrt/fx/lower.html +++ b/docs/_modules/torch_tensorrt/fx/lower.html @@ -9,7 +9,7 @@ - torch_tensorrt.fx.lower — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.fx.lower — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/fx/trt_module.html b/docs/_modules/torch_tensorrt/fx/trt_module.html index 4e478c0e3f..4bfee52b9d 100644 --- a/docs/_modules/torch_tensorrt/fx/trt_module.html +++ b/docs/_modules/torch_tensorrt/fx/trt_module.html @@ -9,7 +9,7 @@ - torch_tensorrt.fx.trt_module — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.fx.trt_module — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/logging.html b/docs/_modules/torch_tensorrt/logging.html index cba5eab6be..596164f828 100644 --- a/docs/_modules/torch_tensorrt/logging.html +++ b/docs/_modules/torch_tensorrt/logging.html @@ -9,7 +9,7 @@ - torch_tensorrt.logging — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.logging — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/runtime/_multi_device_safe_mode.html b/docs/_modules/torch_tensorrt/runtime/_multi_device_safe_mode.html index d5cd752294..fdac722346 100644 --- a/docs/_modules/torch_tensorrt/runtime/_multi_device_safe_mode.html +++ b/docs/_modules/torch_tensorrt/runtime/_multi_device_safe_mode.html @@ -9,7 +9,7 @@ - torch_tensorrt.runtime._multi_device_safe_mode — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.runtime._multi_device_safe_mode — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/ts/_compile_spec.html b/docs/_modules/torch_tensorrt/ts/_compile_spec.html index 3a4a6994ad..2fae159cd6 100644 --- a/docs/_modules/torch_tensorrt/ts/_compile_spec.html +++ b/docs/_modules/torch_tensorrt/ts/_compile_spec.html @@ -9,7 +9,7 @@ - torch_tensorrt.ts._compile_spec — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.ts._compile_spec — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/ts/_compiler.html b/docs/_modules/torch_tensorrt/ts/_compiler.html index 8b84d0f15c..fbfd21f740 100644 --- a/docs/_modules/torch_tensorrt/ts/_compiler.html +++ b/docs/_modules/torch_tensorrt/ts/_compiler.html @@ -9,7 +9,7 @@ - torch_tensorrt.ts._compiler — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.ts._compiler — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_modules/torch_tensorrt/ts/ptq.html b/docs/_modules/torch_tensorrt/ts/ptq.html index a3bd2cd53f..8cc48525b0 100644 --- a/docs/_modules/torch_tensorrt/ts/ptq.html +++ b/docs/_modules/torch_tensorrt/ts/ptq.html @@ -9,7 +9,7 @@ - torch_tensorrt.ts.ptq — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.ts.ptq — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/_sources/index.rst.txt b/docs/_sources/index.rst.txt index d1a91beabc..757acc2011 100644 --- a/docs/_sources/index.rst.txt +++ b/docs/_sources/index.rst.txt @@ -44,13 +44,12 @@ User Guide :hidden: user_guide/torch_tensorrt_explained - user_guide/getting_started user_guide/dynamic_shapes user_guide/saving_models user_guide/runtime user_guide/using_dla tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage - tutorials/_rendered_examples/dynamo/vgg16_fp8_ptq + tutorials/_rendered_examples/dynamo/vgg16_ptq tutorials/_rendered_examples/dynamo/engine_caching_example tutorials/_rendered_examples/dynamo/refit_engine_example @@ -113,13 +112,11 @@ Tutorials tutorials/notebooks tutorials/_rendered_examples/dynamo/torch_compile_resnet_example tutorials/_rendered_examples/dynamo/torch_compile_transformers_example - tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage tutorials/_rendered_examples/dynamo/torch_compile_stable_diffusion tutorials/_rendered_examples/dynamo/torch_export_cudagraphs tutorials/_rendered_examples/dynamo/custom_kernel_plugins tutorials/_rendered_examples/distributed_inference/data_parallel_gpt2 tutorials/_rendered_examples/distributed_inference/data_parallel_stable_diffusion - tutorials/_rendered_examples/dynamo/vgg16_fp8_ptq tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example Python API Documentation diff --git a/docs/_static/documentation_options.js b/docs/_static/documentation_options.js index 1540ac6f48..6b61b71ddb 100644 --- a/docs/_static/documentation_options.js +++ b/docs/_static/documentation_options.js @@ -1,6 +1,6 @@ var DOCUMENTATION_OPTIONS = { URL_ROOT: document.getElementById("documentation_options").getAttribute('data-url_root'), - VERSION: 'v2.5.0.dev0+8759736', + VERSION: 'v2.5.0.dev0+a4a9419', LANGUAGE: 'en', COLLAPSE_INDEX: false, BUILDER: 'html', diff --git a/docs/cli/torchtrtc.html b/docs/cli/torchtrtc.html index 5309cb19d6..e9ad12f49b 100644 --- a/docs/cli/torchtrtc.html +++ b/docs/cli/torchtrtc.html @@ -10,7 +10,7 @@ - torchtrtc — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torchtrtc — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/conversion.html b/docs/contributors/conversion.html index 4a53b164b3..aa60dd09fb 100644 --- a/docs/contributors/conversion.html +++ b/docs/contributors/conversion.html @@ -10,7 +10,7 @@ - Conversion Phase — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Conversion Phase — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/dynamo_converters.html b/docs/contributors/dynamo_converters.html index 02205bdac1..bbd1ce46eb 100644 --- a/docs/contributors/dynamo_converters.html +++ b/docs/contributors/dynamo_converters.html @@ -10,7 +10,7 @@ - Writing Dynamo Converters — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Writing Dynamo Converters — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/lowering.html b/docs/contributors/lowering.html index 4138e50a9d..a5bd7f4b60 100644 --- a/docs/contributors/lowering.html +++ b/docs/contributors/lowering.html @@ -10,7 +10,7 @@ - Lowering Phase — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Lowering Phase — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/partitioning.html b/docs/contributors/partitioning.html index cba9deb117..a23c3419ad 100644 --- a/docs/contributors/partitioning.html +++ b/docs/contributors/partitioning.html @@ -10,7 +10,7 @@ - Partitioning Phase — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Partitioning Phase — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/phases.html b/docs/contributors/phases.html index 7c1f991d95..9ab6e44afe 100644 --- a/docs/contributors/phases.html +++ b/docs/contributors/phases.html @@ -10,7 +10,7 @@ - Compiler Phases — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Compiler Phases — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/runtime.html b/docs/contributors/runtime.html index 84f0a1fa20..3e500669a5 100644 --- a/docs/contributors/runtime.html +++ b/docs/contributors/runtime.html @@ -10,7 +10,7 @@ - Runtime Phase — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Runtime Phase — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/system_overview.html b/docs/contributors/system_overview.html index c2fa91573b..dd53036784 100644 --- a/docs/contributors/system_overview.html +++ b/docs/contributors/system_overview.html @@ -10,7 +10,7 @@ - System Overview — Torch-TensorRT v2.5.0.dev0+8759736 documentation + System Overview — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/ts_converters.html b/docs/contributors/ts_converters.html index cd1871800d..60b38cd7d1 100644 --- a/docs/contributors/ts_converters.html +++ b/docs/contributors/ts_converters.html @@ -10,7 +10,7 @@ - Writing TorchScript Converters — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Writing TorchScript Converters — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/useful_links.html b/docs/contributors/useful_links.html index 874f4968e7..1bcdb512e3 100644 --- a/docs/contributors/useful_links.html +++ b/docs/contributors/useful_links.html @@ -10,7 +10,7 @@ - Useful Links for Torch-TensorRT Development — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Useful Links for Torch-TensorRT Development — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/contributors/writing_dynamo_aten_lowering_passes.html b/docs/contributors/writing_dynamo_aten_lowering_passes.html index da2a3bb37d..f57921e28a 100644 --- a/docs/contributors/writing_dynamo_aten_lowering_passes.html +++ b/docs/contributors/writing_dynamo_aten_lowering_passes.html @@ -10,7 +10,7 @@ - Writing Dynamo ATen Lowering Passes — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Writing Dynamo ATen Lowering Passes — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/dynamo/dynamo_export.html b/docs/dynamo/dynamo_export.html index c1c66ebaa0..a6cf0086c7 100644 --- a/docs/dynamo/dynamo_export.html +++ b/docs/dynamo/dynamo_export.html @@ -10,7 +10,7 @@ - Compiling Exported Programs with Torch-TensorRT — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Compiling Exported Programs with Torch-TensorRT — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/dynamo/torch_compile.html b/docs/dynamo/torch_compile.html index 9715f4e330..f378c18a1d 100644 --- a/docs/dynamo/torch_compile.html +++ b/docs/dynamo/torch_compile.html @@ -10,7 +10,7 @@ - TensorRT Backend for torch.compile — Torch-TensorRT v2.5.0.dev0+8759736 documentation + TensorRT Backend for torch.compile — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/fx/getting_started_with_fx_path.html b/docs/fx/getting_started_with_fx_path.html index 7dd4a4c566..2d5138e628 100644 --- a/docs/fx/getting_started_with_fx_path.html +++ b/docs/fx/getting_started_with_fx_path.html @@ -10,7 +10,7 @@ - Torch-TensorRT (FX Frontend) User Guide — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Torch-TensorRT (FX Frontend) User Guide — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/genindex.html b/docs/genindex.html index 031c2c262e..046b5950c9 100644 --- a/docs/genindex.html +++ b/docs/genindex.html @@ -9,7 +9,7 @@ - Index — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Index — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/getting_started/installation.html b/docs/getting_started/installation.html index 8bd9697cd5..cccc499826 100644 --- a/docs/getting_started/installation.html +++ b/docs/getting_started/installation.html @@ -10,7 +10,7 @@ - Installation — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Installation — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/getting_started/quick_start.html b/docs/getting_started/quick_start.html index 779abb5d8f..f9c498f6bc 100644 --- a/docs/getting_started/quick_start.html +++ b/docs/getting_started/quick_start.html @@ -10,7 +10,7 @@ - Quick Start — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Quick Start — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/index.html b/docs/index.html index f4b8d2cadd..5c7f1983df 100644 --- a/docs/index.html +++ b/docs/index.html @@ -10,7 +10,7 @@ - Torch-TensorRT — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Torch-TensorRT — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -274,7 +274,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -315,6 +315,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -342,7 +343,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/indices/supported_ops.html b/docs/indices/supported_ops.html index a308008a58..ccbeebb6af 100644 --- a/docs/indices/supported_ops.html +++ b/docs/indices/supported_ops.html @@ -10,7 +10,7 @@ - Operators Supported — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Operators Supported — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -274,7 +274,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -315,6 +315,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -342,7 +343,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/objects.inv b/docs/objects.inv index a7252e6dc451664cfe67cfa44c70f94d0c92b984..4d29df817b9c4abe6ddc809d9f143e0e6923b1f9 100644 GIT binary patch delta 20 bcmezQgYoAN#tDAxi6)7bCWe+9Lse=3W!?xq delta 20 bcmezQgYoAN#tDAx7Urgw=Ei0lLse=3V!sFN diff --git a/docs/py-modindex.html b/docs/py-modindex.html index 8904d613d2..8e212e6470 100644 --- a/docs/py-modindex.html +++ b/docs/py-modindex.html @@ -9,7 +9,7 @@ - Python Module Index — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Python Module Index — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/py_api/dynamo.html b/docs/py_api/dynamo.html index feadef8950..833d4d2462 100644 --- a/docs/py_api/dynamo.html +++ b/docs/py_api/dynamo.html @@ -10,7 +10,7 @@ - torch_tensorrt.dynamo — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.dynamo — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/py_api/fx.html b/docs/py_api/fx.html index db34bd08d6..0d3fbeb8d0 100644 --- a/docs/py_api/fx.html +++ b/docs/py_api/fx.html @@ -10,7 +10,7 @@ - torch_tensorrt.fx — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.fx — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/py_api/logging.html b/docs/py_api/logging.html index dbcf27f9b6..783fd8871d 100644 --- a/docs/py_api/logging.html +++ b/docs/py_api/logging.html @@ -10,7 +10,7 @@ - torch_tensorrt.logging — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.logging — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/py_api/ptq.html b/docs/py_api/ptq.html index 9d60615179..2353ec0ee7 100644 --- a/docs/py_api/ptq.html +++ b/docs/py_api/ptq.html @@ -10,7 +10,7 @@ - torch_tensorrt.ts.ptq — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.ts.ptq — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/py_api/runtime.html b/docs/py_api/runtime.html index 562414e95c..c17c6893a2 100644 --- a/docs/py_api/runtime.html +++ b/docs/py_api/runtime.html @@ -10,7 +10,7 @@ - torch_tensorrt.runtime — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.runtime — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/py_api/torch_tensorrt.html b/docs/py_api/torch_tensorrt.html index f798bc6729..57cc479e11 100644 --- a/docs/py_api/torch_tensorrt.html +++ b/docs/py_api/torch_tensorrt.html @@ -10,7 +10,7 @@ - torch_tensorrt — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/py_api/ts.html b/docs/py_api/ts.html index 58284e02cd..358eed4dc4 100644 --- a/docs/py_api/ts.html +++ b/docs/py_api/ts.html @@ -10,7 +10,7 @@ - torch_tensorrt.ts — Torch-TensorRT v2.5.0.dev0+8759736 documentation + torch_tensorrt.ts — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • @@ -681,7 +681,7 @@

    Functions
    -torch_tensorrt.ts.TensorRTCompileSpec(inputs: Optional[List[torch.Tensor | Input]] = None, input_signature: Optional[Any] = None, device: torch.device | Device = Device(type=DeviceType.GPU, gpu_id=0), disable_tf32: bool = False, sparse_weights: bool = False, enabled_precisions: Optional[Set[Union[dtype, dtype]]] = None, refit: bool = False, debug: bool = False, capability: EngineCapability = EngineCapability.STANDARD, num_avg_timing_iters: int = 1, workspace_size: int = 0, dla_sram_size: int = 1048576, dla_local_dram_size: int = 1073741824, dla_global_dram_size: int = 536870912, truncate_long_and_double: bool = False, calibrator: object = None, allow_shape_tensors: bool = False) <torch.ScriptClass object at 0x7f14103e9130>[source]
    +torch_tensorrt.ts.TensorRTCompileSpec(inputs: Optional[List[torch.Tensor | Input]] = None, input_signature: Optional[Any] = None, device: torch.device | Device = Device(type=DeviceType.GPU, gpu_id=0), disable_tf32: bool = False, sparse_weights: bool = False, enabled_precisions: Optional[Set[Union[dtype, dtype]]] = None, refit: bool = False, debug: bool = False, capability: EngineCapability = EngineCapability.STANDARD, num_avg_timing_iters: int = 1, workspace_size: int = 0, dla_sram_size: int = 1048576, dla_local_dram_size: int = 1073741824, dla_global_dram_size: int = 536870912, truncate_long_and_double: bool = False, calibrator: object = None, allow_shape_tensors: bool = False) <torch.ScriptClass object at 0x7fad0f3e7bf0>[source]

    Utility to create a formatted spec dictionary for using the PyTorch TensorRT backend

    Keyword Arguments
    diff --git a/docs/search.html b/docs/search.html index c8894e97a8..ce3a3704bf 100644 --- a/docs/search.html +++ b/docs/search.html @@ -9,7 +9,7 @@ - Search — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Search — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -272,7 +272,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -313,6 +313,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -340,7 +341,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/searchindex.js b/docs/searchindex.js index b9f82f35a6..f1b4178aa4 100644 --- a/docs/searchindex.js +++ b/docs/searchindex.js @@ -1 +1 @@ -Search.setIndex({"docnames": ["_cpp_api/classtorch__tensorrt_1_1DataType", "_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType", "_cpp_api/classtorch__tensorrt_1_1TensorFormat", "_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator", "_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator", "_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502", "_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268", "_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e", "_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827", "_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b", "_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da", "_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59", "_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883", "_cpp_api/dir_cpp", "_cpp_api/dir_cpp_include", "_cpp_api/dir_cpp_include_torch_tensorrt", "_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558", "_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb", "_cpp_api/file_cpp_include_torch_tensorrt_logging.h", "_cpp_api/file_cpp_include_torch_tensorrt_macros.h", "_cpp_api/file_cpp_include_torch_tensorrt_ptq.h", "_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc", "_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c", "_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2", "_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528", "_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384", "_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1", "_cpp_api/namespace_torch_tensorrt", "_cpp_api/namespace_torch_tensorrt__logging", "_cpp_api/namespace_torch_tensorrt__ptq", "_cpp_api/namespace_torch_tensorrt__torchscript", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h", "_cpp_api/structtorch__tensorrt_1_1Device", "_cpp_api/structtorch__tensorrt_1_1GraphInputs", "_cpp_api/structtorch__tensorrt_1_1Input", "_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec", "_cpp_api/torch_tensort_cpp", "_cpp_api/unabridged_orphan", "cli/torchtrtc", "contributors/conversion", "contributors/dynamo_converters", "contributors/lowering", "contributors/partitioning", "contributors/phases", "contributors/runtime", "contributors/system_overview", "contributors/ts_converters", "contributors/useful_links", "contributors/writing_dynamo_aten_lowering_passes", "dynamo/dynamo_export", "dynamo/torch_compile", "fx/getting_started_with_fx_path", "getting_started/installation", "getting_started/quick_start", "index", "indices/supported_ops", "py_api/dynamo", "py_api/fx", "py_api/logging", "py_api/ptq", "py_api/runtime", "py_api/torch_tensorrt", "py_api/ts", "sg_execution_times", "src/pytorch-sphinx-theme/docs/changelog", "src/pytorch-sphinx-theme/docs/configuring", "src/pytorch-sphinx-theme/docs/demo/api", "src/pytorch-sphinx-theme/docs/demo/demo", "src/pytorch-sphinx-theme/docs/demo/lists_tables", "src/pytorch-sphinx-theme/docs/demo/long", "src/pytorch-sphinx-theme/docs/demo/structure", "src/pytorch-sphinx-theme/docs/index", "src/pytorch-sphinx-theme/docs/installing", "ts/creating_torchscript_module_in_python", "ts/getting_started_with_cpp_api", "ts/getting_started_with_python_api", "ts/ptq", "ts/torchscript_frontend_from_pytorch", "tutorials/_rendered_examples/dynamo/custom_kernel_plugins", "tutorials/_rendered_examples/dynamo/engine_caching_bert_example", "tutorials/_rendered_examples/dynamo/engine_caching_example", "tutorials/_rendered_examples/dynamo/index", "tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example", "tutorials/_rendered_examples/dynamo/refit_engine_example", "tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage", "tutorials/_rendered_examples/dynamo/torch_compile_resnet_example", "tutorials/_rendered_examples/dynamo/torch_compile_stable_diffusion", "tutorials/_rendered_examples/dynamo/torch_compile_transformers_example", "tutorials/_rendered_examples/dynamo/torch_export_cudagraphs", "tutorials/_rendered_examples/dynamo/torch_export_gpt2", "tutorials/_rendered_examples/dynamo/torch_export_llama2", "tutorials/_rendered_examples/dynamo/vgg16_ptq", "tutorials/_rendered_examples/index", "tutorials/notebooks", "tutorials/serving_torch_tensorrt_with_triton", "user_guide/dynamic_shapes", "user_guide/runtime", "user_guide/saving_models", "user_guide/torch_tensorrt_explained", "user_guide/using_dla"], "filenames": ["_cpp_api/classtorch__tensorrt_1_1DataType.rst", "_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.rst", "_cpp_api/classtorch__tensorrt_1_1TensorFormat.rst", "_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.rst", "_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.rst", "_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.rst", "_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.rst", "_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.rst", "_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.rst", "_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.rst", "_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.rst", "_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.rst", "_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.rst", "_cpp_api/dir_cpp.rst", "_cpp_api/dir_cpp_include.rst", "_cpp_api/dir_cpp_include_torch_tensorrt.rst", "_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.rst", "_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.rst", "_cpp_api/file_cpp_include_torch_tensorrt_logging.h.rst", "_cpp_api/file_cpp_include_torch_tensorrt_macros.h.rst", "_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.rst", "_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.rst", "_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.rst", "_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.rst", "_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.rst", "_cpp_api/namespace_torch_tensorrt.rst", "_cpp_api/namespace_torch_tensorrt__logging.rst", "_cpp_api/namespace_torch_tensorrt__ptq.rst", "_cpp_api/namespace_torch_tensorrt__torchscript.rst", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.rst", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.rst", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.rst", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.rst", "_cpp_api/structtorch__tensorrt_1_1Device.rst", "_cpp_api/structtorch__tensorrt_1_1GraphInputs.rst", "_cpp_api/structtorch__tensorrt_1_1Input.rst", "_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.rst", "_cpp_api/torch_tensort_cpp.rst", "_cpp_api/unabridged_orphan.rst", "cli/torchtrtc.rst", "contributors/conversion.rst", "contributors/dynamo_converters.rst", "contributors/lowering.rst", "contributors/partitioning.rst", "contributors/phases.rst", "contributors/runtime.rst", "contributors/system_overview.rst", "contributors/ts_converters.rst", "contributors/useful_links.rst", "contributors/writing_dynamo_aten_lowering_passes.rst", "dynamo/dynamo_export.rst", "dynamo/torch_compile.rst", "fx/getting_started_with_fx_path.rst", "getting_started/installation.rst", "getting_started/quick_start.rst", "index.rst", "indices/supported_ops.rst", "py_api/dynamo.rst", "py_api/fx.rst", "py_api/logging.rst", "py_api/ptq.rst", "py_api/runtime.rst", "py_api/torch_tensorrt.rst", "py_api/ts.rst", "sg_execution_times.rst", "src/pytorch-sphinx-theme/docs/changelog.rst", "src/pytorch-sphinx-theme/docs/configuring.rst", "src/pytorch-sphinx-theme/docs/demo/api.rst", "src/pytorch-sphinx-theme/docs/demo/demo.rst", "src/pytorch-sphinx-theme/docs/demo/lists_tables.rst", "src/pytorch-sphinx-theme/docs/demo/long.rst", "src/pytorch-sphinx-theme/docs/demo/structure.rst", "src/pytorch-sphinx-theme/docs/index.rst", "src/pytorch-sphinx-theme/docs/installing.rst", "ts/creating_torchscript_module_in_python.rst", "ts/getting_started_with_cpp_api.rst", "ts/getting_started_with_python_api.rst", "ts/ptq.rst", "ts/torchscript_frontend_from_pytorch.rst", "tutorials/_rendered_examples/dynamo/custom_kernel_plugins.rst", "tutorials/_rendered_examples/dynamo/engine_caching_bert_example.rst", "tutorials/_rendered_examples/dynamo/engine_caching_example.rst", "tutorials/_rendered_examples/dynamo/index.rst", "tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example.rst", "tutorials/_rendered_examples/dynamo/refit_engine_example.rst", "tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage.rst", "tutorials/_rendered_examples/dynamo/torch_compile_resnet_example.rst", "tutorials/_rendered_examples/dynamo/torch_compile_stable_diffusion.rst", "tutorials/_rendered_examples/dynamo/torch_compile_transformers_example.rst", "tutorials/_rendered_examples/dynamo/torch_export_cudagraphs.rst", "tutorials/_rendered_examples/dynamo/torch_export_gpt2.rst", "tutorials/_rendered_examples/dynamo/torch_export_llama2.rst", "tutorials/_rendered_examples/dynamo/vgg16_ptq.rst", "tutorials/_rendered_examples/index.rst", "tutorials/notebooks.rst", "tutorials/serving_torch_tensorrt_with_triton.rst", "user_guide/dynamic_shapes.rst", "user_guide/runtime.rst", "user_guide/saving_models.rst", "user_guide/torch_tensorrt_explained.rst", "user_guide/using_dla.rst"], "titles": ["Class DataType", "Class Device::DeviceType", "Class TensorFormat", "Template Class Int8CacheCalibrator", "Template Class Int8Calibrator", "Define STR", "Define TORCH_TENSORRT_PATCH_VERSION", "Define TORCH_TENSORRT_MAJOR_VERSION", "Define TORCH_TENSORRT_MINOR_VERSION", "Define TORCHTRT_API", "Define XSTR", "Define TORCHTRT_HIDDEN", "Define TORCH_TENSORRT_VERSION", "Directory cpp", "Directory include", "Directory torch_tensorrt", "Enum Level", "Enum EngineCapability", "File logging.h", "File macros.h", "File ptq.h", "File torch_tensorrt.h", "Function torch_tensorrt::logging::get_logging_prefix", "Function torch_tensorrt::logging::get_reportable_log_level", "Function torch_tensorrt::logging::get_is_colored_output_on", "Function torch_tensorrt::logging::set_reportable_log_level", "Function torch_tensorrt::logging::log", "Function torch_tensorrt::logging::set_is_colored_output_on", "Function torch_tensorrt::logging::set_logging_prefix", "Template Function torch_tensorrt::ptq::make_int8_cache_calibrator", "Template Function torch_tensorrt::ptq::make_int8_calibrator", "Function torch_tensorrt::torchscript::check_method_operator_support", "Function torch_tensorrt::torchscript::compile", "Function torch_tensorrt::torchscript::embed_engine_in_new_module", "Function torch_tensorrt::torchscript::convert_method_to_trt_engine", "Function torch_tensorrt::get_build_info", "Function torch_tensorrt::set_device", "Function torch_tensorrt::dump_build_info", "Namespace torch_tensorrt", "Namespace torch_tensorrt::logging", "Namespace torch_tensorrt::ptq", "Namespace torch_tensorrt::torchscript", "Program Listing for File logging.h", "Program Listing for File macros.h", "Program Listing for File ptq.h", "Program Listing for File torch_tensorrt.h", "Struct Device", "Struct GraphInputs", "Struct Input", "Struct CompileSpec", "Torch-TensorRT C++ API", "Full API", "torchtrtc", "Conversion Phase", "Writing Dynamo Converters", "Lowering Phase", "Partitioning Phase", "Compiler Phases", "Runtime Phase", "System Overview", "Writing TorchScript Converters", "Useful Links for Torch-TensorRT Development", "Writing Dynamo ATen Lowering Passes", "Compiling Exported Programs with Torch-TensorRT", "TensorRT Backend for torch.compile", "Torch-TensorRT (FX Frontend) User Guide", "Installation", "Quick Start", "Torch-TensorRT", "Operators Supported", "torch_tensorrt.dynamo", "torch_tensorrt.fx", "torch_tensorrt.logging", "torch_tensorrt.ts.ptq", "torch_tensorrt.runtime", "torch_tensorrt", "torch_tensorrt.ts", "Computation times", "Changelog", "Configuration", "5. :mod:`test_py_module`", "3. Paragraph Level Markup", "4. Lists & Tables", "1. Long Sticky Nav", "1. Structural Elements", "<no title>", "Installation", "Creating a TorchScript Module", "Using Torch-TensorRT in C++", "Using Torch-TensorRT in Python", "Post Training Quantization (PTQ)", "Using Torch-TensorRT TorchScript Frontend Directly From PyTorch", "Using Custom Kernels within TensorRT Engines with Torch-TensorRT", "Engine Caching (BERT)", "Engine Caching", "Dynamo / torch.compile", "Mutable Torch TensorRT Module", "Refitting Torch-TensorRT Programs with New Weights", "Torch Compile Advanced Usage", "Compiling ResNet using the Torch-TensorRT torch.compile Backend", "Torch Compile Stable Diffusion", "Compiling a Transformer using torch.compile and TensorRT", "Torch Export with Cudagraphs", "Compiling GPT2 using the Torch-TensorRT with dynamo backend", "Compiling Llama2 using the Torch-TensorRT with dynamo backend", "Deploy Quantized Models using Torch-TensorRT", "Torch-TensorRT Tutorials", "Example notebooks", "Serving a Torch-TensorRT model with Triton", "Dynamic shapes with Torch-TensorRT", "Deploying Torch-TensorRT Programs", "Saving models compiled with Torch-TensorRT", "Torch-TensorRT Explained", "DLA"], "terms": {"defin": [0, 1, 2, 3, 4, 33, 43, 46, 47, 48, 49, 51, 52, 54, 65, 67, 74, 75, 79, 87, 88, 89, 90, 92, 94, 98, 101, 102, 103, 104, 107], "file": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 46, 47, 48, 49, 52, 54, 56, 58, 59, 64, 65, 66, 67, 70, 71, 73, 75, 76, 77, 79, 80, 82, 86, 88, 90, 108, 109, 111], "torch_tensorrt": [0, 1, 2, 14, 16, 17, 42, 43, 44, 46, 47, 48, 49, 50, 51, 52, 54, 56, 62, 63, 64, 65, 67, 68, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 100, 103, 104, 105, 106, 107, 108, 109, 110, 111, 113], "h": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 15, 46, 47, 48, 49, 50, 51, 52, 55, 67, 75, 88, 90], "support": [0, 1, 2, 27, 31, 46, 48, 49, 52, 54, 56, 61, 63, 65, 67, 68, 71, 74, 75, 76, 79, 80, 87, 88, 89, 92, 97, 99, 101, 103, 104, 105, 108, 112, 113], "data": [0, 2, 3, 4, 29, 30, 44, 46, 48, 49, 52, 53, 56, 57, 59, 60, 64, 65, 69, 70, 71, 73, 75, 76, 81, 85, 89, 90, 92, 94, 105, 107], "type": [0, 1, 2, 30, 49, 50, 52, 53, 56, 58, 60, 62, 63, 64, 65, 70, 71, 73, 74, 75, 76, 81, 88, 89, 90, 92, 94, 105, 107, 111], "can": [0, 1, 4, 29, 30, 34, 46, 47, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 66, 67, 70, 73, 74, 75, 76, 79, 81, 87, 88, 89, 90, 91, 92, 94, 95, 96, 97, 98, 101, 102, 105, 106, 107, 108, 109, 110, 111, 112], "us": [0, 1, 2, 3, 4, 29, 30, 32, 34, 36, 43, 44, 45, 46, 48, 49, 52, 53, 54, 56, 58, 59, 60, 62, 63, 64, 65, 66, 67, 68, 70, 71, 73, 74, 75, 76, 77, 79, 80, 81, 82, 87, 90, 94, 95, 96, 97, 106, 108, 110, 111, 112, 113], "tensorrt": [0, 1, 3, 4, 29, 30, 31, 32, 33, 34, 37, 44, 45, 46, 48, 49, 52, 53, 54, 55, 56, 57, 59, 60, 62, 67, 70, 71, 73, 74, 75, 76, 77, 87, 90, 94, 95, 98, 100, 102], "engin": [0, 1, 17, 32, 33, 34, 45, 46, 48, 49, 52, 53, 56, 57, 59, 62, 63, 64, 70, 71, 74, 75, 76, 79, 88, 89, 90, 91, 95, 97, 99, 101, 106, 109, 110, 112, 113], "thi": [0, 1, 2, 29, 30, 42, 43, 44, 45, 46, 47, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 66, 67, 70, 71, 74, 75, 76, 79, 80, 81, 83, 84, 87, 88, 90, 91, 92, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112], "compat": [0, 1, 46, 55, 58, 64, 65, 70, 74, 75, 76, 112], "c10": [0, 1, 45, 46, 48, 49, 88, 90], "check": [0, 1, 31, 46, 52, 55, 60, 65, 70, 74, 76, 88, 92, 96, 97, 108, 110], "trt": [0, 1, 3, 4, 46, 48, 53, 55, 58, 60, 62, 64, 65, 67, 69, 70, 74, 75, 88, 92, 101, 103, 104, 109, 110, 111], "so": [0, 44, 52, 53, 54, 55, 58, 59, 60, 62, 64, 65, 66, 71, 74, 75, 80, 81, 82, 88, 90, 92, 94, 98, 99, 101, 103, 104, 109], "should": [0, 3, 4, 29, 45, 49, 52, 53, 54, 55, 56, 57, 59, 60, 63, 64, 65, 70, 74, 75, 76, 79, 81, 84, 90, 92, 93, 94, 97, 102, 108], "reason": [0, 65, 87, 92, 94, 112], "you": [0, 1, 2, 29, 30, 46, 48, 49, 52, 53, 54, 55, 56, 58, 59, 60, 63, 65, 66, 67, 70, 74, 75, 76, 79, 81, 82, 83, 87, 88, 89, 90, 91, 92, 94, 95, 96, 97, 105, 106, 107, 108, 109, 110, 111, 112], "need": [0, 1, 2, 25, 29, 43, 46, 53, 54, 55, 60, 65, 66, 70, 71, 74, 75, 81, 88, 89, 90, 92, 93, 94, 96, 97, 107, 108, 109, 110], "explictli": 0, "public": [0, 1, 2, 3, 4, 44, 45, 46, 47, 48, 49, 82, 90], "enum": [0, 1, 2, 42, 45, 46, 70, 76, 90], "valu": [0, 1, 2, 16, 17, 45, 46, 48, 53, 56, 58, 60, 63, 69, 70, 73, 75, 79, 88, 96, 98, 99, 101, 107], "underli": [0, 1, 2, 46, 60], "In": [0, 1, 2, 46, 53, 54, 56, 57, 58, 59, 60, 64, 65, 66, 74, 75, 81, 82, 84, 89, 90, 92, 95, 96, 106, 107, 108, 109, 110, 111], "case": [0, 1, 2, 46, 49, 53, 54, 56, 58, 60, 62, 64, 65, 66, 74, 75, 90, 92, 96, 97, 109, 110], "itself": [0, 1, 2, 46, 52, 55, 91, 108], "interfac": [0, 1, 2, 46, 58, 59, 60, 64, 68, 90], "vs": [0, 1, 2, 46, 55, 66, 70, 75, 76, 91], "normal": [0, 1, 2, 46, 65, 81, 87, 88, 90, 96, 97, 102, 105, 108, 113], "instatin": [0, 1, 2, 46], "ex": [0, 1, 2, 33, 46, 76, 82, 84], "kfloat": [0, 45, 49], "enumer": [0, 1, 2, 16, 17, 46], "klong": [0, 45], "int64": [0, 75, 76], "kdoubl": [0, 45], "fp64": [0, 75], "fp32": [0, 48, 49, 52, 64, 65, 70, 75, 76, 90, 107, 108], "khalf": [0, 45, 88], "fp16": [0, 48, 49, 52, 64, 65, 71, 75, 88, 89, 96, 100, 113], "kchar": [0, 45], "int8": [0, 44, 48, 49, 52, 64, 70, 75, 76, 90, 105, 113], "kint": [0, 45], "int": [0, 3, 4, 36, 44, 45, 49, 52, 54, 56, 63, 64, 69, 70, 71, 75, 76, 79, 88, 92, 105], "kbool": [0, 45], "bool": [0, 1, 2, 3, 4, 24, 27, 30, 31, 42, 44, 45, 46, 49, 55, 60, 64, 69, 70, 71, 73, 74, 75, 76, 79, 88, 90, 92], "kunknown": [0, 2, 45], "sentinel": [0, 2, 75], "function": [0, 1, 2, 3, 4, 46, 48, 49, 54, 55, 56, 58, 60, 62, 64, 65, 66, 87, 88, 90, 91, 92, 97, 98, 101, 102, 103, 104, 107, 108, 109, 110, 112, 113], "default": [0, 1, 2, 3, 4, 16, 29, 30, 33, 43, 45, 46, 48, 49, 52, 54, 56, 62, 64, 65, 66, 70, 71, 74, 75, 76, 79, 80, 81, 88, 89, 90, 91, 92, 94, 105, 109, 110, 111, 112], "construct": [0, 1, 2, 3, 4, 46, 48, 49, 53, 54, 55, 57, 59, 60, 65, 73, 74, 75, 81, 82, 88, 90, 92, 94, 109], "new": [0, 1, 2, 3, 4, 32, 33, 46, 48, 49, 56, 58, 59, 60, 62, 64, 65, 67, 70, 76, 81, 88, 94, 95, 96, 99, 101, 102, 106, 108, 110], "object": [0, 1, 2, 3, 4, 46, 48, 49, 52, 58, 60, 62, 63, 64, 70, 74, 75, 76, 90, 91, 109, 111], "inlin": [0, 1, 2, 3, 4, 29, 30, 44, 46, 48, 55, 82, 85, 88], "constexpr": [0, 1, 2, 45, 46, 92], "t": [0, 1, 2, 45, 46, 55, 60, 65, 66, 69, 75, 79, 81, 82, 87, 88, 90, 92, 105, 108, 109], "constructor": [0, 2, 46, 48, 49, 58, 87], "from": [0, 1, 2, 3, 4, 29, 30, 44, 46, 48, 49, 52, 53, 55, 56, 57, 58, 59, 60, 63, 64, 65, 68, 70, 71, 74, 75, 76, 77, 79, 80, 81, 82, 87, 88, 90, 92, 93, 94, 96, 97, 100, 101, 103, 104, 105, 107, 108, 110, 111, 112], "torchtrt_api": [0, 2, 19, 22, 23, 24, 25, 26, 27, 28, 31, 32, 33, 34, 35, 36, 37, 42, 43, 44, 45, 48, 49, 50], "scalartyp": [0, 45, 69], "torch": [0, 1, 2, 4, 20, 21, 29, 30, 31, 32, 33, 34, 37, 44, 45, 46, 47, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 70, 71, 73, 74, 75, 76, 77, 87, 90, 93, 94, 113], "paramet": [0, 1, 2, 3, 4, 25, 26, 27, 29, 30, 31, 32, 33, 34, 36, 46, 48, 49, 53, 54, 55, 60, 64, 65, 70, 71, 73, 74, 75, 76, 85, 87, 88, 97, 103, 104], "oper": [0, 1, 2, 3, 4, 31, 44, 45, 46, 49, 52, 53, 55, 56, 57, 58, 59, 60, 62, 63, 65, 68, 70, 75, 76, 89, 90, 97, 99, 101, 112, 113], "const": [0, 1, 2, 3, 4, 29, 30, 31, 32, 33, 34, 36, 44, 45, 46, 55, 60, 69, 88, 90], "get": [0, 1, 2, 3, 4, 23, 35, 44, 46, 55, 56, 60, 62, 63, 65, 74, 75, 88, 90, 94, 103, 104, 107, 108], "return": [0, 1, 2, 3, 4, 23, 24, 29, 30, 31, 32, 33, 34, 35, 42, 43, 44, 45, 46, 54, 55, 56, 57, 58, 59, 60, 62, 64, 65, 70, 71, 74, 75, 76, 87, 88, 89, 90, 92, 94, 97, 98, 105, 108, 109], "explicit": [0, 1, 2, 3, 4, 45, 46, 55, 65, 71, 74, 81, 90, 112], "delet": [0, 1, 2, 45, 46, 55], "other": [0, 1, 2, 45, 46, 52, 53, 55, 58, 62, 64, 65, 66, 69, 70, 74, 75, 80, 81, 88, 89, 110], "comparis": [0, 2], "true": [0, 1, 2, 4, 46, 49, 55, 56, 60, 62, 64, 65, 69, 70, 71, 74, 75, 76, 79, 82, 88, 90, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 108, 110, 113], "fals": [0, 1, 2, 3, 4, 44, 45, 46, 49, 54, 62, 64, 65, 69, 70, 71, 74, 75, 76, 79, 80, 81, 82, 88, 90, 91, 92, 93, 94, 96, 97, 98, 100, 101, 102, 103, 104, 105, 110], "struct": [1, 21, 38, 41, 45, 54, 90], "onli": [1, 3, 4, 16, 29, 44, 46, 48, 52, 54, 55, 56, 59, 60, 64, 65, 67, 70, 71, 74, 75, 81, 90, 92, 96, 97, 110, 113], "applic": [1, 29, 46, 52, 55, 59, 64, 70, 74, 75, 88, 89, 91, 110, 113], "kcuda": [1, 46, 56, 88], "which": [1, 2, 29, 32, 34, 46, 49, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 66, 70, 71, 73, 74, 75, 76, 79, 81, 82, 87, 88, 89, 90, 91, 92, 94, 98, 99, 102, 103, 104, 107, 108, 109, 110, 111, 112], "map": [1, 46, 53, 54, 55, 57, 59, 60, 65, 75, 88, 90, 91, 94, 98, 107, 108], "kgpu": [1, 45, 46], "To": [1, 46, 52, 54, 56, 64, 66, 70, 79, 87, 88, 89, 91, 92, 97, 108], "datatyp": [1, 21, 38, 45, 46, 48, 49, 50, 70, 75, 76, 89, 92, 108], "target": [1, 33, 45, 46, 48, 49, 52, 54, 56, 58, 59, 64, 65, 66, 68, 70, 74, 75, 76, 89, 90, 91, 92, 97, 112, 113], "gpu": [1, 32, 34, 36, 45, 46, 52, 64, 65, 70, 74, 75, 76, 88, 90, 91, 92, 103, 104, 108, 110, 112, 113], "run": [1, 34, 46, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 64, 65, 66, 67, 70, 71, 74, 75, 76, 81, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 107, 108, 109, 110, 111, 112, 113], "kdla": [1, 45, 46, 113], "dla": [1, 45, 46, 49, 52, 64, 68, 70, 75, 76], "intern": [1, 16, 46, 60, 63, 72, 74, 81, 88], "note": [1, 46, 48, 54, 60, 62, 65, 66, 74, 75, 79, 81, 88, 92, 97, 109, 113], "The": [1, 46, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 66, 70, 74, 75, 76, 79, 82, 87, 89, 90, 91, 92, 93, 94, 96, 97, 99, 102, 106, 107, 108, 109, 111, 112], "valid": [1, 46, 56, 60, 62, 70, 74, 75], "kcpu": [1, 46], "comparison": [1, 46], "an": [2, 3, 4, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 64, 65, 66, 67, 70, 71, 73, 74, 75, 76, 79, 81, 82, 87, 88, 89, 90, 92, 94, 97, 98, 102, 103, 104, 107, 108, 109, 110, 111, 112], "memeori": 2, "layout": [2, 48, 69, 70, 75, 76], "store": [2, 4, 49, 52, 53, 58, 60, 64, 65, 70, 74, 75, 76, 87, 88, 92, 94, 97], "tensor": [2, 33, 44, 45, 48, 49, 52, 53, 54, 55, 56, 58, 60, 62, 63, 64, 65, 69, 70, 71, 74, 75, 76, 87, 88, 89, 90, 92, 98, 107], "kcontigu": [2, 45, 48], "contigu": [2, 48, 49, 52, 70, 75, 76], "nchw": [2, 70, 75, 76], "linear": [2, 56, 69, 75, 87, 92, 105], "kchannelslast": [2, 45], "channel": [2, 75, 80], "last": [2, 55, 65, 75, 105], "nhwc": [2, 52], "memoryformat": [2, 45], "ptq": [3, 4, 15, 18, 19, 38, 50, 51, 52, 68, 70, 75, 76, 95, 106], "privat": [3, 4, 44, 45, 90], "algorithm": [3, 4, 29, 30, 44, 65, 73, 90], "typenam": [3, 4, 29, 30, 44], "gener": [3, 4, 29, 52, 55, 58, 59, 60, 62, 64, 65, 66, 70, 71, 79, 81, 82, 85, 87, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 105, 106, 110], "int8calibr": [3, 20, 30, 40, 44, 50], "implement": [3, 4, 55, 56, 58, 63, 65, 74, 80, 88, 90, 92, 94, 110], "specifi": [3, 4, 33, 52, 54, 60, 64, 65, 66, 70, 75, 76, 79, 81, 89, 91, 108, 109, 111, 112], "calibr": [3, 4, 29, 30, 44, 49, 52, 70, 73, 75, 76, 88, 90], "read": [3, 4, 29, 30, 44, 79, 81, 90], "nvinfer1": [3, 4, 29, 30, 44, 45, 49, 60, 90], "iint8calibr": [3, 4, 29, 30, 44, 45, 49, 70, 75, 76, 90], "iint8entropycalibrator2": [3, 4, 29, 30, 44, 90], "std": [3, 4, 22, 26, 28, 29, 30, 31, 33, 34, 35, 42, 44, 45, 47, 48, 49, 56, 88, 90, 108, 113], "string": [3, 4, 18, 20, 21, 22, 26, 28, 29, 30, 31, 33, 34, 35, 42, 44, 45, 49, 54, 56, 58, 60, 64, 70, 75, 79, 88, 90], "cache_file_path": [3, 4, 29, 30, 44], "8": [3, 52, 55, 63, 64, 66, 74, 75, 81, 82, 85, 88, 92, 99, 102, 108, 109], "cach": [3, 4, 29, 30, 44, 52, 64, 65, 70, 71, 73, 75, 88, 90, 95, 106, 110], "getbatchs": [3, 4, 44], "noexceptoverrid": [3, 4], "batch": [3, 4, 44, 64, 65, 71, 74, 90, 94, 99, 101, 105, 108, 109, 113], "size": [3, 4, 44, 48, 49, 52, 55, 56, 64, 65, 69, 70, 71, 75, 76, 79, 88, 90, 92, 94, 99, 101, 105, 107, 109], "next": [3, 4, 53, 54, 58, 63, 71, 75, 79, 81, 82, 90, 98, 102, 105, 108], "alwai": [3, 4, 27, 52, 75, 81, 97], "1": [3, 4, 33, 44, 45, 48, 49, 52, 54, 55, 56, 58, 60, 62, 63, 64, 65, 66, 69, 70, 71, 73, 74, 75, 76, 78, 79, 81, 82, 85, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 99, 101, 102, 104, 105, 107, 109, 111, 113], "due": [3, 4, 66, 80, 81, 105], "issu": [3, 4, 64, 70, 75, 88, 98, 101], "getbatch": [3, 4, 44], "void": [3, 4, 25, 26, 27, 28, 36, 37, 42, 44, 45], "bind": [3, 4, 33, 44, 74, 76, 81], "char": [3, 4, 44, 52, 88], "name": [3, 4, 31, 33, 34, 44, 54, 56, 58, 60, 65, 66, 71, 73, 74, 75, 76, 81, 82, 87, 88, 91, 92, 97, 102, 105, 108], "nbbind": [3, 4, 44], "Not": 3, "arrai": [3, 4, 33, 53, 54, 75, 76, 92], "pointer": [3, 4, 90], "fed": [3, 4, 48], "buffer": [3, 4, 65, 92], "each": [3, 4, 49, 53, 55, 56, 58, 60, 64, 65, 66, 70, 71, 74, 79, 81, 88, 97, 110], "input": [3, 4, 21, 29, 33, 38, 44, 45, 47, 49, 50, 52, 53, 54, 55, 56, 58, 60, 62, 63, 64, 65, 67, 69, 70, 71, 72, 74, 75, 76, 82, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 102, 103, 104, 105, 107, 108, 109, 110, 111, 112, 113], "number": [3, 4, 49, 52, 54, 55, 56, 60, 63, 64, 65, 70, 71, 75, 76, 79, 88, 89, 92, 95, 97, 99, 101, 106, 107, 112], "readcalibrationcach": [3, 4, 44], "size_t": [3, 4, 44, 90], "length": [3, 4, 44, 65, 69, 82], "how": [3, 4, 66, 81, 83, 85, 87, 91, 92, 94, 96, 98, 105, 107, 108, 109, 110], "enabl": [3, 4, 24, 49, 52, 54, 56, 57, 59, 64, 65, 66, 70, 71, 73, 74, 75, 76, 79, 94, 96, 97, 99, 101, 102, 110], "use_cach": [3, 4, 30, 44, 73, 90, 103, 104], "set": [3, 4, 16, 21, 25, 27, 29, 32, 34, 36, 45, 46, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 65, 66, 70, 71, 74, 75, 76, 79, 83, 86, 87, 88, 89, 90, 92, 97, 105, 107, 109, 110, 112, 113], "writecalibrationcach": [3, 4, 44], "write": [3, 4, 29, 30, 44, 65, 68, 81, 88, 90, 108], "provid": [3, 4, 49, 52, 54, 56, 58, 60, 62, 64, 65, 66, 67, 70, 71, 74, 75, 76, 81, 88, 89, 90, 91, 94, 95, 97, 98, 102, 106, 108, 109, 110, 111, 112], "cast": [3, 4, 55], "convienc": [3, 4, 49], "convert": [3, 4, 31, 32, 34, 52, 55, 56, 57, 59, 63, 64, 68, 70, 75, 76, 89, 91, 92, 99, 101, 107, 110], "easili": [3, 4, 96], "assign": [3, 4, 80], "ptq_calibr": [3, 4, 45, 49, 90], "field": [3, 4, 63, 71, 75, 90], "compilespec": [3, 4, 21, 32, 34, 41, 45, 50, 56, 76, 88, 90, 113], "dataloaderuniqueptr": [4, 44], "libtorch": [4, 37, 60, 66, 67, 88, 90, 112], "dataload": [4, 29, 30, 44, 49, 73, 90, 105], "unique_ptr": [4, 30], "unqiue_ptr": 4, "A": [4, 29, 30, 32, 33, 47, 48, 54, 55, 56, 60, 65, 66, 70, 71, 75, 76, 82, 90, 100, 108], "uniqu": [4, 89], "what": [4, 54, 55, 65, 67, 75, 81, 87, 88, 89, 104, 112], "make_data_load": [4, 90], "factori": [4, 29, 30, 64, 70, 90], "path": [4, 13, 14, 15, 29, 30, 52, 64, 65, 66, 70, 73, 75, 87, 88, 90, 94, 102, 105, 108, 112], "find": [4, 65, 66, 88, 92], "whether": [4, 52, 54, 64, 65, 70, 71, 75, 80, 90, 99, 101, 110], "exist": [4, 31, 32, 34, 54, 63, 64, 65, 70, 73, 75, 76, 90, 94, 107], "There": [4, 53, 54, 59, 60, 62, 63, 65, 66, 82, 87, 90, 97, 107, 108, 109, 110], "consum": [4, 53, 87], "macro": [5, 6, 7, 8, 9, 10, 11, 12, 15, 18, 20, 21, 42, 44, 45, 50, 51], "x": [5, 10, 33, 43, 55, 56, 66, 67, 74, 76, 82, 87, 88, 92, 94, 98, 102, 105, 109, 111], "includ": [13, 15, 16, 35, 37, 42, 43, 44, 45, 51, 52, 54, 56, 57, 58, 59, 62, 64, 65, 66, 67, 70, 71, 74, 75, 79, 81, 87, 88, 90, 92, 95, 106, 110], "parent": [14, 15, 18, 19, 20, 21], "cpp": [14, 15, 42, 43, 44, 45, 51, 55, 59, 66, 88, 90], "log": [15, 16, 19, 20, 38, 44, 50, 51, 55, 60, 64, 65, 68, 69, 70, 71, 75, 99, 101], "emum": [16, 17], "messag": [16, 25, 26, 52, 72], "sever": [16, 26, 72], "kinternal_error": [16, 42], "print": [16, 31, 44, 62, 64, 70, 76, 81, 88, 91, 92, 93, 94, 96, 97, 99, 101, 103, 104, 105, 108], "error": [16, 49, 52, 53, 55, 59, 64, 65, 70, 72, 75, 76, 81, 88, 109], "kerror": [16, 42], "all": [16, 42, 43, 44, 45, 49, 52, 54, 55, 56, 58, 62, 64, 65, 66, 70, 72, 74, 75, 77, 81, 82, 87, 88, 89, 90, 92, 106, 107, 108, 110, 112], "kwarn": [16, 42], "warn": [16, 44, 52, 60, 72, 74], "kinfo": [16, 42, 44], "info": [16, 32, 34, 45, 52, 60, 72, 74, 75], "kdebug": [16, 42, 44], "debug": [16, 27, 45, 49, 52, 60, 62, 64, 70, 72, 74, 75, 76, 91, 92, 93, 94, 96, 97, 98, 99, 101, 105], "kgraph": [16, 42, 55], "everyth": [16, 64, 70, 75], "intermedi": [16, 49, 52, 54, 64, 70, 72, 75, 76, 87, 112], "graph": [16, 31, 32, 34, 45, 49, 52, 53, 54, 56, 57, 59, 60, 62, 63, 64, 65, 70, 71, 72, 75, 76, 87, 88, 92, 94, 95, 96, 97, 99, 101, 102, 106, 107, 109, 110], "lower": [16, 54, 63, 65, 68, 70, 71, 72, 75, 82, 92, 94, 99, 101, 107, 112], "phase": [16, 60, 63, 88, 97, 109, 112], "select": [17, 29, 30, 34, 49, 52, 58, 64, 65, 66, 69, 70, 75, 76, 80, 83, 89, 90, 92, 112], "capabl": [17, 45, 49, 52, 58, 70, 75, 76, 91], "kstandard": [17, 45, 49], "ksafeti": [17, 45], "kdla_standalon": [17, 45], "directori": [18, 19, 20, 21, 42, 43, 44, 45, 50, 66, 70, 90, 94], "program": [18, 19, 20, 21, 29, 51, 52, 57, 58, 59, 68, 70, 87, 94, 95, 104, 106, 109], "list": [18, 19, 20, 21, 31, 49, 51, 53, 56, 58, 60, 62, 63, 65, 67, 69, 70, 71, 74, 75, 76, 85, 88, 89, 92, 108], "torchscript": [19, 21, 38, 43, 45, 49, 50, 52, 56, 57, 58, 59, 63, 67, 70, 71, 73, 74, 75, 76, 89, 107, 109, 113], "str": [19, 43, 44, 50, 54, 64, 65, 69, 70, 73, 74, 75, 76, 92, 94, 105], "torch_tensorrt_major_vers": [19, 43, 50], "torch_tensorrt_minor_vers": [19, 43, 50], "torch_tensorrt_patch_vers": [19, 43, 50], "torch_tensorrt_vers": [19, 43, 50], "torchtrt_hidden": [19, 43, 50], "xstr": [19, 43, 50], "nvinfer": [20, 44], "fstream": [20, 44], "iostream": [20, 21, 44, 45, 88], "iter": [20, 44, 49, 52, 53, 64, 70, 73, 75, 76, 93, 94, 105], "memori": [20, 21, 44, 45, 55, 60, 70, 75, 76, 88, 89, 92, 94, 103, 104], "sstream": [20, 44], "vector": [20, 21, 33, 44, 45, 47, 48, 49, 56, 58, 75, 88, 90, 113], "templat": [20, 40, 44, 45, 50, 79, 88], "int8cachecalibr": [20, 29, 40, 44, 50], "cuda_runtim": [21, 45], "custom_class": [21, 45], "devic": [21, 33, 36, 38, 45, 49, 50, 52, 58, 64, 69, 70, 71, 73, 74, 75, 76, 89, 90, 91, 92, 96, 100, 103, 104, 107, 113], "graphinput": [21, 38, 45, 49, 50], "devicetyp": [21, 38, 45, 46, 50, 74, 75, 76, 90, 91, 92, 113], "tensorformat": [21, 38, 45, 48, 50, 75, 92], "level": [23, 25, 26, 39, 42, 44, 50, 54, 55, 56, 59, 64, 65, 70, 75, 76, 85, 87, 92, 108, 112], "current": [23, 54, 56, 58, 60, 62, 63, 64, 65, 66, 70, 71, 74, 75, 76, 79, 92, 96, 103, 104, 105, 110], "report": [23, 44, 74], "Is": [24, 75], "color": [24, 27, 81], "output": [24, 27, 33, 49, 52, 53, 54, 55, 56, 58, 60, 62, 63, 64, 65, 66, 70, 72, 74, 75, 76, 79, 81, 82, 88, 92, 94, 96, 97, 100, 107, 108, 109, 111], "lvl": [25, 26, 42], "inform": [25, 33, 35, 37, 48, 52, 53, 56, 58, 62, 64, 65, 66, 70, 71, 72, 75, 81, 87, 88, 90, 91, 92, 94, 109], "ad": [25, 52, 53, 54, 56, 62, 65, 66, 92, 96], "abov": [25, 54, 56, 62, 65, 66, 72, 80, 81, 88, 92, 99, 101, 111], "msg": [26, 42], "add": [26, 53, 54, 55, 56, 60, 63, 66, 69, 79, 81, 86, 88, 89, 92], "global": [26, 52, 64, 70, 75, 88], "colored_output_on": [27, 42], "prefix": [27, 28, 42, 81], "help": [27, 52, 53, 60, 64, 65, 88, 94, 105, 107, 110], "when": [27, 44, 45, 46, 52, 53, 55, 56, 57, 58, 59, 60, 64, 65, 66, 70, 74, 75, 76, 79, 81, 83, 87, 88, 90, 92, 94, 96, 97, 107, 109, 110], "termin": [27, 52, 88], "If": [27, 33, 53, 54, 55, 56, 62, 63, 64, 65, 66, 67, 70, 71, 75, 79, 81, 88, 89, 90, 92, 94, 97, 98, 102, 108, 109, 110, 112, 113], "build": [29, 30, 35, 49, 52, 53, 57, 59, 60, 63, 64, 65, 70, 74, 75, 80, 85, 88, 90, 92, 99, 101, 109], "post": [29, 30, 49, 52, 63, 68, 88, 94], "train": [29, 30, 49, 52, 68, 69, 88, 89, 94], "quantiz": [29, 30, 52, 64, 68, 73, 75, 88, 95, 106], "creat": [29, 30, 33, 52, 53, 54, 56, 58, 60, 65, 68, 75, 76, 81, 88, 92, 95, 97, 106, 108], "previous": [29, 33, 88, 94, 97], "therefor": [29, 58, 65, 66, 74, 81, 88, 107, 110], "have": [29, 33, 44, 52, 53, 54, 55, 56, 60, 62, 63, 65, 66, 70, 71, 73, 74, 75, 76, 81, 87, 88, 89, 90, 92, 99, 101, 105, 107, 108, 109], "requir": [29, 49, 52, 53, 54, 55, 63, 64, 65, 66, 70, 75, 76, 79, 88, 90, 92, 105, 108, 109, 110], "dataset": [29, 73, 90, 107], "save": [29, 44, 52, 58, 64, 65, 67, 68, 70, 74, 75, 76, 88, 89, 93, 94, 97, 100, 107, 108, 110, 112], "later": [29, 70, 88, 97, 111, 112], "differ": [29, 55, 56, 59, 64, 65, 66, 75, 79, 87, 92, 94, 96, 110, 112], "scratch": [29, 94, 97], "depend": [29, 35, 53, 59, 64, 65, 67, 70, 88, 89, 108, 110], "howev": [29, 66, 79, 80, 88, 92, 94, 108, 109, 112], "network": [29, 30, 54, 60, 65, 75, 88, 90, 92, 107, 108, 113], "also": [29, 53, 54, 60, 62, 64, 66, 67, 79, 81, 82, 88, 89, 90, 94, 102, 105, 106, 107], "recalibr": 29, "its": [29, 53, 56, 58, 60, 66, 74, 75, 81, 92, 105, 108, 110, 112], "structur": [29, 46, 49, 56, 59, 60, 64, 70, 75, 79, 81, 85, 87, 92, 108], "chang": [29, 55, 56, 59, 62, 64, 65, 74, 75, 76, 79, 90, 94, 96, 97, 108, 110, 112], "respons": [29, 54, 58, 81, 110], "ensur": [29, 54, 55, 56, 62, 64, 66, 74], "By": [29, 30, 51, 56, 66, 79, 87, 94, 109], "entropi": [29, 30, 90], "v2": [29, 30, 81], "perform": [29, 30, 54, 62, 63, 70, 74, 75, 90, 92, 102, 107, 108, 110, 111, 112], "recommend": [29, 30, 65, 66, 75, 81, 88, 92, 108, 109], "feed": [29, 30, 88], "forward": [29, 30, 32, 33, 56, 58, 60, 64, 67, 70, 74, 75, 76, 87, 88, 89, 90, 91, 92, 98, 105, 109], "overrid": [29, 30, 44, 54, 65, 90], "minmax": [29, 30, 90], "recomend": [29, 30], "nlp": [29, 30, 90], "task": [29, 30, 65, 90, 107], "call": [29, 30, 32, 49, 54, 55, 58, 60, 65, 70, 71, 74, 75, 76, 81, 87, 88, 91, 92, 94, 96, 98, 101, 107, 109, 110, 112], "make_int8_calibr": [29, 40, 44, 50, 90], "class": [29, 30, 44, 45, 46, 51, 58, 60, 64, 65, 72, 76, 81, 82, 87, 88, 89, 90, 92, 94, 98, 105, 107, 109], "e": [29, 30, 52, 55, 60, 65, 66, 67, 71, 75, 87, 88, 90, 92, 94, 97], "g": [29, 30, 52, 55, 65, 66, 71, 75, 81, 90, 92, 97], "iint8minmaxcalibr": [29, 30, 90], "calibration_cache_fil": [29, 30, 90], "move": [30, 44, 55, 58, 76, 88, 90, 103, 104], "calibration_dataload": [30, 90], "contain": [30, 31, 52, 53, 54, 55, 56, 60, 65, 66, 71, 74, 75, 81, 82, 87, 88, 90, 92, 94, 108, 110], "jit": [31, 32, 33, 34, 45, 47, 49, 52, 53, 55, 56, 57, 58, 59, 60, 61, 64, 67, 68, 70, 74, 75, 76, 87, 88, 89, 91, 92, 97, 108, 111, 112], "modul": [31, 32, 33, 34, 45, 49, 52, 56, 57, 58, 59, 60, 64, 65, 66, 67, 68, 70, 71, 73, 74, 75, 76, 80, 81, 82, 89, 90, 91, 92, 95, 97, 98, 105, 106, 107, 109, 111, 113], "method_nam": [31, 34, 45, 52, 75, 76, 88], "see": [31, 55, 56, 58, 62, 64, 65, 66, 75, 76, 81, 87, 88, 89, 92, 94, 97, 98], "fulli": [31, 52, 55, 64, 70, 74, 75, 76, 88, 90, 92, 113], "compil": [31, 34, 41, 45, 49, 50, 52, 54, 55, 56, 58, 60, 62, 65, 70, 71, 72, 74, 75, 76, 77, 79, 87, 89, 90, 91, 92, 93, 96, 105, 108, 110, 113], "take": [31, 32, 33, 34, 53, 54, 57, 58, 59, 60, 62, 65, 70, 71, 74, 75, 76, 79, 81, 88, 90, 91, 92, 98, 107, 109], "method": [31, 32, 33, 34, 48, 52, 55, 60, 66, 70, 75, 76, 81, 87, 88, 91, 94, 107], "pure": [31, 70, 75], "Will": 31, "out": [31, 44, 53, 55, 56, 57, 59, 60, 64, 66, 70, 75, 76, 81, 88, 92, 96, 105, 108, 109], "unsupport": [31, 49, 54, 64, 75, 92, 112], "script": [31, 55, 56, 67, 75, 76, 87, 88, 89, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 110, 112], "nvidia": [32, 34, 42, 43, 44, 45, 52, 61, 64, 65, 66, 70, 75, 76, 88, 98, 101, 108, 112, 113], "configur": [32, 34, 48, 62, 66, 70, 74, 75, 76, 85, 88, 90, 92, 108, 109], "equival": [32, 57, 59, 60, 70, 75, 76, 87, 88, 90, 92, 99, 101], "specif": [32, 49, 54, 55, 57, 59, 62, 64, 70, 75, 76, 81, 106, 107, 112], "traget": 32, "input_binding_nam": [33, 45, 74, 76], "output_binding_nam": [33, 45, 74, 76], "emb": [33, 52, 63, 76, 82], "pre": [33, 55, 73, 76, 90, 94, 110], "built": [33, 52, 58, 59, 64, 66, 70, 74, 75, 76, 94, 97], "serial": [33, 34, 52, 57, 59, 66, 70, 74, 75, 76, 88, 92, 94, 112], "regist": [33, 54, 58, 60, 65, 74, 76, 92], "execut": [33, 49, 52, 55, 57, 58, 59, 63, 64, 65, 66, 68, 70, 71, 74, 75, 76, 77, 87, 88, 90, 92, 108], "must": [33, 48, 49, 52, 54, 55, 56, 60, 62, 65, 66, 70, 71, 75, 76, 81, 82, 88, 94, 109, 110, 112], "follow": [33, 52, 54, 56, 58, 62, 63, 64, 65, 66, 76, 79, 81, 82, 86, 87, 88, 90, 92, 94, 95, 99, 106, 107, 108, 109, 110], "format": [33, 45, 48, 49, 52, 69, 70, 75, 76, 81, 82, 89, 92, 94, 105, 107, 108, 111], "symbol": [33, 65, 66, 76, 81, 110], "index": [33, 61, 62, 66, 68, 69, 76, 79, 85, 90, 92], "0": [33, 43, 44, 45, 49, 52, 54, 56, 59, 60, 62, 64, 65, 66, 68, 69, 70, 71, 73, 74, 75, 76, 77, 78, 80, 81, 88, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 108, 109, 113], "2": [33, 43, 54, 56, 60, 63, 64, 65, 66, 68, 69, 70, 73, 74, 75, 76, 79, 81, 82, 85, 87, 88, 90, 92, 93, 94, 95, 96, 97, 98, 99, 101, 104, 105, 106, 109, 111], "y": [33, 56, 76, 82, 92, 98], "compilesepc": 33, "order": [33, 49, 54, 56, 60, 62, 65, 66, 70, 71, 74, 75, 76, 88, 89, 94], "pass": [33, 53, 54, 56, 57, 58, 59, 60, 63, 64, 65, 66, 68, 72, 73, 74, 75, 76, 87, 88, 90, 92, 94, 97], "origin": [33, 65, 71, 75, 92, 94, 96, 112], "pytorch": [33, 48, 49, 52, 54, 55, 56, 57, 58, 59, 60, 63, 64, 66, 67, 70, 73, 74, 75, 76, 87, 88, 89, 90, 94, 95, 96, 97, 105, 106, 108, 109, 110, 111, 112], "assum": [33, 74, 91, 92], "convent": 33, "below": [33, 56, 60, 62, 63, 64, 65, 66, 81, 88, 89, 94, 100, 108], "equivil": 34, "librari": [35, 42, 43, 44, 45, 52, 54, 57, 58, 59, 60, 75, 88, 92], "version": [35, 37, 59, 62, 64, 65, 70, 74, 75, 79, 82, 92, 107, 108, 111], "gpu_id": [36, 45, 46, 52, 74, 75, 76, 90, 91, 92, 113], "id": [36, 45, 52, 75, 79, 80, 84, 113], "cudasetdevic": 36, "dump": [37, 52, 92], "base": [37, 50, 58, 63, 64, 66, 70, 71, 75, 81, 87, 89, 90, 93, 97, 101, 107, 112], "stdout": [37, 74], "enginecap": [38, 45, 49, 50, 64, 70, 74, 75, 76, 91, 92], "dump_build_info": [38, 45, 50], "get_build_info": [38, 45, 50], "set_devic": [38, 45, 50, 110], "get_is_colored_output_on": [39, 42, 50], "get_logging_prefix": [39, 42, 50], "get_reportable_log_level": [39, 42, 50], "set_is_colored_output_on": [39, 42, 50], "set_logging_prefix": [39, 42, 50], "set_reportable_log_level": [39, 42, 50], "make_int8_cache_calibr": [40, 44, 50, 90], "check_method_operator_support": [41, 45, 50], "convert_method_to_trt_engin": [41, 45, 50, 75, 76, 88, 91], "embed_engine_in_new_modul": [41, 45, 50, 76], "document": [42, 43, 44, 45, 50, 59, 79, 81, 82, 86, 87, 88, 90, 91, 108, 109, 110], "copyright": [42, 43, 44, 45, 82, 88], "c": [42, 43, 44, 45, 52, 59, 64, 69, 70, 71, 74, 75, 82, 89, 92, 96, 108, 110, 113], "corpor": [42, 43, 44, 45], "right": [42, 43, 44, 45, 55, 59, 60, 81], "reserv": [42, 43, 44, 45, 103, 104], "licens": [42, 43, 44, 45, 88], "under": [42, 43, 44, 45, 59, 65, 81, 99, 112], "bsd": [42, 43, 44, 45], "style": [42, 43, 44, 45, 64, 67, 79, 81, 82], "found": [42, 43, 44, 45, 63, 66, 74, 81, 88, 90, 92, 94, 110], "root": [42, 43, 44, 45, 66, 79, 90, 105], "sourc": [42, 43, 44, 45, 54, 59, 64, 65, 70, 71, 72, 73, 74, 75, 76, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "tree": [42, 43, 44, 45, 79, 90, 105, 110], "pragma": [42, 43, 44, 45, 90], "onc": [42, 43, 44, 45, 53, 55, 56, 58, 64, 65, 66, 75, 90, 92, 108, 110], "namespac": [42, 43, 44, 45, 51, 55, 68, 75, 90, 92], "ar": [42, 46, 49, 52, 53, 54, 55, 56, 58, 59, 60, 62, 63, 64, 65, 66, 70, 73, 74, 75, 76, 79, 81, 82, 83, 87, 88, 90, 91, 92, 93, 94, 96, 97, 99, 107, 108, 109, 110, 111, 112], "ones": [42, 56, 57, 59, 66, 81, 88, 92, 112], "necessari": [42, 62, 64, 66, 74, 97, 110], "user": [42, 48, 54, 56, 57, 58, 59, 62, 63, 64, 66, 70, 81, 82, 88, 89, 90, 94, 97, 106, 108, 109, 110, 112], "dont": 42, "know": [42, 60, 79, 81, 92], "we": [42, 44, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 71, 74, 79, 81, 87, 88, 90, 92, 94, 95, 96, 97, 98, 99, 101, 102, 103, 104, 105, 106, 107, 108, 109, 111, 112], "want": [42, 56, 65, 66, 67, 71, 87, 88, 90, 91, 92, 97, 98, 108], "use_cmake_generated_export_head": 43, "torch_tensorrt_export": 43, "els": [43, 44, 48, 76, 81, 82, 93, 94, 105], "__gnuc__": 43, "__attribute__": 43, "__visibility__": 43, "hidden": [43, 79], "endif": [43, 44, 45], "doe": [43, 44, 55, 56, 60, 62, 65, 66, 75, 81, 90, 92, 99, 101], "gaurd": 43, "someth": [43, 55, 81, 108], "5": [43, 52, 56, 58, 59, 64, 65, 66, 70, 74, 75, 81, 82, 85, 87, 88, 92, 96, 98, 108], "setup": [43, 90, 108], "alias": 43, "eas": 43, "ts": [43, 52, 56, 67, 68, 75, 87, 88, 89, 91, 109, 111], "torchtrt": [43, 56, 92, 105], "ifndef": [44, 45], "doxygen_should_skip_thi": [44, 45], "get_batch_impl": 44, "element_typ": 44, "super": [44, 87, 92, 98, 105, 109], "batchtyp": 44, "dataloader_": 44, "cache_file_path_": 44, "use_cache_": 44, "auto": [44, 56, 60, 64, 67, 70, 81, 82, 88, 90, 103, 104, 113], "batched_data_": 44, "push_back": [44, 56], "it_": 44, "begin": [44, 65, 66, 81, 98, 102], "noexcept": [44, 90], "hack": 44, "explict": 44, "work": [44, 55, 59, 60, 64, 65, 67, 70, 73, 74, 75, 81, 82, 90, 92, 97, 98, 102, 109], "here": [44, 53, 54, 56, 58, 63, 64, 65, 66, 67, 79, 81, 82, 87, 88, 90, 92, 102, 103, 104, 105, 108, 109, 110, 111], "explic": 44, "just": [44, 45, 55, 56, 64, 65, 68, 72, 74, 81, 83, 87, 88, 89, 91, 92, 94, 96, 107, 110], "still": [44, 56, 65, 66, 90, 98, 112], "static_cast": 44, "option": [44, 48, 52, 56, 57, 59, 62, 63, 64, 65, 70, 74, 75, 76, 81, 85, 90, 92, 93, 94, 98, 100, 110, 111, 113], "batch_siz": [44, 90, 105], "end": [44, 52, 60, 62, 69, 70, 75, 76, 81, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "statu": [44, 82], "reset": [44, 93, 94, 98, 101, 110], "incas": 44, "go": [44, 55, 56, 65, 67, 87, 88, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 107, 108, 112], "again": [44, 58, 60, 81, 92, 96], "stringstream": 44, "ss": 44, "cache_": 44, "clear": 44, "ifstream": 44, "io": [44, 108], "binari": [44, 90], "noskipw": 44, "good": [44, 60, 65, 81, 94], "copi": [44, 60, 65, 69, 73, 82, 108], "istream_iter": 44, "back_insert": 44, "nullptr": [44, 45, 49], "ofstream": [44, 88], "cache_fil": [44, 73, 90], "reinterpret_cast": 44, "cache_size_": 44, "int8_t": 45, "arrayref": [45, 48, 49], "friend": 45, "ostream": 45, "os": [45, 94], "dtype": [45, 48, 49, 52, 63, 64, 65, 69, 70, 71, 74, 75, 76, 89, 92, 93, 99, 101, 102, 107, 109], "device_typ": [45, 46, 75, 90, 91, 113], "int64_t": [45, 46, 48, 49, 90, 113], "core": [45, 52, 55, 56, 59, 64, 70, 75, 88, 112, 113], "agx": 45, "platform": [45, 52, 59, 66, 108, 113], "xavier": [45, 113], "dla_cor": [45, 46, 52, 75, 90, 91, 113], "allow_gpu_fallback": [45, 46, 70, 75, 76, 90, 91, 113], "customclasshold": [45, 48], "min_shap": [45, 48, 63, 65, 70, 75, 76, 89, 99, 102, 107, 109], "opt_shap": [45, 48, 63, 70, 75, 76, 89, 99, 102, 107, 109], "max_shap": [45, 48, 63, 65, 70, 75, 76, 89, 99, 102, 107, 109], "shape": [45, 47, 48, 49, 52, 56, 60, 63, 65, 68, 69, 70, 71, 74, 75, 76, 89, 92, 102, 105, 108, 110, 113], "doubl": [45, 48, 49, 52, 63, 70, 75, 76, 81, 110], "tensor_domain": [45, 48, 75], "input_is_dynam": 45, "ivalu": [45, 47, 49, 53, 58, 60, 88], "input_signatur": [45, 47, 49, 76, 89], "nest": [45, 49, 50, 81, 82], "full": [45, 49, 52, 60, 64, 70, 72, 75, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 108, 110, 113], "spec": [45, 48, 49, 52, 72, 75, 76, 91, 94], "flatten": [45, 47, 69, 87, 88, 105], "fixed_s": [45, 49], "reflect": [45, 75], "builderconfig": 45, "graph_input": [45, 49], "enabled_precis": [45, 49, 63, 64, 70, 74, 75, 76, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 103, 104, 105, 108, 113], "disable_tf32": [45, 49, 64, 70, 74, 75, 76, 90, 92, 103, 104], "sparse_weight": [45, 49, 64, 65, 70, 74, 75, 76, 92], "refit": [45, 49, 64, 70, 75, 76, 91, 92, 94, 95, 96, 106], "truncate_long_and_doubl": [45, 49, 63, 64, 76, 100], "allow_shape_tensor": [45, 49, 76], "uint64_t": [45, 49], "num_avg_timing_it": [45, 49, 64, 70, 74, 75, 76, 91, 92], "workspace_s": [45, 49, 52, 64, 70, 74, 75, 76, 92, 97, 99, 101], "dla_sram_s": [45, 49, 52, 64, 70, 74, 75, 76, 92], "1048576": [45, 49, 64, 70, 74, 75, 76, 92], "dla_local_dram_s": [45, 49, 52, 64, 70, 74, 75, 76, 92], "1073741824": [45, 49, 64, 70, 74, 75, 76, 92], "dla_global_dram_s": [45, 49, 52, 64, 70, 74, 75, 76, 92], "536870912": [45, 49, 64, 70, 74, 75, 76, 92], "require_full_compil": [45, 49, 64, 70, 74, 75, 76, 92], "min_block_s": [45, 49, 56, 63, 64, 70, 74, 75, 76, 92, 93, 94, 97, 98, 99, 101, 104, 105], "3": [45, 49, 52, 55, 56, 58, 63, 64, 65, 67, 69, 70, 73, 75, 76, 81, 82, 85, 87, 88, 90, 91, 92, 93, 94, 96, 97, 99, 102, 105, 107, 109, 111, 113], "torch_executed_op": [45, 49, 56, 63, 64, 70, 74, 75, 76, 92, 97, 98, 99, 101], "torch_executed_modul": [45, 49, 56, 70, 75, 76], "member": [46, 47, 48, 49], "hold": [46, 47, 48, 53, 60, 75, 90], "relat": [46, 81, 98, 101], "let": [46, 52, 55, 60, 65, 70, 75, 76, 79, 81, 107, 108, 112], "layer": [46, 49, 52, 53, 55, 60, 62, 64, 65, 70, 74, 75, 76, 88, 90, 92, 105, 107, 108, 109, 112, 113], "thei": [46, 52, 53, 54, 55, 58, 60, 65, 73, 74, 75, 79, 81, 89, 94], "complex": [47, 49, 64, 66, 87, 89, 96], "either": [47, 48, 52, 60, 62, 70, 75, 76, 79, 81, 87, 88, 89, 92, 94, 111], "one": [47, 54, 55, 60, 64, 65, 70, 74, 75, 81, 87, 88, 89, 92, 98, 101, 103, 104, 108], "rang": [48, 49, 52, 65, 75, 92, 93, 94, 99, 107, 109], "optim": [48, 52, 63, 64, 65, 68, 70, 71, 73, 75, 87, 88, 89, 97, 99, 100, 101, 107, 109, 112], "profil": [48, 71, 74], "singl": [48, 52, 55, 56, 65, 75, 81, 87, 88, 90, 110], "repres": [48, 49, 54, 60, 65, 67, 81], "signifi": [48, 55], "static": [48, 49, 53, 60, 63, 64, 70, 75, 76, 79, 88, 105, 109], "three": [48, 57, 59, 65, 71, 75, 81, 82, 107, 108], "min": [48, 52, 60, 69, 75, 94, 99, 109], "optimin": 48, "max": [48, 52, 60, 69, 75, 79, 94, 99, 105, 109], "allow": [48, 49, 52, 53, 54, 55, 56, 62, 64, 65, 66, 70, 75, 76, 79, 92, 94, 97, 99, 101, 110], "argument": [48, 52, 54, 55, 58, 60, 62, 64, 65, 70, 74, 75, 76, 81, 82, 88, 89, 92, 109], "expect": [48, 54, 55, 60, 75, 88, 89, 107], "tradit": [48, 70, 75, 76, 90], "convect": 48, "produc": [48, 53, 54, 58, 60, 63, 75, 81, 88, 107], "low": [48, 65, 96], "high": [48, 55, 56, 79, 92, 112], "weight": [48, 49, 52, 53, 64, 65, 69, 70, 75, 76, 81, 88, 94, 95, 96, 100, 106, 107], "first": [48, 53, 54, 55, 65, 67, 81, 82, 88, 89, 90, 92, 94, 96, 98, 108, 109, 111, 112], "calcul": [48, 53, 56, 88, 92], "detect": [48, 58, 75], "float32": [48, 49, 52, 63, 64, 65, 70, 75, 76, 92, 96, 100, 103, 104, 109], "dynam": [48, 49, 63, 65, 68, 70, 71, 75, 76, 94, 98, 100, 101, 104, 110], "opt": [48, 66, 74, 75, 102], "minimum": [48, 49, 52, 56, 63, 64, 70, 75, 76, 92], "maximum": [48, 49, 52, 64, 65, 70, 71, 75, 76, 99, 101, 108], "accept": [48, 52, 54, 58, 60, 66, 75, 88, 89, 98, 111], "exampl": [48, 56, 58, 59, 60, 65, 66, 68, 70, 72, 74, 75, 76, 77, 79, 80, 82, 85, 87, 88, 89, 90, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 108, 109, 110, 111], "s": [48, 49, 53, 56, 58, 60, 63, 65, 66, 68, 70, 71, 74, 75, 79, 81, 82, 87, 88, 90, 92, 94, 107, 108, 109, 110, 111], "cannot": [48, 55, 56, 65, 66, 70, 74, 75, 76, 80, 87, 92], "through": [48, 53, 54, 55, 56, 58, 64, 65, 70, 72, 73, 81, 88, 89, 92, 96, 97, 107, 112], "altern": [48, 56, 62, 63, 75, 89, 102, 107, 111], "refer": [48, 54, 57, 59, 65, 80, 85, 88, 90, 92, 105, 108, 109, 111], "given": [48, 49, 52, 54, 55, 65, 70, 71, 73, 75, 76, 87, 88, 89, 91, 109], "kernel": [48, 49, 52, 60, 64, 65, 70, 75, 76, 95, 106, 110], "ani": [48, 52, 53, 54, 60, 62, 64, 65, 69, 70, 73, 74, 75, 76, 79, 81, 88, 89, 90, 92, 99, 109], "event": [48, 64, 93, 94], "place": [48, 55, 62, 65, 81, 82, 83, 90, 92, 105], "variabl": [48, 65, 74, 75], "dimens": [48, 55, 65, 71, 75, 99, 107, 109], "domain": [48, 75, 82, 90], "convien": 49, "fix": [49, 65, 81, 92, 110, 113], "describ": [49, 56, 60, 75, 87, 91, 95, 106, 108], "entri": [49, 60, 94], "okai": 49, "ha": [49, 53, 54, 55, 56, 57, 59, 60, 62, 64, 65, 66, 70, 71, 75, 81, 82, 87, 88, 90, 94, 97, 105, 107, 109, 112], "flaten": 49, "precis": [49, 52, 63, 64, 65, 70, 75, 88, 89, 90, 99, 101, 113], "dure": [49, 52, 54, 56, 60, 63, 64, 70, 73, 75, 90, 107, 109, 110], "prevent": [49, 52, 54, 56], "tf32": [49, 52, 64, 70], "comput": [49, 64, 65, 66, 70, 74, 81, 90, 107], "inner": [49, 82, 107], "product": [49, 75], "round": [49, 70, 75, 76, 92], "10": [49, 66, 70, 71, 75, 76, 85, 87, 88, 90, 105, 107, 108, 109], "bit": [49, 60, 65, 66, 70, 75, 76, 88], "mantissa": [49, 70, 75, 76], "befor": [49, 54, 55, 56, 59, 60, 65, 70, 75, 76, 88, 108, 109], "multipli": [49, 70, 75, 76], "accumul": [49, 70, 75, 76], "sum": [49, 65, 69, 70, 75, 76, 92, 105], "23": [49, 55, 70, 75, 76, 82], "behavior": [49, 56, 65, 70, 75, 76, 109, 110, 111], "sparsiti": [49, 52, 65, 70, 75, 76], "conv": [49, 52, 88, 92], "fc": [49, 52, 55], "truncat": [49, 52, 63, 64, 70, 75, 76], "long": [49, 52, 53, 63, 75, 81, 82], "float": [49, 52, 63, 64, 69, 75, 87, 88, 89, 90, 91, 92, 93, 94, 97, 98, 101, 102], "ishap": 49, "restrict": [49, 64, 70, 75, 76, 109], "cuda": [49, 58, 63, 65, 67, 70, 71, 74, 75, 88, 89, 90, 91, 92, 93, 94, 96, 97, 99, 100, 102, 103, 104, 105, 108, 109, 110, 111], "safeti": [49, 52, 75], "averag": [49, 52, 64, 70, 75, 76, 92], "time": [49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 64, 65, 66, 67, 68, 70, 71, 74, 75, 76, 79, 81, 88, 90, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "workspac": [49, 52, 64, 65, 66, 70, 71, 75, 76, 92, 98, 99, 101], "fast": [49, 52, 64, 67, 70, 75, 76], "softwar": [49, 52, 64, 70, 75, 76, 81], "manag": [49, 52, 53, 55, 57, 59, 60, 64, 70, 72, 74, 75, 76, 88, 102, 110], "ram": [49, 52, 64, 70, 75, 76], "commun": [49, 52, 64, 70, 75, 76, 88], "within": [49, 52, 57, 59, 64, 70, 74, 75, 76, 79, 81, 95, 106], "host": [49, 52, 64, 66, 70, 75, 76, 92, 108], "share": [49, 52, 64, 66, 70, 74, 75, 76, 94], "across": [49, 52, 55, 56, 64, 70, 75, 76, 79], "metadata": [49, 52, 54, 58, 60, 64, 70, 75, 76, 79, 97, 109], "quantizatiom": 49, "instead": [49, 52, 53, 54, 55, 66, 70, 74, 75, 88, 97, 105, 110], "potenti": [49, 70, 75, 84], "subgraph": [49, 52, 53, 54, 55, 60, 62, 88, 92, 94, 112], "aten": [49, 54, 55, 56, 60, 61, 64, 68, 69, 70, 75, 76, 88, 98, 112], "thrown": [49, 70, 75, 76], "empti": [49, 70, 71, 75, 76, 82, 87, 92], "torch_tensorrtnamespac": 50, "loggingenum": 50, "levelnamespac": 50, "ptqtemplat": 50, "int8cachecalibratortempl": 50, "int8calibratornamespac": 50, "torchscriptstruct": 50, "compilespecstruct": 50, "deviceclass": 50, "devicetypestruct": 50, "graphinputsclass": 50, "datatypeclass": 50, "tensorformatenum": 50, "enginecapabilitystruct": 50, "cppdirectori": 50, "includedirectori": 50, "torch_tensorrtfil": 50, "hfile": 50, "relationship": 50, "inherit": [50, 65, 70, 90], "subdirectori": 51, "definit": [51, 54, 60, 81], "cli": [52, 89], "It": [52, 54, 55, 56, 57, 59, 60, 65, 66, 68, 75, 79, 81, 92, 107, 110, 112], "serv": [52, 58, 65, 68, 70, 75], "easi": [52, 53, 55, 88, 90], "wai": [52, 64, 65, 66, 87, 88, 90, 92, 94, 95, 97, 106, 107, 110, 111], "command": [52, 64, 66, 81, 82, 87, 88, 108], "line": [52, 66, 82, 88, 96], "quickli": [52, 88, 90], "part": [52, 56, 59, 65, 74, 79, 80, 81, 92, 94], "deploy": [52, 74, 88, 89, 90, 107, 108, 110, 113], "pipelin": [52, 88, 96, 100, 113], "basic": [52, 56, 65, 82, 106, 108], "featur": [52, 56, 65, 66, 88, 90, 91, 100, 105, 107, 112], "though": [52, 59, 60, 87, 88, 112], "alreadi": [52, 53, 54, 55, 88, 90, 92, 109], "two": [52, 55, 60, 62, 64, 65, 66, 75, 81, 82, 86, 87, 89, 90, 94, 108, 109], "embed": [52, 54, 58, 69, 76, 81, 113], "plan": [52, 59, 63, 64, 70], "after": [52, 53, 55, 56, 62, 65, 70, 74, 75, 87, 88, 89, 98, 101, 108, 110], "link": [52, 53, 62, 68, 79, 80, 85, 88, 92, 110], "against": [52, 88], "libtorchtrt": [52, 66, 88], "python": [52, 56, 59, 62, 64, 65, 70, 71, 74, 75, 76, 81, 82, 88, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 110, 113], "import": [52, 55, 56, 63, 64, 65, 66, 67, 74, 79, 81, 87, 88, 89, 91, 92, 93, 94, 96, 108, 109, 110, 111], "packag": [52, 55, 64, 88], "aspect": 52, "ident": [52, 62, 70, 75, 97], "standard": [52, 58, 66, 68, 70, 74, 75, 76, 81, 91, 92, 96, 107, 110], "load": [52, 56, 58, 64, 65, 67, 70, 73, 74, 75, 76, 88, 89, 90, 91, 92, 93, 94, 96, 97, 107, 108, 110, 112], "like": [52, 53, 55, 58, 60, 65, 66, 67, 75, 80, 81, 87, 88, 89, 90, 92, 94, 96, 97, 108, 110], "would": [52, 54, 60, 64, 65, 66, 74, 88, 89, 91, 92, 108, 110], "input_file_path": [52, 113], "output_file_path": [52, 113], "input_spec": [52, 65, 71], "displai": [52, 62, 64, 72, 79, 110], "menu": [52, 79, 81], "verbios": 52, "v": [52, 82, 105, 108], "verbos": [52, 64, 65, 70, 71, 82, 99, 101], "about": [52, 53, 58, 60, 66, 74, 79, 88, 108, 109], "process": [52, 56, 64, 75, 80, 81, 87, 90, 91, 97, 98, 102, 107, 108, 110], "onto": [52, 58], "consol": 52, "w": [52, 66, 75], "disabl": [52, 64, 66, 70, 74, 79, 80, 94, 110], "i": [52, 55, 60, 66, 67, 69, 81, 82, 87, 88, 90, 92, 93, 94, 105], "debugg": [52, 70, 75, 76], "fallback": [52, 57, 59, 60, 97, 113], "model": [52, 56, 58, 63, 67, 68, 70, 71, 72, 73, 75, 87, 88, 89, 90, 91, 93, 94, 95, 96, 106, 109, 110, 112], "throw": [52, 55, 75, 88], "spars": [52, 54, 64, 69, 70], "p": [52, 69, 88, 108, 113], "repeat": [52, 69], "f32": [52, 70, 74, 75, 92], "half": [52, 64, 75, 81, 88, 89, 90, 91, 92, 98, 99, 108, 113], "float16": [52, 75, 92, 96, 100], "f16": [52, 75, 88, 113], "i8": [52, 75], "d": [52, 75, 81, 82, 88, 113], "multi": [52, 74], "dlacor": 52, "avail": [52, 54, 60, 62, 64, 65, 66, 70, 74, 75, 79, 92, 112, 113], "dla_standalon": [52, 75], "file_path": [52, 75, 111], "teo": 52, "op_nam": 52, "op": [52, 53, 54, 55, 56, 57, 59, 60, 62, 63, 64, 74, 75, 88, 98, 110, 112], "partial": [52, 81], "tem": 52, "module_nam": 52, "mod": [52, 56, 65, 70, 85, 88, 90], "mb": [52, 77], "num_op": 52, "block": [52, 53, 55, 56, 64, 70, 85, 112], "treat": 52, "num": 52, "avg": 52, "num_it": 52, "sram": 52, "local": [52, 55, 66, 79, 88], "dram": 52, "atol": 52, "absolut": [52, 66], "toler": 52, "threshold": 52, "numer": [52, 65, 82], "deviat": 52, "1e": [52, 96, 97], "rtol": 52, "rel": [52, 56], "skip": 52, "complianc": 52, "64bit": 52, "32bit": 52, "custom": [52, 62, 63, 65, 66, 95, 103, 104, 106], "dll": 52, "n": [52, 60, 62, 75, 88, 90, 92, 93], "min_n": 52, "min_c": 52, "min_h": 52, "min_w": 52, "opt_n": 52, "opt_c": 52, "opt_h": 52, "opt_w": 52, "max_n": 52, "max_c": 52, "max_h": 52, "max_w": 52, "32": [52, 75, 87, 88, 89, 90, 103, 104, 105, 113], "flag": [52, 56, 57, 59, 66, 73, 75, 89, 102, 110, 111], "forc": [52, 63, 65, 70, 75, 76, 79], "posit": [52, 54, 65, 75, 79], "test": [52, 56, 59, 65, 66, 70, 75, 81, 82, 90, 105, 107, 108], "ssd_trace": 52, "pt": [52, 65, 88, 103, 104, 108], "ssd_trt": 52, "300": [52, 91], "512": [52, 70, 75, 76, 105, 107], "1024": [52, 70, 75, 76, 103, 107], "simplifi": [53, 92], "form": [53, 74, 75, 81, 89, 108], "up": [53, 55, 56, 57, 58, 59, 62, 65, 66, 70, 75, 81, 87, 92, 94, 95, 97, 98, 101, 106, 107], "context": [53, 57, 58, 59, 64, 72, 74, 102, 110], "inetworkdefinit": [53, 54], "record": [53, 87, 93, 94, 102, 110], "togeth": [53, 60, 88], "start": [53, 56, 65, 69, 73, 75, 82, 88, 91, 92, 93, 94, 107], "look": [53, 54, 55, 67, 70, 75, 87, 90, 91, 94, 108, 109], "assembl": [53, 62, 88], "resourc": [53, 90, 92], "coupl": [53, 59, 65, 110], "state": [53, 54, 60, 62, 74, 88, 96], "been": [53, 60, 64, 66, 73, 82, 88, 94, 97, 112], "evaluated_value_map": [53, 60], "stage": [53, 65], "arg": [53, 54, 62, 65, 70, 73, 74, 75, 85, 88, 92, 94, 105, 107], "itensor": [53, 54, 60, 65, 88, 92], "value_tensor_map": [53, 60], "typic": [53, 60, 75, 108], "abl": [53, 55, 60, 62, 65, 90, 91, 92, 97], "system": [53, 60, 62, 64, 68, 70, 74, 75, 76, 92, 94, 97, 112], "registri": [53, 54, 88, 92], "enter": [53, 75], "recurs": 53, "resolv": [53, 55, 57, 59, 98, 101], "until": [53, 56, 59, 60, 66, 70, 75, 112], "final": [53, 56, 57, 59, 66, 92, 98, 101, 107], "some": [53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 65, 66, 75, 80, 81, 88, 90, 92, 94, 109, 112], "These": [53, 54, 56, 58, 62, 64, 66, 70, 73, 74, 75, 79, 81, 90, 108, 112], "those": [53, 54, 62, 64, 81], "do": [53, 54, 55, 56, 60, 63, 65, 80, 82, 87, 88, 89, 90, 92, 113], "theori": [53, 81], "kind": [53, 65], "common": [53, 55, 65, 71, 81, 94], "prim": [53, 55, 56, 58, 69, 87, 88], "constant": [53, 54, 55, 56, 88, 92], "emit": 53, "listconstruct": [53, 56, 58, 88], "make": [53, 54, 65, 66, 70, 75, 81, 83, 88, 89, 90, 92, 94, 95, 106, 107, 108, 113], "associ": [53, 60, 88, 94, 110], "where": [53, 54, 55, 60, 62, 64, 65, 70, 74, 75, 76, 82, 88, 90, 97], "result": [53, 55, 56, 66, 67, 70, 72, 74, 75, 76, 79, 87, 89, 92, 96, 97, 108, 112], "done": [53, 56, 59, 92, 97, 108, 111], "mai": [53, 54, 56, 58, 59, 65, 66, 70, 74, 75, 76, 81, 82, 87, 88, 89, 90, 92, 97, 98, 101, 108, 110], "For": [53, 56, 62, 63, 64, 65, 66, 67, 71, 75, 79, 81, 82, 87, 88, 90, 91, 92, 96, 98, 105, 107, 108, 110, 111], "more": [53, 64, 65, 66, 68, 70, 75, 79, 82, 87, 88, 89, 90, 91, 92, 94, 96, 99, 101, 108, 110], "writing_convert": [53, 88], "locat": [54, 62, 66, 90, 92], "py": [54, 55, 59, 62, 65, 66, 77, 79, 81, 86, 87, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 109], "convers": [54, 55, 56, 58, 63, 64, 65, 70, 75, 76, 88, 92, 109], "decror": 54, "dynamo_tensorrt_convert": [54, 92], "signatur": [54, 76], "leaky_relu": [54, 69], "def": [54, 62, 65, 81, 87, 89, 92, 93, 94, 98, 105, 108, 109], "leaky_relu_convert": 54, "ctx": [54, 60, 88, 92], "conversionctx": [54, 60, 88], "tupl": [54, 58, 63, 65, 70, 71, 74, 75, 76, 89, 92, 94, 97, 109], "kwarg": [54, 65, 70, 73, 74, 75, 92, 107], "dict": [54, 70, 74, 75, 76, 92, 94], "union": [54, 60, 64, 70, 74, 75, 76, 88], "sequenc": [54, 62, 65, 70, 71, 74, 75, 76, 81, 92, 107], "decor": [54, 62, 65], "kei": [54, 81, 87, 94, 108, 109], "node": [54, 55, 56, 57, 59, 60, 62, 65, 71, 88, 92, 105, 107, 109], "capability_valid": 54, "lambda": [54, 60, 81, 88, 108], "fx": [54, 62, 63, 70, 74, 75, 88, 89, 92, 97, 111], "determin": [54, 55, 64, 65, 75, 109, 110], "properli": [54, 66], "handl": [54, 55, 56, 58, 64, 65, 74, 75, 92], "partition": [54, 70, 75, 92], "sure": [54, 66, 88, 89, 108, 113], "prioriti": 54, "develop": [54, 65, 66, 68, 81, 82, 88, 92], "bodi": [54, 81, 82], "nativ": [54, 59, 61, 88, 92, 97], "numpi": [54, 75, 92, 93, 94, 96, 97, 108], "frozen": 54, "attribut": [54, 55, 56, 58, 65, 75, 81, 88], "previou": [54, 79, 98], "correspond": [54, 60, 65, 66, 74, 75, 94, 96, 105, 110], "edg": [54, 81], "well": [54, 63, 66, 68, 72, 74, 81, 88, 90, 94, 102, 111], "being": [54, 65, 66, 70, 88, 92, 97], "truth": 54, "http": [54, 61, 64, 66, 79, 81, 87, 88, 90, 92, 96, 98, 101, 105, 107, 108, 109, 110], "github": [54, 61, 64, 66, 79, 88, 90, 98, 101, 105, 108, 110], "com": [54, 61, 64, 66, 88, 90, 96, 98, 101, 105, 108, 110], "blob": [54, 61, 66, 79, 90, 94], "main": [54, 55, 56, 57, 58, 59, 60, 63, 65, 66, 79, 81, 83, 88, 92, 105], "src": [54, 58, 61, 69], "native_funct": [54, 61], "yaml": [54, 61], "sinc": [54, 55, 64, 65, 74, 81, 87, 88, 90, 93, 94, 97], "mani": [54, 56, 64, 65, 79, 81, 82, 94, 97, 112], "composit": [54, 88], "raw": [54, 79], "impl": 54, "subpackag": 54, "chain": [54, 60], "primarili": [54, 59, 66, 88], "manipul": [54, 62, 75], "net": [54, 60, 81, 82, 88, 92], "addit": [54, 55, 64, 65, 74, 75, 88, 92, 94, 97, 107, 109], "call_modul": 54, "call_funct": [54, 62, 65], "eg": [54, 108], "aten_": 54, "_leaky_relu": 54, "opoverloadpacket": 54, "while": [54, 56, 66, 74, 90, 96, 107, 108, 110, 112], "opoverload": 54, "particular": [54, 64, 94], "collect": [54, 56, 64, 70, 75, 76, 88, 89, 105], "trtinterpret": [54, 65, 71], "along": [54, 75], "match": [54, 55, 97], "special": [54, 56], "account": [54, 108], "illustr": [54, 65, 99, 107], "scale_grad_by_freq": [54, 69], "embedding_param_valid": 54, "establish": 54, "subset": [54, 64, 70, 75, 90, 107], "converter_util": [54, 92], "enforce_tensor_typ": 54, "dictionari": [54, 75, 76, 91, 98], "between": [54, 55, 56, 60, 66, 75, 81, 82, 90, 94, 96], "possibl": [54, 66, 81, 92, 94, 107, 108], "prefer": [54, 64, 66, 88], "keyword": [54, 62, 70, 74, 75, 76, 98, 101], "both": [54, 56, 64, 66, 68, 70, 71, 74, 75, 79, 81, 87, 90, 92, 94], "enforc": [54, 88], "situat": 54, "partit": [54, 55, 63, 64, 70, 75, 112], "greater": [54, 70, 72, 75], "than": [54, 55, 64, 66, 70, 75, 80, 81, 93, 94, 96, 107, 110], "3d": [54, 65], "autocast": 54, "therebi": [54, 58, 92, 107], "limit": [54, 55, 72, 80, 90, 94, 112], "author": [54, 82], "conv_nod": 54, "7": [54, 56, 58, 59, 74, 75, 85, 88, 92, 98, 99, 101, 105, 109], "ignor": [54, 70, 74, 75, 92], "misc": [54, 92], "trttensor": 54, "np": [54, 92, 93, 94, 96, 97, 108], "ndarrai": [54, 92], "aten_ops_convolut": 54, "conversioncontext": [54, 92], "side": [54, 55, 79, 88], "effect": [54, 55, 64, 65, 70, 79, 88, 90, 92, 107], "term": [54, 75, 81, 82, 90, 92, 107], "getitem": 54, "categor": 54, "modif": [54, 62, 75], "op_evalu": 54, "capbility_valid": 54, "opcod": 54, "decompos": 54, "suboper": 54, "separ": [54, 56, 57, 59, 66], "Such": 54, "via": [54, 64, 65, 68, 70, 74, 75, 76, 79, 85, 89, 90, 98, 99, 101, 107, 109, 110, 111, 112], "register_torch_trt_decomposit": 54, "addmm_replac": 54, "replac": [54, 56, 62, 66, 73, 92, 105, 112], "input_": 54, "mat1": 54, "mat2": [54, 69], "beta": [54, 65, 69, 76], "alpha": [54, 65, 69, 82], "mul": [54, 56, 69], "matmul": [54, 55, 69, 88, 109], "modifi": [54, 56, 62, 65, 82, 95, 96, 106, 109], "edit": [54, 66, 79], "torch_enabled_decomposit": 54, "torch_disabled_decomposit": 54, "disjoint": 54, "preced": [54, 81], "over": [54, 57, 59, 65, 81, 105, 108, 112], "much": [54, 60, 79, 81, 90], "significantli": [54, 55, 79, 94], "easier": [54, 57, 59, 60, 65, 70, 74, 75, 88, 90, 92, 96], "tri": 54, "made": [55, 57, 59, 75, 81], "represent": [55, 60, 65, 87, 107, 112], "instanc": [55, 62, 64, 66, 70, 73, 74, 87, 88, 107, 110], "idea": [55, 81], "reduc": [55, 56, 57, 59, 65, 70, 75, 90, 92, 94, 107, 110], "actual": [55, 58, 60, 65, 87, 88, 92], "aim": [55, 112], "closer": 55, "scope": [55, 92, 98, 101], "csrc": [55, 61], "common_subexpression_elimin": 55, "subexpress": 55, "dead_code_elimin": 55, "exception_elimin": 55, "wa": [55, 58, 62, 64, 65, 70, 74, 75, 81, 88, 112], "1013": 55, "ne": [55, 69], "1012": 55, "24": 55, "lib": [55, 66, 88], "python3": [55, 66, 88], "6": [55, 56, 58, 66, 69, 85, 87, 88, 92], "site": [55, 66, 81, 88], "nn": [55, 61, 65, 70, 71, 74, 75, 76, 87, 88, 89, 92, 98, 105, 109, 112], "batchnorm": 55, "248": 55, "11": [55, 66, 81, 85, 88, 108], "block0": 55, "raiseexcept": 55, "249": 55, "12": [55, 56, 81, 85, 87, 88, 99, 108, 109], "block1": 55, "guard_elimin": 55, "whose": [55, 65, 99], "freeze_modul": 55, "propag": 55, "fuse_addmm_branch": 55, "variant": [55, 110], "caught": 55, "ret": 55, "622": 55, "self": [55, 58, 60, 69, 74, 75, 87, 88, 89, 92, 94, 98, 105, 107, 109, 113], "bia": [55, 69, 88, 105], "x9": 55, "3677": 55, "output0": 55, "add_": [55, 69, 88], "fuse_linear": 55, "back": [55, 56, 58, 59, 74, 75, 81, 87, 88, 92, 112], "fuse_flatten_linear": 55, "implicitli": [55, 75], "connect": [55, 70, 75, 76, 81, 96, 108, 113], "higher": [55, 64, 70, 75, 79, 81, 87], "1d": 55, "lower_graph": 55, "access": [55, 60, 65, 79, 88, 91, 112], "rather": 55, "getattr": [55, 58, 87, 88], "trainabl": 55, "remain": [55, 75, 90, 112], "lower_tupl": 55, "lowersimpletupl": 55, "tupleconstruct": [55, 58], "tupleunpack": 55, "leav": [55, 62, 64, 70], "statement": [55, 81], "loweralltupl": 55, "_all_": 55, "rais": [55, 65, 75], "onnx": 55, "module_fallback": 55, "consist": [55, 65, 81, 92, 110, 112], "pair": [55, 60, 66, 81, 90, 107], "delimit": 55, "around": [55, 58, 60, 66, 74, 81, 84, 87, 92], "second": [55, 65, 81, 89, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "mark": [55, 56, 79, 94], "notatemoduleforfallback": 55, "marknodesforfallback": 55, "tell": [55, 56, 57, 58, 59, 60, 81, 112], "them": [55, 56, 58, 63, 64, 65, 66, 70, 74, 79, 88, 92, 94, 107, 109, 112], "peephole_optimz": 55, "intent": [55, 81], "catch": [55, 75, 88], "small": [55, 92, 93, 108], "might": [55, 66, 79, 97, 109], "interest": [55, 81], "now": [55, 56, 59, 60, 65, 66, 75, 81, 88, 91, 92, 94, 97, 110], "expand": [55, 69], "simpli": [55, 98, 107], "remove_contigu": 55, "remove_dropout": 55, "infer": [55, 64, 65, 70, 75, 76, 88, 90, 95, 97, 98, 106, 107, 109, 110, 111, 112], "remove_to": 55, "unpack_addmm": 55, "reus": [55, 65, 90, 94], "dedic": [55, 82], "unpack_log_softmax": 55, "softmax": [55, 65, 69, 105], "loop_unrol": 55, "suffici": [55, 66, 75], "short": [55, 64, 70, 81, 82, 97], "tile_to_repeat": 55, "instruct": [56, 57, 59, 65, 66, 88, 108], "criteria": [56, 57, 59, 64], "lack": [56, 57, 59, 65, 92], "explicitli": [56, 57, 59, 66, 76, 89, 90, 91], "On": 56, "segment": [56, 63, 92, 99, 101, 107], "verifi": [56, 70, 92, 97], "Then": [56, 90, 91, 97], "roughli": 56, "analysi": 56, "everi": [56, 71, 74, 75, 88, 110], "complet": [56, 63, 70, 75, 87, 88], "mean": [56, 60, 65, 69, 71, 98, 108, 112], "trace": [56, 65, 70, 74, 76, 87, 88, 109, 111, 112], "tensorlist": [56, 60], "figur": [56, 82, 84], "our": [56, 59, 63, 87, 88, 108], "stitch": [56, 88], "altogeth": [56, 79], "brief": 56, "descript": [56, 82, 105], "partitioninfo": 56, "api": [56, 59, 60, 62, 63, 64, 65, 74, 75, 76, 80, 88, 89, 90, 91, 92, 95, 98, 99, 102, 106, 107, 108, 109, 110, 111], "maintain": [56, 58, 60, 75, 96, 112], "code": [56, 59, 62, 64, 65, 66, 80, 82, 87, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 109], "mymodel": [56, 63, 67, 89, 92, 109, 111], "ts_model": [56, 88], "trt_model": [56, 91, 92, 99, 103, 104, 105, 108, 111], "off": [56, 58, 102], "consecut": [56, 63], "satisfi": [56, 62, 65], "forced_fallback_op": 56, "randn": [56, 63, 67, 70, 75, 76, 88, 91, 94, 99, 102, 109, 111], "224": [56, 63, 67, 70, 71, 75, 76, 88, 94, 96, 97, 99, 102, 107, 108, 109, 111], "trt_ts_modul": [56, 89], "input_s": 56, "inputrang": 56, "cfg": [56, 88], "relu": [56, 69, 87, 88, 98, 105], "trt_mod": [56, 67, 88, 90, 113], "consid": [56, 76, 88, 92], "segmentmodelwithdependencyawar": 56, "test_segment": 56, "20": [56, 85, 97, 99, 101], "x_lgamma": 56, "lgamma": 56, "y_lgamma": 56, "div": [56, 69], "div_lgamma": 56, "27": [56, 88], "cat": [56, 66, 69, 105], "greedi": [56, 103, 104], "strategi": [56, 75], "travers": [56, 57, 59, 64], "gather": 56, "same": [56, 58, 62, 64, 65, 66, 70, 75, 79, 81, 87, 88, 91, 92, 94, 97, 99, 101, 108, 109, 110, 111], "encount": [56, 64, 66, 98, 101], "4": [56, 58, 63, 64, 65, 66, 69, 75, 77, 79, 81, 82, 85, 88, 92, 98, 100, 101, 102, 105, 109], "suboptim": 56, "arithmet": 56, "split": [56, 65, 69], "own": [56, 60, 64, 66, 70, 81, 88, 94, 105, 108], "could": [56, 64, 65, 92, 99, 101, 110], "rewrit": [56, 62], "portion": [56, 81, 92, 100], "without": [56, 60, 67, 70, 79, 81, 88, 90, 92, 93, 94, 97, 110], "reorder": 56, "seri": 56, "cleanli": 56, "approach": [56, 94], "achiev": [56, 107], "hit": 56, "larger": [56, 70, 75, 79, 107], "boundari": [56, 73, 75], "guarante": [56, 74], "trigger": [56, 64, 65, 75, 88, 94, 96, 97, 112], "appear": [56, 81], "adjac": [56, 70, 75, 81], "As": [56, 65, 66, 75, 88, 92, 94, 97, 112], "clean": [56, 62, 81, 98, 101], "step": [56, 65, 69, 75, 90, 92, 97, 107], "consolid": [56, 87], "further": [56, 64, 65, 110, 112], "merg": 56, "identifi": 56, "do_not_merg": 56, "combin": [56, 64, 65], "condit": [56, 81, 112], "loop": [56, 64, 65, 103, 104], "ir": [57, 59, 60, 63, 64, 67, 70, 75, 87, 88, 89, 95, 98, 99, 101, 102, 106, 109], "larg": [57, 59, 79, 81, 88, 90, 97, 107], "opset": [57, 59], "compon": [57, 59, 66, 73, 87, 110, 112], "evalu": [57, 58, 59, 105], "deploi": [57, 59, 68, 88, 90, 95, 106, 108], "instanti": [57, 58, 59, 60, 88, 100], "wrap": [57, 58, 59, 65, 81, 84, 88, 91, 98, 101], "extend": [57, 59, 60, 69, 88, 94, 107], "providi": [57, 59], "stand": [58, 81], "interpret": [58, 65, 81], "execute_engin": [58, 74, 88], "stack": [58, 69, 90, 105, 112], "machin": [58, 66, 90, 108], "pop": 58, "push": 58, "element": [58, 65, 81, 82, 85], "realiz": 58, "abstract": [58, 60, 82], "__torch__": [58, 87, 88], "portabl": [58, 66, 76], "serializ": [58, 64, 87, 112], "instnanti": 58, "whatev": [58, 65, 92], "self_1": [58, 88], "torchvis": [58, 90, 91, 94, 96, 97, 99, 102, 105, 108], "resnet": [58, 77, 95, 96, 106, 107, 108], "___torch_mangle_4847": 58, "resnet_trt": 58, "input_0": [58, 88], "__torch___torchvision_models_resnet____torch_mangle_4847_resnet_trt_engin": 58, "listunpack": [58, 88], "multipl": [58, 66, 70, 74, 75, 81, 82, 90, 108, 110], "repack": 58, "ssd": 58, "ssd300_trt": 58, "__torch___pytorch_detection_ssd_src_model_ssd300_trt_engin": 58, "holder": [58, 83], "torchbind": 58, "pickler": 58, "seril": 58, "zip": [58, 66, 96, 97, 106], "depickl": 58, "encod": [58, 107], "sm": 58, "correct": [58, 66, 79, 96, 97, 105], "bazel": [59, 66], "linux": [59, 88], "x86_64": [59, 66], "aarch64": 59, "gcc": [59, 88], "untest": 59, "try": [59, 75, 81, 82, 88, 91, 92, 94, 112], "older": 59, "repositori": [59, 66, 79, 86, 108], "notebook": [59, 68, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "doc": [59, 61, 66, 79, 80, 81, 86, 92, 109], "docsrc": 59, "third_parti": [59, 66], "toolchain": [59, 66], "unstabl": 59, "subject": [59, 62, 112], "matur": 59, "most": [59, 65, 66, 71, 92, 97, 108, 110, 112], "hood": [59, 99, 112], "major": [59, 65, 75], "top": [59, 79, 83], "coordin": [59, 75], "ingest": 59, "flow": [60, 65, 81, 87, 107], "ilay": 60, "analogu": 60, "goal": [60, 64, 94], "registernodeconversionpattern": [60, 88], "helper": 60, "pattern": [60, 75, 88], "schema": [60, 88, 92], "caus": [60, 64, 79, 98, 99, 101, 110], "acthardtanh": 60, "torchtrt_unus": 60, "hardtanh": [60, 69], "scalar": [60, 69], "min_val": [60, 69], "max_val": [60, 69], "unwraptodoubl": 60, "new_lay": 60, "addactiv": 60, "activationtyp": [60, 65], "kclip": 60, "torchtrt_check": 60, "unabl": [60, 88, 92], "setalpha": 60, "setbeta": 60, "setnam": [60, 88], "util": [60, 62, 73, 76, 88, 90, 95, 98, 101, 103, 104, 105, 106, 107, 108, 112], "node_info": [60, 88], "c_str": [60, 88], "out_tensor": [60, 88], "associatevalueandtensor": [60, 88], "getoutput": [60, 88], "log_debug": 60, "getdimens": [60, 88], "accord": [60, 64, 76], "unwrap": 60, "tool": [60, 64, 65, 66, 88, 94, 107], "don": [60, 65, 79, 81, 82, 90, 105, 108, 109], "annot": [60, 88], "your": [60, 63, 64, 66, 67, 74, 79, 81, 82, 86, 87, 88, 89, 91, 94, 109, 110], "Its": [60, 81], "track": [60, 90], "sort": [60, 69, 91], "live": [60, 81], "directli": [60, 62, 63, 66, 68, 73, 75, 90, 92, 95, 98, 106, 111], "associatevalueandivalu": 60, "inspect": [60, 87, 88], "dataflow": [60, 88], "mechan": [60, 64, 65, 92, 97, 107], "safe": [60, 64, 70, 74, 75, 76], "unsur": 60, "deep": [60, 64, 68, 79, 90, 92, 113], "straight": 60, "chanc": 60, "none": [60, 64, 65, 69, 70, 71, 73, 74, 75, 76, 79, 81, 92, 94, 98, 105], "wrapper": [60, 65, 111], "similar": [60, 63, 64, 65, 66, 88, 91, 92, 103, 104], "tocustomclass": 60, "tensorcontain": 60, "istensor": 60, "iscustomclass": 60, "lot": [60, 63], "singular": 60, "becaus": [60, 65, 66, 71, 87, 88, 92, 93, 94], "alloc": 60, "freed": 60, "destructor": 60, "destroi": [60, 82], "realli": 60, "think": [60, 81], "becom": [60, 66, 96], "benefit": [60, 88, 94], "deal": [60, 94], "quit": [60, 66, 88, 107], "effici": 60, "batch_norm": [60, 69], "fusion": [60, 62, 65], "deeplearn": [61, 65], "sdk": [61, 112], "matrix": 61, "html": [61, 66, 81, 87, 90, 92, 109], "c_api": 61, "python_api": 61, "org": [61, 66, 79, 81, 87, 88, 90, 92, 109, 110], "stabl": [61, 76, 77, 79, 95, 106, 109], "master": [61, 66, 90, 110], "overview": [61, 68, 98, 102], "md": 61, "appli": [62, 63, 90, 97], "desir": [62, 70, 82, 90, 94], "coalesc": 62, "insert": [62, 88, 90, 94, 97], "graphmodul": [62, 63, 70, 71, 75, 88, 89, 92, 97, 111, 112], "caller": 62, "invok": [62, 64, 65, 87, 88, 110], "lint": 62, "recompil": [62, 70, 75, 94, 97, 101, 109, 112], "repair": 62, "disallow": 62, "repair_input_as_output": 62, "gm": [62, 70], "sample_input": [62, 65, 98], "scenario": [62, 64, 96], "clone": [62, 66, 69, 92], "modified_graph": 62, "extract": [62, 88, 107], "placehold": 62, "isinst": [62, 65, 92, 105], "issubclass": 62, "direct": [62, 85, 97, 110], "len": [62, 69, 92], "direct_output": 62, "inserting_aft": 62, "cloned_placehold": 62, "replace_input_with": 62, "date": [62, 82, 112], "eliminate_dead_cod": 62, "logger": [62, 72], "f": [62, 64, 65, 75, 81, 87, 92, 105], "__init__": [62, 74, 75, 81, 87, 92, 94, 98, 105, 109], "pass_manag": 62, "passmanag": 62, "backend": [62, 67, 76, 77, 80, 91, 93, 94, 95, 98, 100, 101, 105, 106, 109], "offer": [62, 64], "registr": [62, 65], "conveni": [62, 90, 101, 107, 110, 112], "control": [62, 65, 87, 97, 110], "_aten_lowering_pass": 62, "my_custom_pass": 62, "front": [62, 70], "passlist": 62, "arbitrari": [62, 74], "remov": [62, 63, 70, 79, 93, 94, 105], "dump_lowering_pass": 62, "apply_lowering_pass": 62, "graph_modul": [62, 70], "_remove_lowering_pass": 62, "evolv": 62, "introduc": [63, 65, 107], "exportedprogram": [63, 67, 70, 75, 97, 103, 104, 109, 112], "dynamo": [63, 64, 66, 67, 73, 74, 75, 77, 88, 92, 93, 94, 97, 98, 99, 101, 102, 105, 109], "frontend": [63, 70, 73, 89, 92, 95, 99, 101, 105, 106, 109], "simpl": [63, 64, 65, 81, 82, 87, 107, 108, 109], "usag": [63, 65, 73, 77, 81, 88, 95, 106, 109, 111], "eval": [63, 67, 88, 89, 93, 94, 96, 97, 98, 99, 101, 102, 103, 104, 105, 108, 109, 111], "exp_program": [63, 94, 97, 105, 109], "trt_gm": [63, 67, 94, 97, 109, 111], "interact": [63, 81, 96, 98, 99, 100, 101, 102, 103, 104], "ideal": 63, "discuss": [63, 64, 108], "section": [63, 65, 79, 81, 82, 83, 85, 88, 90, 108, 111], "frequent": 63, "builder": [63, 64, 65, 70], "respect": [63, 66, 75], "releas": [63, 64, 81, 95, 106], "insid": [63, 81, 92, 95, 106, 108], "decomposit": [63, 64, 70, 75, 92], "downstream": [63, 107], "constraint": 63, "guid": [64, 80, 106], "present": [64, 97], "learn": [64, 66, 68, 88, 90, 92, 108, 113], "acceler": [64, 71, 75, 95, 106, 110, 112, 113], "workflow": [64, 65, 67, 68, 70, 71, 75, 88, 91, 94, 96, 99, 100, 101, 103, 104, 107], "wide": [64, 75, 85], "varieti": [64, 108], "primari": [64, 94, 111], "simplic": 64, "optimized_model": [64, 67, 93, 98, 99, 101], "depth": [64, 79, 107], "challeng": [64, 96, 108], "addition": [64, 92], "fit": [64, 81], "compilationset": [64, 70, 74, 92, 98], "_enum": [64, 70], "callabl": [64, 70, 75], "pass_through_build_failur": [64, 70, 74, 75, 92], "max_aux_stream": [64, 70, 74, 75, 92], "version_compat": [64, 70, 74, 75, 92], "optimization_level": [64, 70, 74, 75, 92, 98], "use_python_runtim": [64, 70, 74, 75, 92, 93, 94, 96, 97, 98], "truncate_doubl": [64, 70, 74, 75, 92, 93, 103, 104], "use_fast_partition": [64, 70, 74, 75, 92], "enable_experimental_decomposit": [64, 70, 74, 75, 92], "_devic": [64, 70], "assume_dynamic_shape_support": [64, 70, 74, 75], "make_refit": [64, 70, 74, 75, 93, 94, 96, 97], "engine_cap": [64, 70, 74, 75, 92], "dryrun": [64, 70, 74, 75, 92], "hardware_compat": [64, 70, 74, 75, 92], "timing_cache_path": [64, 70, 74, 75, 94], "tmp": [64, 70, 74, 75, 88, 93], "torch_tensorrt_engine_cach": [64, 70, 74, 75], "timing_cach": [64, 65, 70, 74, 75], "bin": [64, 66, 70, 74, 75], "lazy_engine_init": [64, 70, 74, 75], "cache_built_engin": [64, 70, 74, 93, 94], "reuse_cached_engin": [64, 70, 74, 93, 94], "dpython": [64, 70, 75, 76], "per": [64, 70, 92, 110], "regardless": [64, 70, 82, 99, 101], "fail": [64, 70, 75, 88, 96, 97, 105, 113], "auxiliari": [64, 70], "stream": [64, 70, 75, 92], "impli": [64, 70], "longer": [64, 66, 70, 75, 79, 110], "search": [64, 68, 70, 75, 79], "strictli": [64, 70], "runtim": [64, 66, 67, 68, 70, 75, 88, 96, 98, 101, 102, 112], "presenc": [64, 70], "preferenti": [64, 70], "choos": [64, 65, 70, 87], "float64": [64, 70, 75, 76], "refitt": [64, 70], "toggl": [64, 70, 75], "mode": [64, 65, 70, 74, 75, 89, 90, 102, 105], "detail": [64, 65, 70, 87, 88, 92, 94, 108, 110], "natur": [64, 70, 81], "architectur": [64, 66, 68, 70, 75, 94, 107], "amper": [64, 70, 75], "newer": [64, 66, 70, 75], "storag": [64, 70, 90], "sub": [64, 69, 81, 87, 98], "slate": 64, "futur": [64, 65, 70, 75, 76, 110], "occur": 64, "first_output": 64, "subsequ": [64, 94], "second_output": 64, "session": [64, 67, 81, 94, 102], "point": [64, 66, 75, 79, 80, 81, 88, 105, 108], "cover": [64, 106, 107], "benchmark": [64, 69], "automat": [64, 75, 81, 88, 97, 109, 112], "vari": [64, 71, 109], "distribut": [64, 88, 90, 110], "inf": 64, "dynamo_convers": 64, "contribut": 64, "demonstr": [64, 81, 82, 83, 90, 92, 94, 95, 96, 105, 106, 107, 108], "break": [64, 65, 70, 74, 75, 81, 92], "successfulli": [64, 96, 97], "_dynamo": [64, 93, 94, 98, 99, 101, 109], "explain": [64, 65, 68], "veri": [64, 65, 82, 83, 90, 91, 103, 104, 108], "explan": [64, 65], "graph_break_count": 64, "furthermor": 64, "durat": [64, 81], "latter": [64, 74], "logic": [64, 65], "guard": 64, "compos": [65, 87, 90, 105, 108], "variou": [65, 113], "etc": [65, 79, 81, 92, 113], "environ": [65, 67, 108], "research": 65, "few": [65, 66, 75], "nightli": 65, "lower_exampl": 65, "welcom": [65, 88], "finish": 65, "converison": 65, "pleas": [65, 75, 81, 88, 105, 108, 109], "max_batch_s": [65, 71, 108], "2048": [65, 71], "max_workspace_s": [65, 71], "33554432": [65, 71], "explicit_batch_dimens": [65, 71], "lower_precis": [65, 71], "lowerprecis": [65, 71], "verbose_log": [65, 71], "timing_cache_prefix": [65, 71], "save_timing_cach": [65, 71], "cuda_graph_batch_s": [65, 71], "dynamic_batch": [65, 71], "turn": [65, 71, 102], "trtmodul": [65, 71], "otherwis": [65, 66, 71, 94, 110], "implicit": [65, 69, 71, 81], "config": [65, 66, 71, 108], "updat": [65, 66, 70, 71, 75, 92, 95, 97, 106], "dim": [65, 69, 71, 92, 94, 105, 108, 109], "fx2trt_exampl": 65, "acc_trac": 65, "come": [65, 66, 80, 92, 96, 108], "my_pytorch_model": 65, "build_model": 65, "prepar": [65, 108], "acc_mod": 65, "earli": [65, 97], "deprec": [65, 69], "continu": [65, 81, 110], "backward": [65, 74, 92, 112], "vision": [65, 108], "activ": [65, 74, 76, 81, 88, 90, 107, 110, 113], "except": [65, 70, 75], "permut": [65, 69], "transpos": [65, 69, 109], "ll": [65, 94], "inputtensorspec": [65, 71, 75], "experiment": [65, 75, 76], "dataclass": [65, 98], "re": [65, 75, 81, 94, 96, 102, 110], "manual": [65, 75, 80, 81, 97], "sampl": [65, 70, 81, 89, 90, 96, 97, 98, 99, 100, 101, 102, 103, 104, 108], "rand": [65, 88, 94, 96, 97, 98], "from_tensor": [65, 75], "slightli": [65, 92], "promis": 65, "optimize_target_shap": 65, "input_tensor_spec": 65, "shape_rang": [65, 71], "100": [65, 71, 92, 94, 105], "accordingli": [65, 79, 109, 110], "trtinterpreterresult": [65, 71], "namedtupl": 65, "input_nam": [65, 71], "output_nam": [65, 71], "serialized_cach": [65, 71], "bytearrai": [65, 74, 76], "afford": 65, "temporari": [65, 94], "best": [65, 70, 75, 81, 96], "perforamnc": 65, "examin": 65, "suitabl": 65, "force_fp32_output": 65, "strict_type_constraint": 65, "usual": [65, 66, 79], "unless": 65, "certain": [65, 66, 98, 110], "algorithm_selector": 65, "profiling_verbos": 65, "trt_interpreter_result": 65, "64": [65, 75, 89, 104, 105, 109], "25": [65, 71, 88], "runtimeerror": [65, 105], "xxx": 65, "One": [65, 81, 82, 88, 107, 110], "reload_trt_mod": 65, "reload_model_output": 65, "far": [65, 81], "give": [65, 79, 81], "convtert": 65, "scheme": [65, 70, 75], "action": [65, 81], "tensort": [65, 112], "thing": [65, 66, 81], "compar": [65, 70, 75, 89, 97], "vanilla": 65, "mainli": 65, "builtin": 65, "purpos": [65, 107, 108], "acc_op": 65, "leverag": [65, 90, 95, 106], "power": [65, 81, 88, 107], "goe": [65, 81], "whole": 65, "sigmoid": [65, 69], "tensorrt_convert": 65, "acc_ops_sigmoid": 65, "rest": [65, 81, 82], "input_v": 65, "receiv": 65, "region": 65, "add_activ": 65, "get_output": [65, 92], "wherev": 65, "rememb": [65, 66], "mapper": 65, "todo": [65, 79], "logist": 65, "down": [65, 66, 79], "happen": [65, 87, 96, 99, 109], "acc_norm": 65, "foo": [65, 81, 82], "register_acc_op": 65, "register_acc_op_map": 65, "this_arg_is_opt": 65, "op_and_target": 65, "arg_replacement_tupl": 65, "rule": [65, 66, 76], "third": [65, 82], "boolean": [65, 75], "matter": [65, 92], "register_custom_acc_mapper_fn": 65, "design": [65, 73, 96, 107, 113], "redund": 65, "throught": 65, "custom_mapp": 65, "_": [65, 81, 92, 105], "foo_kwarg": 65, "inserting_befor": 65, "foo_nod": 65, "meta": [65, 85, 104], "children": 65, "unit": [65, 75], "test_acc_trac": 65, "acc_op_convert": 65, "essenti": 65, "plugin": [65, 92, 95, 106], "yet": [65, 107], "folder": 65, "center": 66, "pypi": 66, "m": [66, 82, 105], "pip": [66, 108], "upload": [66, 108], "x86": [66, 110], "extra": [66, 74, 88, 92, 96], "url": [66, 79, 108], "download": [66, 85, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 108], "whl": 66, "cu118": 66, "cu124": 66, "tarbal": [66, 88, 90], "easiest": [66, 92], "bazelisk": 66, "bazelbuild": 66, "export": [66, 68, 70, 75, 94, 95, 97, 99, 103, 104, 105, 106, 111, 112], "bazel_vers": 66, "path_to_torchtrt_root": 66, "bazelvers": 66, "mkdir": 66, "cd": [66, 108], "curl": [66, 81], "fssl": 66, "o": [66, 81, 108], "dist": 66, "unzip": 66, "bash": 66, "sh": 66, "cp": [66, 92], "usr": 66, "driver": 66, "branch": 66, "4e5b0f6e860910eb510fa70a76ee3eb9825e7a4d": 66, "l46": 66, "pull": [66, 94, 108], "latest": [66, 79], "l53c1": 66, "fact": 66, "reproduc": 66, "l71": 66, "http_archiv": 66, "build_fil": 66, "archiv": 66, "sha256": 66, "strip_prefix": 66, "OR": 66, "TO": [66, 88], "gnu": 66, "tar": [66, 81, 90], "gz": [66, 81, 82, 90], "ld_library_path": 66, "comment": [66, 81], "uncom": 66, "l114c1": 66, "l124c3": 66, "wheel": 66, "dep": 66, "lighter": 66, "executor": 66, "avoid": [66, 92, 97, 109], "implic": 66, "python_onli": 66, "legaci": [66, 73], "mainten": 66, "torchdynamo": [66, 109, 112], "technolog": [66, 112], "project": [66, 80, 85], "exclud": [66, 92], "speed": [66, 94, 95, 97, 106], "no_torchscript": 66, "dbg": 66, "pre_cxx11_abi": 66, "complic": 66, "incompat": 66, "popular": [66, 80, 107], "ngc": [66, 108], "tabl": [66, 85], "bdist_wheel": 66, "preinstal": 66, "forum": 66, "correctli": [66, 92], "declar": 66, "intend": [66, 98, 99, 100, 101, 102, 103, 104], "microsoft": 66, "2022": [66, 68], "open": [66, 107, 108], "app": 66, "x64": 66, "prompt": [66, 96, 100, 103, 104], "admin": 66, "privileg": 66, "launcher": 66, "chocolatei": 66, "navig": [66, 79], "ninja": 66, "setuptool": 66, "r": [66, 81], "txt": 66, "distutils_use_sdk": 66, "cuda_win": 66, "libtorch_win": 66, "tensorrt_win": 66, "non": [66, 75, 82, 84, 110], "similarli": [66, 94, 102, 110], "ci_workspac": 66, "win": 66, "tmpl": 66, "torchtrtc": [66, 68, 113], "websit": 66, "finder": 66, "dcmake_module_path": 66, "doesn": [66, 81, 87, 88], "dtorch_dir": 66, "dtensorrt_root": 66, "choic": [66, 73], "b": [66, 69, 75, 82, 108], "dcmake_build_typ": 66, "72048": 66, "jp_workspac": 66, "new_local_repositori": 66, "sudo": 66, "home": 66, "unlik": [66, 91], "libtorch_pre_cxx11_abi": 66, "shift": [66, 69, 81], "jetpack": 66, "jetpack_x": 66, "jetpack_5": 66, "drop": [66, 79, 105], "anywher": 67, "ahead": [67, 68, 88, 96], "ep": [67, 69, 97, 111], "output_format": [67, 75, 111], "input_tensor": [67, 92, 105], "fill": 67, "aot": [68, 88, 96, 97, 112], "integr": [68, 95, 96, 98, 106], "seamlessli": [68, 75], "ecosystem": [68, 112], "hybrid": [68, 70, 75, 76, 112], "instal": [68, 85, 88, 108, 110], "triton": [68, 92], "page": [68, 83, 85, 108], "introductori": 68, "blog": [68, 110], "gtc": 68, "2020": [68, 88], "talk": 68, "fall": [68, 75, 92], "2021": 68, "dai": 68, "confer": 68, "_convolut": [69, 88], "stride": [69, 75, 92, 105], "pad": [69, 75, 92, 105], "dilat": 69, "output_pad": 69, "group": [69, 81, 82], "determinist": 69, "cudnn_en": 69, "allow_tf32": 69, "ab": 69, "aco": 69, "acosh": 69, "adaptive_avg_pool1d": 69, "output_s": 69, "adaptive_avg_pool2d": 69, "adaptive_avg_pool3d": 69, "adaptive_max_pool1d": 69, "adaptive_max_pool2d": 69, "adaptive_max_pool3d": 69, "argmax": 69, "keepdim": 69, "argmin": 69, "asin": 69, "asinh": 69, "atan": 69, "atanh": 69, "avg_pool1d": 69, "kernel_s": [69, 92, 105], "ceil_mod": 69, "count_include_pad": 69, "avg_pool2d": 69, "divisor_overrid": 69, "avg_pool3d": 69, "gamma": 69, "var": 69, "momentum": 69, "bitwise_not": 69, "bmm": 69, "ceil": 69, "clamp": 69, "clamp_max": 69, "clamp_min": 69, "constant_pad_nd": 69, "co": [69, 82, 107], "cosh": 69, "cumsum": 69, "tensor_mod": 69, "rounding_mod": 69, "div_": 69, "elu": 69, "scale": [69, 90, 107], "input_scal": 69, "indic": [69, 79, 81, 97, 99, 109], "padding_idx": 69, "eq": [69, 81], "erf": 69, "exp": 69, "expand_a": 69, "fake_quantize_per_channel_affin": 69, "zero_point": 69, "axi": [69, 75], "quant_min": 69, "quant_max": 69, "fake_quantize_per_tensor_affin": 69, "using_int": [69, 88], "start_dim": [69, 88], "end_dim": [69, 88], "floor": 69, "floor_divid": 69, "ge": 69, "gru_cel": 69, "hx": 69, "w_ih": 69, "w_hh": 69, "b_ih": 69, "b_hh": 69, "gt": 69, "hardtanh_": 69, "instance_norm": 69, "running_mean": 69, "running_var": 69, "use_input_stat": 69, "layer_norm": 69, "normalized_shap": 69, "le": 69, "negative_slop": 69, "01": [69, 82, 88, 105], "leaky_relu_": 69, "lstm_cell": 69, "lt": 69, "masked_fil": 69, "mask": [69, 92], "max_pool1d": 69, "max_pool2d": [69, 87, 88], "max_pool3d": 69, "mul_": 69, "narrow": 69, "neg": [69, 96], "norm": 69, "scalaropt_dim": 69, "pixel_shuffl": 69, "upscale_factor": 69, "pow": 69, "tensor_scalar": 69, "expon": 69, "tensor_tensor": 69, "prelu": 69, "prod": [69, 92], "dim_int": 69, "reciproc": 69, "reflection_pad1d": 69, "reflection_pad2d": 69, "relu_": 69, "repeat_interleav": 69, "self_int": 69, "replication_pad1d": 69, "replication_pad2d": 69, "replication_pad3d": 69, "reshap": [69, 92, 108], "roll": 69, "rsub": 69, "scatter": 69, "sigmoid_": 69, "sin": [69, 81], "sinh": 69, "slice": 69, "split_siz": 69, "split_with_s": 69, "sqrt": 69, "squar": 69, "squeez": [69, 107], "sub_": 69, "dim_intlist": 69, "tan": 69, "tanh": 69, "tanh_": 69, "non_block": [69, 105], "memory_format": [69, 75], "prim_devic": 69, "topk": 69, "k": [69, 90, 105], "largest": 69, "dim0": [69, 94], "dim1": 69, "unbind": 69, "unsqueez": 69, "upsample_bilinear2d": 69, "align_corn": 69, "scales_h": 69, "scales_w": 69, "vec": 69, "scale_factor": 69, "upsample_linear1d": 69, "upsample_nearest1d": 69, "upsample_nearest2d": 69, "upsample_nearest3d": 69, "scales_d": 69, "upsample_trilinear3d": 69, "view": [69, 79], "__and__": 69, "__derive_index": 69, "idx": 69, "__getitem__": 69, "__is__": 69, "t1": 69, "t2": 69, "obj": 69, "__isnot__": 69, "__not__": 69, "__or__": 69, "__range_length": 69, "lo": 69, "hi": [69, 81, 82], "__round_to_zero_floordiv": 69, "__xor__": 69, "append": [69, 93, 94, 105], "el": 69, "arang": [69, 92], "pin_memori": 69, "start_step": 69, "copy_": 69, "float_int": 69, "int_float": 69, "floordiv": 69, "is_floating_point": 69, "numel": 69, "l": [69, 105], "9223372036854775807": 69, "requires_grad": 69, "tupleindex": 69, "tup": 69, "exported_program": [70, 75, 111], "arg_input": [70, 75, 97], "kwarg_input": [70, 75, 97], "engine_cache_dir": [70, 93, 94], "engine_cache_s": [70, 93, 94], "custom_engine_cach": [70, 94], "baseenginecach": [70, 94], "int32": [70, 75, 76, 92, 93, 101, 107], "channel_last": [70, 75, 76, 107], "244": [70, 75, 76], "alia": [70, 75], "better": [70, 75, 87, 107, 112], "understand": [70, 75, 109], "convolut": [70, 75, 76, 90, 92, 113], "_c": [70, 75, 76, 91], "oppos": [70, 75, 76], "lean": [70, 75], "spend": [70, 75], "integ": [70, 75, 84], "faster": [70, 75, 93, 94, 107], "parition": [70, 75], "increas": [70, 75, 94], "amount": [70, 75], "defer": [70, 75, 112], "lead": [70, 75, 81, 110], "oversubscript": [70, 75], "hard": [70, 97], "disk": [70, 75, 94], "space": [70, 81, 82, 90], "byte": [70, 74, 75, 76, 92, 94, 107], "1gb": [70, 93, 94], "exce": 70, "oldest": 70, "gear": [70, 90], "toward": [70, 90], "refit_module_weight": [70, 97], "compiled_modul": [70, 97], "new_weight_modul": [70, 97], "verify_output": [70, 97], "use_weight_map_cach": [70, 97], "in_plac": [70, 97], "compmil": 70, "coverag": [70, 92], "min_acc_module_s": 71, "is_aten": 71, "use_experimental_fx_rt": 71, "correctness_atol": 71, "correctness_rtol": 71, "minim": [71, 90, 92], "submodul": [71, 87, 92], "fx2trt": 71, "cpu": [71, 103, 104], "has_batch_dim": 71, "dtyep": 71, "prop": 71, "min_input_shap": 71, "optimized_input_shap": 71, "max_input_shap": 71, "popul": 71, "225": [71, 108], "explicit_precis": 71, "logger_level": 71, "model_trt": 72, "model_torchtrt": 72, "internal_error": 72, "toolkit": 73, "dataloadercalibr": [73, 90], "preprocess": [73, 90, 108], "algo_typ": [73, 90], "calibrationalgo": [73, 90], "cachecalibr": [73, 90], "qualnam": [73, 75], "entropy_calibr": 73, "entropy_calibration_2": [73, 90], "legacy_calibr": 73, "minmax_calibr": 73, "set_multi_device_safe_mod": [74, 110], "_multidevicesafemodecontextmanag": 74, "impact": 74, "suppress": 74, "unsaf": 74, "trt_compiled_modul": 74, "torchtensorrtmodul": [74, 92], "encompass": [74, 76], "simpili": 74, "de": 74, "initi": [74, 75, 81, 97, 98, 99, 101, 102, 103, 104], "scriptmodul": [74, 75, 76, 88, 89, 111, 112], "overridden": [74, 75], "subclass": 74, "although": [74, 81], "recip": [74, 90], "afterward": 74, "former": 74, "care": 74, "hook": 74, "silent": 74, "get_extra_st": 74, "state_dict": [74, 75, 96], "set_extra_st": 74, "picklabl": 74, "pickl": [74, 92, 94], "load_state_dict": [74, 96, 105], "pythontorchtensorrtmodul": 74, "serialized_engin": [74, 76], "_set": [74, 98], "weight_name_map": 74, "trt_modul": 74, "engine_str": 74, "my_modul": 74, "current_devic": 74, "cudagraphs_validate_shap": 74, "versu": 74, "disable_profil": 74, "enable_profil": 74, "iprofil": 74, "spent": 74, "get_layer_info": 74, "request": [75, 88, 108], "decid": 75, "deseri": [75, 76, 88, 92], "retrac": 75, "strict": [75, 110], "valueerror": 75, "mutabletorchtensorrtmodul": [75, 95, 96, 106], "pytorch_model": 75, "regular": 75, "whenev": 75, "refit_gm": 75, "shape_mod": 75, "_shapemod": 75, "interv": 75, "notat": 75, "bound": 75, "torch_tensor": 75, "tracer": 75, "example_tensor": 75, "optimization_profile_field": 75, "classmethod": 75, "disable_memory_format_check": 75, "core_id": 75, "schedul": [75, 108], "use_default": 75, "try_to": 75, "anoth": [75, 81, 82, 87, 89, 97], "typeerror": 75, "unknown": 75, "succe": 75, "float_dtyp": 75, "failur": 75, "bf16": 75, "try_from": [75, 92], "complex128": 75, "16": [75, 85, 87, 88, 89, 99, 102], "brain": 75, "bfloat16": 75, "f64": 75, "f8": 75, "fp8": [75, 95, 106], "float8": 75, "i32": 75, "sign": [75, 108], "i64": 75, "u8": 75, "unsign": 75, "uint8": 75, "trt_dla": 75, "torchtrt_dla": 75, "_from": 75, "torchtrt_dla_ec": 75, "torchtrt_safety_ec": 75, "saefti": 75, "trt_dla_ec": 75, "standalon": [75, 81], "certifi": 75, "tf": 75, "torchtrt_linear": 75, "cdhw32": 75, "thirti": 75, "row": [75, 82], "spatial": 75, "31": [75, 88], "subscript": [75, 81], "chw16": 75, "sixteen": 75, "15": [75, 81, 85], "chw2": 75, "chw32": 75, "chw4": 75, "four": [75, 81, 82], "dhwc": 75, "equivi": 75, "channels_last_3d": 75, "dhwc8": 75, "eight": 75, "dla_hwc4": 75, "imag": [75, 90, 92, 96, 100, 105, 108], "roundup": 75, "elements": 75, "dla_linear": 75, "planar": 75, "hwc": 75, "channels_last": 75, "hwc16": 75, "hwc8": 75, "least": [75, 81, 82], "ishapelay": 76, "check_method_op_support": 76, "seriali": 76, "put_binding_nam": 76, "tensorrtcompilespec": [76, 91], "scriptclass": 76, "0x7f14103e9130": 76, "_jit_to_tensorrt": 76, "00": 77, "000": [77, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "total": [77, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "galleri": [77, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "mem": 77, "advanc": [77, 82, 90, 95, 106], "torch_compile_advanced_usag": [77, 98], "torch_compile_resnet_exampl": [77, 99], "diffus": [77, 95, 106], "torch_compile_stable_diffus": [77, 100], "transform": [77, 88, 90, 93, 95, 97, 103, 104, 105, 106, 108, 111], "torch_compile_transformers_exampl": [77, 101], "v0": [78, 108], "pytorch_sphinx_them": [79, 86], "conf": [79, 86], "html_theme_opt": 79, "canonical_url": 79, "analytics_id": 79, "logo_onli": 79, "display_vers": 79, "prev_next_buttons_loc": 79, "bottom": 79, "style_external_link": 79, "vcs_pageview_mod": 79, "collapse_navig": 79, "sticky_navig": [79, 83], "navigation_depth": 79, "includehidden": 79, "titles_onli": 79, "canon": 79, "rank": 79, "trail": 79, "slash": 79, "googl": 79, "analyt": 79, "With": [79, 81, 88, 90, 94, 108], "isn": [79, 81, 92], "shown": [79, 81, 88], "sidebar": [79, 85], "button": [79, 81], "icon": [79, 81], "extern": [79, 81], "display_github": 79, "display_gitlab": 79, "gitlab": 79, "bitbucket": 79, "bar": [79, 81], "www": [79, 81, 88, 90, 108], "sphinx": [79, 80, 81, 82, 86, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "en": 79, "toctre": 79, "lose": 79, "scroll": [79, 83], "unlimit": 79, "header": [79, 81, 82, 88, 108], "render": 79, "github_url": 79, "bitbucket_url": 79, "gitlab_url": 79, "left": [79, 81], "upon": [79, 98, 101], "rst": [79, 81], "visitor": 79, "revert": 79, "misbuild": 79, "show": [79, 81, 94, 100, 107], "properti": [79, 92], "stick": 79, "screen": 79, "vertic": [79, 81], "too": [79, 81, 82], "sticki": [79, 85], "nav": [79, 85], "At": [80, 97], "django": 80, "payment": 80, "dotpai": 80, "dotpayprovid": 80, "seller_id": 80, "pin": 80, "lock": 80, "lang": 80, "pl": 80, "polish": 80, "gatewai": 80, "transfer": 80, "purchas": 80, "item": [80, 82, 105], "param": 80, "seller": 80, "consult": 80, "ui": 80, "languag": [80, 81, 82, 87, 92, 108], "data_item_1": 80, "emphasi": 81, "strong": 81, "hyperlink": 81, "cross": 81, "uri": 81, "web": 81, "anonym": 81, "label": [81, 90, 105, 107, 108], "substitut": 81, "charact": 81, "exceedingli": 81, "ugli": 81, "problem": 81, "problemat": 81, "ext": [81, 82], "autodoc": [81, 82], "demo": [81, 90], "test_py_modul": [81, 85], "my": 81, "role": 81, "pep": 81, "287": 81, "rfc": 81, "2822": 81, "superscript": 81, "gui": 81, "taken": 81, "height": 81, "interfer": 81, "press": 81, "keyboard": 81, "mous": 81, "mmb": 81, "menuselect": 81, "seen": [81, 82], "whitespac": 81, "signific": [81, 92], "strang": 81, "hyphen": 81, "word": [81, 107], "adjust": 81, "width": [81, 107], "browser": 81, "window": 81, "sentenc": [81, 107], "suppli": [81, 97], "258": 81, "equat": 81, "x_": 81, "x_0": 81, "x_1": 81, "x_2": 81, "x_3": 81, "x_4": 81, "nabla": 81, "frac": 81, "theta": 81, "phi": 81, "restructuredtext": [81, 82], "parser": [81, 105], "colon": 81, "indent": 81, "literal_block": 81, "spaces_and_linebreak": 81, "preserv": [81, 87, 90], "markup_process": 81, "Or": 81, "great": [81, 88, 92, 94, 112], "why": [81, 110], "didn": 81, "blank": 81, "align": 81, "permit": 81, "awai": 81, "eric": 81, "orchestra": 81, "leader": 81, "bee": 81, "philosoph": 81, "ipso": 81, "facto": 81, "But": [81, 88, 97], "got": [81, 88], "vi": 81, "entiti": 81, "said": 81, "entir": [81, 112], "ancient": 81, "injuri": 81, "sing": 81, "elk": 81, "bracket": 81, "miss": [81, 88], "brontosaurus": 81, "thin": 81, "thicker": 81, "middl": 81, "That": [81, 88], "mine": 81, "belong": 81, "me": [81, 82], "ann": 81, "begun": 81, "cut": [81, 97], "past": 81, "pars": [81, 88], "someurl": 81, "dev0": 81, "8759736": 81, "caption": [81, 84], "pane": 81, "shell_command": 81, "echo": 81, "did": 81, "window_nam": 81, "session_nam": 81, "shorthand": 81, "some_funct": 81, "highlight": 81, "THE": 81, "heaven": 81, "hexagram": 81, "six": 81, "unbroken": 81, "primal": 81, "light": [81, 111], "spirit": 81, "weak": 81, "essenc": 81, "energi": 81, "unrestrict": 81, "conceiv": 81, "motion": 81, "regard": [81, 112], "basi": 81, "thu": 81, "persist": 81, "dual": 81, "sens": [81, 88], "univers": 81, "world": 81, "men": 81, "express": 81, "deiti": 81, "human": 81, "denot": [81, 92], "holi": 81, "man": [81, 82], "sage": 81, "ruler": 81, "who": 81, "awaken": 81, "utf": [81, 82], "sphinx_rtd_them": [81, 82], "docstr": [81, 82, 89], "dl": 81, "dt": 81, "tag": [81, 108], "tt": 81, "descnam": 81, "descclassnam": 81, "wrote": 81, "anyth": [81, 82, 110], "programm": 81, "myclass": 81, "dothismethod": 81, "flush": 81, "meth": 81, "capit": 81, "flox": 81, "unreferenc": 81, "nonexist": 81, "extrem": 81, "stuff": 81, "mayb": 81, "bold": 81, "ital": 81, "heck": 81, "put": [81, 107], "13": [81, 85], "backlink": 81, "knowledg": 81, "mind": 81, "ey": 81, "thought": 81, "medium": 81, "peopl": 81, "subsect": 81, "interpol": 81, "indirect": 81, "phrase": 81, "docutil": [81, 82], "sourceforg": [81, 82], "ref": 81, "clickabl": 81, "legend": 81, "revis": [81, 82, 96, 100], "revisit": 81, "enhanc": 81, "structuredtext": 81, "wooden": 81, "nickel": 81, "mad": 81, "scientist": 81, "bigger": 81, "bread": 81, "box": [81, 109, 112], "wash": 81, "behind": 81, "ear": 81, "room": 81, "closet": 81, "bathroom": 81, "trash": 81, "sink": 81, "mother": 81, "g_": 81, "mu": 81, "nu": 81, "pi": 81, "t_": 81, "rho_": 81, "servic": 81, "thing1": 81, "thing2": 81, "thing3": 81, "prose": 81, "provok": 81, "mental": 81, "exert": 81, "reader": 81, "discret": 81, "strongli": 81, "advis": 81, "subtitl": 81, "outsid": 81, "often": 81, "besid": 81, "border": 81, "background": [81, 87], "ok": [81, 88], "transmit": 81, "disconnect": 81, "nonetheless": 81, "semant": 81, "blue": [81, 92], "white": 81, "arab": 82, "roman": 82, "upper": 82, "iii": 82, "iv": 82, "classifi": [82, 87, 88, 105, 107], "paragraph": [82, 85], "z": 82, "commonli": 82, "vm": 82, "david": 82, "goodger": 82, "address": [82, 92, 96], "123": 82, "street": 82, "canada": 82, "a1b": 82, "2c3": 82, "contact": 82, "myself": 82, "organ": 82, "humankind": 82, "2012": 82, "03": 82, "19": [82, 85], "53": 82, "0000": 82, "tue": 82, "jan": 82, "progress": 82, "7302": 82, "wish": 82, "redistribut": 82, "reattribut": 82, "sell": 82, "bui": 82, "rent": 82, "leas": 82, "improv": [82, 110], "quot": 82, "excerpt": 82, "incorpor": 82, "collat": 82, "fold": 82, "stapl": 82, "mutil": 82, "anyon": 82, "heart": 82, "bibliograph": 82, "markup": [82, 85], "literal": 82, "yahoo": 82, "oh": 82, "liter": 82, "heh": 82, "child": 82, "beat": 82, "text": [82, 84, 107], "hehe": 82, "kept": 82, "sai": [82, 107], "cackl": 82, "night": 82, "lone": 82, "guangzhou": 82, "destini": 82, "hope": 82, "dream": 82, "forth": 82, "fifth": 82, "sixth": 82, "lorem": [82, 84], "ipsum": [82, 84], "dolor": [82, 84], "sit": [82, 84], "amet": [82, 84], "consectetur": [82, 84], "adipisc": [82, 84], "elit": [82, 84], "donec": [82, 84], "porttitor": [82, 84], "odio": [82, 84], "posuer": [82, 84], "vita": [82, 84], "ornar": [82, 84], "libero": [82, 84], "matti": 82, "loborti": [82, 84], "justo": [82, 84], "vestibulum": [82, 84], "nibh": [82, 84], "aliquet": [82, 84], "sed": [82, 84], "feugiat": [82, 84], "sagitti": [82, 84], "nequ": [82, 84], "qui": [82, 84], "eleifend": 82, "dui": [82, 84], "rutrum": [82, 84], "lectu": [82, 84], "suscipit": [82, 84], "letter": [82, 107], "column": 82, "cell": 82, "span": 82, "nam": [82, 84], "mauri": [82, 84], "arcu": [82, 84], "stub": 82, "behav": 83, "area": 83, "interdum": 84, "nec": 84, "finibu": 84, "dictum": 84, "velit": 84, "ut": 84, "eu": 84, "efficitur": 84, "aliquam": 84, "erat": 84, "diam": 84, "gravida": 84, "imperdiet": 84, "tellu": 84, "nisl": 84, "praesent": 84, "eget": 84, "elementum": 84, "rhoncu": 84, "tincidunt": 84, "suspendiss": 84, "volutpat": 84, "scelerisqu": 84, "tristiqu": 84, "aenean": 84, "condimentum": 84, "risu": 84, "accumsan": 84, "laoreet": 84, "maximu": 84, "sapien": 84, "ligula": 84, "fringilla": 84, "commodo": 84, "proin": 84, "et": 84, "pharetra": 84, "etiam": 84, "turpi": 84, "ant": 84, "luctu": 84, "vel": 84, "malesuada": 84, "dignissim": 84, "mi": 84, "nunc": 84, "augu": 84, "sem": 84, "cursu": 84, "nulla": 84, "pellentesqu": 84, "habit": 84, "morbi": 84, "senectu": 84, "netu": 84, "fame": 84, "ac": 84, "egesta": 84, "placerat": 84, "tortor": 84, "iaculi": 84, "venenati": 84, "cra": 84, "puru": 84, "ero": 84, "vehicula": 84, "fusc": 84, "auctor": 84, "phasellu": 84, "est": 84, "viverra": 84, "conval": 84, "faucibu": 84, "vulput": 84, "feli": 84, "sodal": 84, "maecena": 84, "congu": 84, "semper": 84, "enim": 84, "blandit": 84, "sollicitudin": 84, "urna": 84, "orci": 84, "lacu": 84, "quisqu": 84, "facilisi": 84, "hendrerit": 84, "curabitur": 84, "variu": 84, "bibendum": 84, "massa": 84, "magna": 84, "tempu": 84, "metu": 84, "nisi": 84, "pretium": 84, "leo": 84, "euismod": 84, "ultric": 84, "dapibu": 84, "lacinia": 84, "vivamu": 84, "molesti": 84, "hac": 84, "habitass": 84, "platea": 84, "dictumst": 84, "git": 85, "content": [85, 90, 108], "changelog": 85, "math": 85, "9": [85, 88, 92, 108], "14": [85, 93, 101, 108], "17": 85, "18": [85, 88, 96], "submenu": 85, "symlink": 86, "subtre": 86, "_theme": 86, "html_theme": 86, "html_theme_path": 86, "optimiz": 87, "tutori": [87, 90, 92, 94, 96, 97], "beginn": 87, "intro_to_torchscript_tutori": 87, "briefli": 87, "lenet": [87, 88], "lenetfeatextractor": 87, "conv1": [87, 88], "conv2d": [87, 92, 105], "conv2": [87, 88], "lenetclassifi": 87, "fc1": [87, 88], "120": [87, 88], "fc2": [87, 88], "84": [87, 88], "fc3": [87, 88], "feat": [87, 88], "obvious": 87, "pathwai": 87, "input_data": [87, 89], "traced_model": 87, "pick": 87, "script_model": [87, 91], "perspect": 87, "___torch_mangle_10": 87, "129": 87, "___torch_mangle_9": 87, "119": 87, "___torch_mangle_5": 87, "137": 87, "callmethod": 87, "138": 87, "38": 87, "39": 87, "torch_script_modul": [87, 88], "in_tensor": 87, "fly": 87, "lenet_script": [87, 88], "haven": 88, "acquir": 88, "dyanmo": 88, "almost": [88, 112], "trt_lenet_script": 88, "apr": 88, "56": 88, "04": [88, 108], "credit": 88, "stop": 88, "argc": 88, "argv": 88, "cerr": 88, "cout": 88, "even": [88, 96], "cppdoc": 88, "pretti": 88, "fashion": [88, 107], "enable_precis": 88, "And": 88, "convertgraphtotrtengin": 88, "engine_converted_from_jit": 88, "close": 88, "saw": 88, "576": 88, "346": 88, "539": 88, "0464": 88, "0383": 88, "0678": 88, "0932": 88, "1045": 88, "0805": 88, "0435": 88, "0818": 88, "0208": 88, "0358": 88, "cudafloattyp": 88, "0530": 88, "1691": 88, "2802": 88, "1502": 88, "1056": 88, "1549": 88, "input0": [88, 89], "1063": 88, "input1": [88, 89], "input2": 88, "28": 88, "29": 88, "33": 88, "35": 88, "36": 88, "37": 88, "compilegraph": [88, 90], "laid": 88, "translat": [88, 97], "aren": 88, "techniqu": [88, 90, 110], "checkmethodoperatorsupport": 88, "modular": 88, "ship": [88, 110], "exhaust": 88, "109": 88, "addlay": 88, "yourself": 88, "question": 88, "outself": 88, "flatten_convert": 88, "unwraptoint": 88, "in_shap": 88, "tovec": 88, "out_shap": 88, "shuffl": [88, 90, 105], "addshuffl": 88, "setreshapedimens": 88, "todim": 88, "extens": [88, 112], "ctype": 88, "cdll": 88, "contributor": 88, "upstream": 88, "pr": 88, "usecas": [89, 106], "sole": [89, 90, 112], "individu": 89, "accuraci": [90, 107], "loss": [90, 107], "infrastructur": [90, 108], "streamlin": 90, "expos": [90, 92], "cpp_frontend": 90, "loading_data_recip": 90, "cifar10": [90, 105], "cstddef": 90, "ktrain": 90, "ktest": 90, "un": 90, "cs": 90, "toronto": 90, "edu": 90, "kriz": 90, "cifar": 90, "is_train": 90, "trim": 90, "use_subset": 90, "new_siz": 90, "mode_": 90, "images_": 90, "targets_": 90, "calibration_dataset": 90, "data_dir": 90, "320": 90, "4914": [90, 105], "4822": [90, 105], "4465": [90, 105], "2023": [90, 105], "1994": [90, 105], "2010": [90, 105], "dataloaderopt": 90, "worker": 90, "simpler": 90, "virtual": 90, "input_shap": [90, 113], "compile_spec": [90, 99, 113], "kf16": [90, 113], "ki8": 90, "vgg16": [90, 95, 105, 106], "testing_dataset": [90, 105], "totensor": [90, 105, 108], "testing_dataload": [90, 105], "num_work": [90, 105], "vgg": [90, 105], "test_ptq_dataloader_calibr": 90, "test_ptq_trt_calibr": 90, "krizhevski": 90, "hinton": 90, "2009": 90, "tini": 90, "simonyan": 90, "zisserman": 90, "2014": 90, "recognit": [90, 107], "arxiv": 90, "preprint": 90, "1409": 90, "1556": 90, "_jit_to_backend": 91, "mobilenet_v2": 91, "pretrain": [91, 94, 96, 99, 102, 107, 108], "cost": [92, 94, 97, 110], "perhap": 92, "overhead": [92, 110], "sake": 92, "circular": 92, "red": 92, "green": 92, "twice": 92, "written": 92, "openai": 92, "formal": 92, "tl": 92, "custom_op": 92, "circ_pad_kernel": 92, "all_pads_0": 92, "all_pads_2": 92, "all_pads_4": 92, "all_pads_6": 92, "orig_dims_0": 92, "orig_dims_1": 92, "orig_dims_2": 92, "orig_dims_3": 92, "y_shape_1": 92, "y_shape_2": 92, "y_shape_3": 92, "x_len": 92, "y_len": 92, "block_siz": 92, "pid": 92, "program_id": 92, "mask_i": 92, "i3": 92, "i2": 92, "i1": 92, "i0": 92, "j0": 92, "j1": 92, "j2": 92, "j3": 92, "load_idx": 92, "mask_x": 92, "launch": [92, 108], "torchtrt_ex": 92, "triton_circular_pad": 92, "mutates_arg": 92, "out_dim": 92, "tolist": 92, "all_pad": 92, "zero": 92, "orig_dim": 92, "blocksiz": 92, "256": [92, 105, 108], "numblock": 92, "ex_input": 92, "tracabl": 92, "prerequisit": 92, "fake": 92, "real": 92, "faketensor": 92, "register_fak": 92, "autograd": 92, "beyond": 92, "register_autograd": 92, "padded_x": 92, "my_model": 92, "2604": 92, "4232": 92, "3041": 92, "0833": 92, "2461": 92, "1270": 92, "2450": 92, "4079": 92, "2887": 92, "2828": 92, "0373": 92, "0332": 92, "3143": 92, "6344": 92, "5638": 92, "1867": 92, "5068": 92, "4363": 92, "7937": 92, "3488": 92, "1350": 92, "7966": 92, "3517": 92, "1379": 92, "5537": 92, "1088": 92, "8950": 92, "0550": 92, "6163": 92, "0109": 92, "5245": 92, "9632": 92, "5686": 92, "3775": 92, "8162": 92, "4216": 92, "4311": 92, "1649": 92, "2091": 92, "3668": 92, "1006": 92, "1447": 92, "0352": 92, "7689": 92, "8131": 92, "_run_on_gpu_0": 92, "_run_on_acc_1": 92, "dry": 92, "50": [92, 107], "count": 92, "__": 92, "were": [92, 97, 110], "aggreg": 92, "stat": 92, "latenc": [92, 110], "abstractli": 92, "pkl": [92, 96], "cupi": 92, "gap": 92, "prealloc": 92, "circularpaddingplugin": 92, "ipluginv2dynamicext": 92, "field_collect": 92, "pluginfieldcollect": 92, "x_shape": 92, "num_output": 92, "plugin_namespac": 92, "plugin_typ": 92, "plugin_vers": 92, "assert": [92, 96, 97], "get_output_datatyp": 92, "input_typ": 92, "get_output_dimens": 92, "output_index": 92, "dimsexpr": 92, "exprbuild": 92, "iexprbuild": 92, "output_dim": 92, "dimensionoper": 92, "configure_plugin": 92, "inp": 92, "dynamicplugintensordesc": 92, "x_dim": 92, "desc": 92, "supports_format_combin": 92, "po": 92, "in_out": 92, "plugintensordesc": 92, "num_input": 92, "enqueu": 92, "input_desc": 92, "output_desc": 92, "in_dtyp": 92, "a_mem": 92, "unownedmemori": 92, "items": 92, "c_mem": 92, "a_ptr": 92, "memorypoint": 92, "c_ptr": 92, "a_d": 92, "memptr": 92, "c_d": 92, "a_t": 92, "as_tensor": 92, "c_t": 92, "cloned_plugin": 92, "__dict__": 92, "circularpaddingplugincr": 92, "iplugincr": 92, "field_nam": 92, "pluginfield": 92, "pluginfieldtyp": 92, "create_plugin": 92, "pluginfieldcollection_": 92, "deserialize_plugin": 92, "pads_dict": 92, "creator": 92, "trt_plugin_registri": 92, "get_plugin_registri": 92, "register_cr": 92, "untyp": 92, "get_trt_tensor": 92, "set_layer_nam": 92, "recal": 92, "intlist": 92, "circular_padding_convert": 92, "retriev": 92, "elsewher": 92, "plugin_registri": 92, "plugin_cr": 92, "get_plugin_cr": 92, "field_config": 92, "eventu": 92, "freez": 92, "_input": 92, "add_plugin_v2": 92, "circular_padding_plugin": 92, "_run_on_acc_0": 92, "grad_fn": 92, "subbackward0": 92, "minut": [92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "custom_kernel_plugin": 92, "jupyt": [92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "ipynb": [92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "engine_caching_exampl": [93, 94], "remove_timing_cach": [93, 94], "bertmodel": [93, 101], "random": [93, 94, 96, 97], "seed": [93, 94, 96, 97], "manual_se": [93, 94, 96, 97], "from_pretrain": [93, 96, 100, 101, 103, 104], "uncas": [93, 101, 107], "return_dict": 93, "randint": [93, 101], "compile_bert": 93, "enable_tim": [93, 94], "1st": [93, 94], "measur": [93, 94], "2nd": [93, 94], "3rd": [93, 94], "slower": [93, 94], "messur": [93, 94], "compilation_kwarg": [93, 101], "torch_trt_bert_engine_cach": 93, "30": [93, 94, 96, 97, 99, 101], "synchron": [93, 94], "elapsed_tim": [93, 94], "millisecond": 93, "__name__": [93, 98, 101], "__main__": [93, 98, 101], "engine_caching_bert_exampl": 93, "paid": 94, "upfront": 94, "invalid": 94, "repeatedli": 94, "mitig": 94, "explor": [94, 106], "resnet18": [94, 96, 97, 99, 102], "torch_trt": [94, 96, 97], "_default": 94, "_engine_cach": 94, "flexibl": [94, 112], "histor": 94, "barrier": 94, "reconstruct": 94, "prior": [94, 109, 110], "ti": 94, "hash": 94, "magnitud": 94, "make_refitt": 94, "torch_compil": [94, 98, 99, 101, 102, 109, 112], "compiled_model": 94, "ms": 94, "dynamo_compil": 94, "example_input": 94, "200": 94, "dynamic_shap": [94, 109], "remot": 94, "systen": 94, "agnost": 94, "implent": 94, "ramenginecach": 94, "held": 94, "engine_cach": 94, "torch_compile_my_cach": 94, "cudagraph": [95, 106], "mutabl": [95, 106], "vgg16_fp8_ptq": [95, 106], "bert": [95, 101, 106], "gpt2": [95, 106], "llama2": [95, 106], "straightforward": 96, "especi": 96, "hug": [96, 103, 104], "face": [96, 103, 104], "difficult": 96, "ever": 96, "walk": [96, 97], "lora": [96, 97], "use_python": 96, "mutable_modul": 96, "model2": [96, 97], "expected_output": [96, 97], "refitted_output": [96, 97], "allclos": [96, 97], "reload": [96, 112], "checkpoint": [96, 105], "civitai": 96, "12597": 96, "moxin": 96, "diffusionpipelin": [96, 100], "no_grad": [96, 103, 104, 105], "model_id": [96, 100], "runwayml": 96, "v1": [96, 100], "hous": 96, "forest": 96, "shuimobysim": 96, "wuchangshuo": 96, "qualiti": 96, "worst": 96, "lowr": 96, "focu": 96, "cloudi": 96, "watermark": 96, "pipe": [96, 100], "torch_dtyp": [96, 100], "unet": [96, 100], "negative_prompt": 96, "num_inference_step": 96, "without_lora_mut": 96, "jpg": [96, 108], "procedur": 96, "load_lora_weight": 96, "stablediffusionapi": 96, "load_lora_embed": 96, "weight_nam": 96, "safetensor": 96, "adapter_nam": 96, "lora1": 96, "set_adapt": 96, "adapter_weight": 96, "fuse_lora": 96, "unload_lora_weight": 96, "with_lora_mut": 96, "mutable_torchtrt_module_exampl": 96, "expens": 97, "involv": 97, "occasion": [97, 98, 101], "adapt": 97, "infeas": 97, "focus": 97, "mostli": 97, "recogn": 97, "behalf": 97, "init": [97, 105], "sett": 97, "randomli": 97, "exp_program2": 97, "compiled_trt_ep": 97, "new_trt_gm": 97, "accomplish": 97, "gaurente": 97, "attempt": [97, 105, 109], "rebuild": 97, "heurist": 97, "refit_engine_exampl": 97, "x_out": 98, "y_out": 98, "x_y_out": 98, "invoc": 98, "sample_inputs_half": 98, "model_half": 98, "backend_kwarg": 98, "optimized_model_custom": 98, "exit": [98, 101, 108], "2052": [98, 101], "compile_engine_and_inf": [98, 101], "new_input": [99, 101], "new_output": [99, 101], "new_batch_size_input": 99, "new_batch_size_output": 99, "inputs_bs8": 99, "mark_dynam": [99, 109], "outputs_bs8": 99, "No": [99, 109], "inputs_bs12": 99, "outputs_bs12": 99, "compvi": 100, "majest": 100, "castl": 100, "cloud": 100, "majestic_castl": 100, "png": 100, "enable_cudagraph": [102, 110], "out_trt": 102, "set_cudagraphs_mod": [102, 110], "inputs_2": 102, "inputs_3": 102, "out_trt_2": 102, "out_trt_3": 102, "torch_export_cudagraph": 102, "automodelforcausallm": [103, 104], "autotoken": [103, 104], "export_llm": [103, 104], "max_token": [103, 104], "kv_cach": [103, 104], "token": [103, 104, 107], "pad_token_id": 103, "eos_token_id": [103, 104], "attn_implement": [103, 104], "eager": [103, 104], "model_input": [103, 104], "return_tensor": [103, 104], "input_id": [103, 104], "regress": [103, 104], "huggingfac": [103, 104, 107], "pyt_gen_token": [103, 104], "gpt2_ep": 103, "max_seq_len": [103, 104], "trt_gen_token": [103, 104], "skip_special_token": [103, 104], "torch_export_gpt2": 103, "llama_path": 104, "llama": 104, "7b": 104, "chat": 104, "hf": 104, "llama2_ep": 104, "batch_decod": 104, "clean_up_tokenization_spac": 104, "torch_export_llama2": 104, "argpars": 105, "modelopt": 105, "mtq": 105, "export_torch_mod": 105, "layer_spec": 105, "num_class": 105, "1000": [105, 108], "init_weight": 105, "in_channel": 105, "pool": [105, 113], "maxpool2d": 105, "batchnorm2d": 105, "sequenti": 105, "avgpool": 105, "adaptiveavgpool2d": 105, "4096": 105, "dropout": 105, "_initialize_weight": 105, "kaiming_normal_": 105, "fan_out": 105, "nonlinear": 105, "constant_": 105, "elif": 105, "normal_": 105, "vgg16_cfg": 105, "128": 105, "argumentpars": 105, "add_argu": 105, "ckpt": 105, "parse_arg": 105, "model_state_dict": 105, "device_count": 105, "ordereddict": 105, "new_state_dict": 105, "forget": 105, "training_dataset": 105, "randomcrop": 105, "randomhorizontalflip": 105, "training_dataload": 105, "drop_last": 105, "crit": 105, "crossentropyloss": 105, "calibrate_loop": 105, "pred": 105, "5f": 105, "acc": 105, "2f": 105, "quantize_typ": 105, "quant_cfg": 105, "int8_default_cfg": 105, "fp8_default_cfg": 105, "forward_loop": 105, "qdq": 105, "incomplet": 105, "functionaltensor": 105, "functionaltensormod": 105, "_trace": 105, "_export": 105, "float8_e4m3fn": 105, "class_prob": 105, "class_pr": 105, "test_prob": 105, "test_pr": 105, "test_loss": 105, "test_acc": 105, "vgg16_ptq": 105, "concept": 106, "_rendered_examples_python": 106, "_rendered_examples_jupyt": 106, "acoust": 107, "speech": 107, "quartznet": 107, "contextnet": 107, "subword": 107, "piec": 107, "excit": 107, "se": 107, "smaller": 107, "audio": 107, "transcrib": 107, "speedup": 107, "obtain": [107, 111], "feedforward": 107, "cnn": 107, "uniformli": 107, "resolut": 107, "highli": [107, 108], "compound": 107, "coeffici": 107, "b0": 107, "corpu": 107, "english": 107, "supervis": 107, "walkthrough": 107, "overal": 107, "jetson": 107, "adopt": 107, "mobilenetv2": 107, "classif": 107, "imagenet": 107, "imagenett": 107, "qat": 107, "simul": 107, "hand": 108, "consider": 108, "concurr": 108, "grpc": 108, "solv": 108, "aforement": 108, "familiar": 108, "resnet50": 108, "torchhub": 108, "docker": 108, "login": 108, "xx": 108, "yy": 108, "mm": 108, "publish": 108, "22": 108, "pwd": 108, "scratch_spac": 108, "nvcr": 108, "py3": 108, "proce": 108, "hub": 108, "_validate_not_a_forked_repo": 108, "suggest": 108, "simplest": 108, "model_repositori": 108, "pbtxt": 108, "pytorch_libtorch": 108, "input__0": 108, "data_typ": 108, "type_fp32": 108, "output__0": 108, "exact": 108, "encourag": 108, "rm": 108, "8000": 108, "8001": 108, "8002": 108, "the_model_repositori": 108, "tritonserv": 108, "spin": 108, "proceed": 108, "flesh": 108, "wget": 108, "img1": 108, "hakaimagazin": 108, "wp": 108, "gulf": 108, "bird": 108, "attrdict": 108, "pyindex": 108, "tritoncli": 108, "jump": 108, "firstli": 108, "resiz": 108, "pil": 108, "httpclient": 108, "triton_to_np_dtyp": 108, "rn50_preprocess": 108, "img_path": 108, "img": 108, "centercrop": 108, "485": 108, "456": 108, "406": 108, "229": 108, "transformed_img": 108, "inferenceservercli": 108, "localhost": 108, "secondli": 108, "inferinput": 108, "set_data_from_numpi": 108, "binary_data": 108, "inferrequestedoutput": 108, "class_count": 108, "lastli": 108, "send": 108, "model_nam": 108, "inference_output": 108, "as_numpi": 108, "468750": 108, "90": 108, "523438": 108, "92": 108, "664062": 108, "429688": 108, "136": 108, "234375": 108, "confidence_scor": 108, "classification_index": 108, "eagerli": 109, "swap": 109, "exactli": 109, "_tracer": 109, "sometim": 109, "queri": 109, "attn_weight": 109, "seq_len": 109, "compiler_dynamic_shap": 109, "inputs_bs2": 109, "libtorchtrt_runtim": 110, "dl_open": 110, "ld_preload": 110, "load_librari": 110, "cxx11": 110, "abi": 110, "wl": 110, "ltorchtrt": 110, "torchtrt_runtime_exampl": 110, "libtorchtrt_plugin": 110, "neglig": 110, "thread": 110, "alert": 110, "switch": 110, "mismatch": 110, "crash": 110, "sacrif": 110, "incur": 110, "intens": 110, "trt_ep": 111, "stai": 111, "trt_t": 111, "ergonom": 112, "deleg": 112, "believ": 112, "amen": 112, "artifact": 112, "pack": 112, "year": 112, "superset": 112, "codebas": 112, "immedi": 112, "traceabl": 112, "scriptabl": 112, "hardwar": 113, "neural": 113, "deconvolut": 113, "scripted_model": 113}, "objects": {"": [[5, 0, 1, "c.STR", "STR"], [9, 0, 1, "c.TORCHTRT_API", "TORCHTRT_API"], [11, 0, 1, "c.TORCHTRT_HIDDEN", "TORCHTRT_HIDDEN"], [7, 0, 1, "c.TORCH_TENSORRT_MAJOR_VERSION", "TORCH_TENSORRT_MAJOR_VERSION"], [8, 0, 1, "c.TORCH_TENSORRT_MINOR_VERSION", "TORCH_TENSORRT_MINOR_VERSION"], [6, 0, 1, "c.TORCH_TENSORRT_PATCH_VERSION", "TORCH_TENSORRT_PATCH_VERSION"], [12, 0, 1, "c.TORCH_TENSORRT_VERSION", "TORCH_TENSORRT_VERSION"], [10, 0, 1, "c.XSTR", "XSTR"], [0, 1, 1, "_CPPv4N14torch_tensorrt8DataTypeE", "torch_tensorrt::DataType"], [0, 2, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeE5Value", "torch_tensorrt::DataType::DataType"], [0, 2, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeEN3c1010ScalarTypeE", "torch_tensorrt::DataType::DataType"], [0, 2, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeEv", "torch_tensorrt::DataType::DataType"], [0, 3, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeE5Value", "torch_tensorrt::DataType::DataType::t"], [0, 3, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeEN3c1010ScalarTypeE", "torch_tensorrt::DataType::DataType::t"], [0, 4, 1, "_CPPv4N14torch_tensorrt8DataType5ValueE", "torch_tensorrt::DataType::Value"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kBoolE", "torch_tensorrt::DataType::Value::kBool"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kCharE", "torch_tensorrt::DataType::Value::kChar"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value7kDoubleE", "torch_tensorrt::DataType::Value::kDouble"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value6kFloatE", "torch_tensorrt::DataType::Value::kFloat"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kHalfE", "torch_tensorrt::DataType::Value::kHalf"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value4kIntE", "torch_tensorrt::DataType::Value::kInt"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kLongE", "torch_tensorrt::DataType::Value::kLong"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value8kUnknownE", "torch_tensorrt::DataType::Value::kUnknown"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kBoolE", "torch_tensorrt::DataType::kBool"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kCharE", "torch_tensorrt::DataType::kChar"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value7kDoubleE", "torch_tensorrt::DataType::kDouble"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value6kFloatE", "torch_tensorrt::DataType::kFloat"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kHalfE", "torch_tensorrt::DataType::kHalf"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value4kIntE", "torch_tensorrt::DataType::kInt"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kLongE", "torch_tensorrt::DataType::kLong"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value8kUnknownE", "torch_tensorrt::DataType::kUnknown"], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypecv5ValueEv", "torch_tensorrt::DataType::operator Value"], [0, 2, 1, "_CPPv4N14torch_tensorrt8DataTypecvbEv", "torch_tensorrt::DataType::operator bool"], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypeneE8DataType", "torch_tensorrt::DataType::operator!="], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypeneEN8DataType5ValueE", "torch_tensorrt::DataType::operator!="], [0, 3, 1, "_CPPv4NK14torch_tensorrt8DataTypeneE8DataType", "torch_tensorrt::DataType::operator!=::other"], [0, 3, 1, "_CPPv4NK14torch_tensorrt8DataTypeneEN8DataType5ValueE", "torch_tensorrt::DataType::operator!=::other"], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypeeqE8DataType", "torch_tensorrt::DataType::operator=="], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypeeqEN8DataType5ValueE", "torch_tensorrt::DataType::operator=="], [0, 3, 1, "_CPPv4NK14torch_tensorrt8DataTypeeqE8DataType", "torch_tensorrt::DataType::operator==::other"], [0, 3, 1, "_CPPv4NK14torch_tensorrt8DataTypeeqEN8DataType5ValueE", "torch_tensorrt::DataType::operator==::other"], [46, 1, 1, "_CPPv4N14torch_tensorrt6DeviceE", "torch_tensorrt::Device"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device6DeviceEv", "torch_tensorrt::Device::Device"], [1, 1, 1, "_CPPv4N14torch_tensorrt6Device10DeviceTypeE", "torch_tensorrt::Device::DeviceType"], [46, 1, 1, "_CPPv4N14torch_tensorrt6Device10DeviceTypeE", "torch_tensorrt::Device::DeviceType"], [1, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeE5Value", "torch_tensorrt::Device::DeviceType::DeviceType"], [1, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEN3c1010DeviceTypeE", "torch_tensorrt::Device::DeviceType::DeviceType"], [1, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEv", "torch_tensorrt::Device::DeviceType::DeviceType"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeE5Value", "torch_tensorrt::Device::DeviceType::DeviceType"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEN3c1010DeviceTypeE", "torch_tensorrt::Device::DeviceType::DeviceType"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEv", "torch_tensorrt::Device::DeviceType::DeviceType"], [1, 3, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeE5Value", "torch_tensorrt::Device::DeviceType::DeviceType::t"], [1, 3, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEN3c1010DeviceTypeE", "torch_tensorrt::Device::DeviceType::DeviceType::t"], [46, 3, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeE5Value", "torch_tensorrt::Device::DeviceType::DeviceType::t"], [46, 3, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEN3c1010DeviceTypeE", "torch_tensorrt::Device::DeviceType::DeviceType::t"], [1, 4, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5ValueE", "torch_tensorrt::Device::DeviceType::Value"], [46, 4, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5ValueE", "torch_tensorrt::Device::DeviceType::Value"], [1, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kDLAE", "torch_tensorrt::Device::DeviceType::Value::kDLA"], [46, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kDLAE", "torch_tensorrt::Device::DeviceType::Value::kDLA"], [1, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kGPUE", "torch_tensorrt::Device::DeviceType::Value::kGPU"], [46, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kGPUE", "torch_tensorrt::Device::DeviceType::Value::kGPU"], [1, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kDLAE", "torch_tensorrt::Device::DeviceType::kDLA"], [1, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kGPUE", "torch_tensorrt::Device::DeviceType::kGPU"], [1, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypecv5ValueEv", "torch_tensorrt::Device::DeviceType::operator Value"], [46, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypecv5ValueEv", "torch_tensorrt::Device::DeviceType::operator Value"], [1, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceTypecvbEv", "torch_tensorrt::Device::DeviceType::operator bool"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceTypecvbEv", "torch_tensorrt::Device::DeviceType::operator bool"], [1, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeneE10DeviceType", "torch_tensorrt::Device::DeviceType::operator!="], [46, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeneE10DeviceType", "torch_tensorrt::Device::DeviceType::operator!="], [1, 3, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeneE10DeviceType", "torch_tensorrt::Device::DeviceType::operator!=::other"], [46, 3, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeneE10DeviceType", "torch_tensorrt::Device::DeviceType::operator!=::other"], [1, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeeqE10DeviceType", "torch_tensorrt::Device::DeviceType::operator=="], [46, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeeqE10DeviceType", "torch_tensorrt::Device::DeviceType::operator=="], [1, 3, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeeqE10DeviceType", "torch_tensorrt::Device::DeviceType::operator==::other"], [46, 3, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeeqE10DeviceType", "torch_tensorrt::Device::DeviceType::operator==::other"], [46, 6, 1, "_CPPv4N14torch_tensorrt6Device18allow_gpu_fallbackE", "torch_tensorrt::Device::allow_gpu_fallback"], [46, 6, 1, "_CPPv4N14torch_tensorrt6Device11device_typeE", "torch_tensorrt::Device::device_type"], [46, 6, 1, "_CPPv4N14torch_tensorrt6Device8dla_coreE", "torch_tensorrt::Device::dla_core"], [46, 6, 1, "_CPPv4N14torch_tensorrt6Device6gpu_idE", "torch_tensorrt::Device::gpu_id"], [17, 4, 1, "_CPPv4N14torch_tensorrt16EngineCapabilityE", "torch_tensorrt::EngineCapability"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability15kDLA_STANDALONEE", "torch_tensorrt::EngineCapability::kDLA_STANDALONE"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability7kSAFETYE", "torch_tensorrt::EngineCapability::kSAFETY"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability9kSTANDARDE", "torch_tensorrt::EngineCapability::kSTANDARD"], [47, 1, 1, "_CPPv4N14torch_tensorrt11GraphInputsE", "torch_tensorrt::GraphInputs"], [47, 6, 1, "_CPPv4N14torch_tensorrt11GraphInputs15input_signatureE", "torch_tensorrt::GraphInputs::input_signature"], [47, 6, 1, "_CPPv4N14torch_tensorrt11GraphInputs6inputsE", "torch_tensorrt::GraphInputs::inputs"], [48, 1, 1, "_CPPv4N14torch_tensorrt5InputE", "torch_tensorrt::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN2at6TensorE", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEv", "torch_tensorrt::Input::Input"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN2at6TensorE", "torch_tensorrt::Input::Input::tensor"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input5dtypeE", "torch_tensorrt::Input::dtype"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input6formatE", "torch_tensorrt::Input::format"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input9max_shapeE", "torch_tensorrt::Input::max_shape"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input9min_shapeE", "torch_tensorrt::Input::min_shape"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input9opt_shapeE", "torch_tensorrt::Input::opt_shape"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input5shapeE", "torch_tensorrt::Input::shape"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input13tensor_domainE", "torch_tensorrt::Input::tensor_domain"], [2, 1, 1, "_CPPv4N14torch_tensorrt12TensorFormatE", "torch_tensorrt::TensorFormat"], [2, 2, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatE5Value", "torch_tensorrt::TensorFormat::TensorFormat"], [2, 2, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatEN2at12MemoryFormatE", "torch_tensorrt::TensorFormat::TensorFormat"], [2, 2, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatEv", "torch_tensorrt::TensorFormat::TensorFormat"], [2, 3, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatE5Value", "torch_tensorrt::TensorFormat::TensorFormat::t"], [2, 3, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatEN2at12MemoryFormatE", "torch_tensorrt::TensorFormat::TensorFormat::t"], [2, 4, 1, "_CPPv4N14torch_tensorrt12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::Value"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value13kChannelsLastE", "torch_tensorrt::TensorFormat::Value::kChannelsLast"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value11kContiguousE", "torch_tensorrt::TensorFormat::Value::kContiguous"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value8kUnknownE", "torch_tensorrt::TensorFormat::Value::kUnknown"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value13kChannelsLastE", "torch_tensorrt::TensorFormat::kChannelsLast"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value11kContiguousE", "torch_tensorrt::TensorFormat::kContiguous"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value8kUnknownE", "torch_tensorrt::TensorFormat::kUnknown"], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormatcv5ValueEv", "torch_tensorrt::TensorFormat::operator Value"], [2, 2, 1, "_CPPv4N14torch_tensorrt12TensorFormatcvbEv", "torch_tensorrt::TensorFormat::operator bool"], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormatneE12TensorFormat", "torch_tensorrt::TensorFormat::operator!="], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormatneEN12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::operator!="], [2, 3, 1, "_CPPv4NK14torch_tensorrt12TensorFormatneE12TensorFormat", "torch_tensorrt::TensorFormat::operator!=::other"], [2, 3, 1, "_CPPv4NK14torch_tensorrt12TensorFormatneEN12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::operator!=::other"], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormateqE12TensorFormat", "torch_tensorrt::TensorFormat::operator=="], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormateqEN12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::operator=="], [2, 3, 1, "_CPPv4NK14torch_tensorrt12TensorFormateqE12TensorFormat", "torch_tensorrt::TensorFormat::operator==::other"], [2, 3, 1, "_CPPv4NK14torch_tensorrt12TensorFormateqEN12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::operator==::other"], [37, 2, 1, "_CPPv4N14torch_tensorrt15dump_build_infoEv", "torch_tensorrt::dump_build_info"], [35, 2, 1, "_CPPv4N14torch_tensorrt14get_build_infoEv", "torch_tensorrt::get_build_info"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability15kDLA_STANDALONEE", "torch_tensorrt::kDLA_STANDALONE"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability7kSAFETYE", "torch_tensorrt::kSAFETY"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability9kSTANDARDE", "torch_tensorrt::kSTANDARD"], [16, 4, 1, "_CPPv4N14torch_tensorrt7logging5LevelE", "torch_tensorrt::logging::Level"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kDEBUGE", "torch_tensorrt::logging::Level::kDEBUG"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kERRORE", "torch_tensorrt::logging::Level::kERROR"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kGRAPHE", "torch_tensorrt::logging::Level::kGRAPH"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level5kINFOE", "torch_tensorrt::logging::Level::kINFO"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level15kINTERNAL_ERRORE", "torch_tensorrt::logging::Level::kINTERNAL_ERROR"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level8kWARNINGE", "torch_tensorrt::logging::Level::kWARNING"], [24, 2, 1, "_CPPv4N14torch_tensorrt7logging24get_is_colored_output_onEv", "torch_tensorrt::logging::get_is_colored_output_on"], [22, 2, 1, "_CPPv4N14torch_tensorrt7logging18get_logging_prefixEv", "torch_tensorrt::logging::get_logging_prefix"], [23, 2, 1, "_CPPv4N14torch_tensorrt7logging24get_reportable_log_levelEv", "torch_tensorrt::logging::get_reportable_log_level"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kDEBUGE", "torch_tensorrt::logging::kDEBUG"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kERRORE", "torch_tensorrt::logging::kERROR"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kGRAPHE", "torch_tensorrt::logging::kGRAPH"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level5kINFOE", "torch_tensorrt::logging::kINFO"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level15kINTERNAL_ERRORE", "torch_tensorrt::logging::kINTERNAL_ERROR"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level8kWARNINGE", "torch_tensorrt::logging::kWARNING"], [26, 2, 1, "_CPPv4N14torch_tensorrt7logging3logE5LevelNSt6stringE", "torch_tensorrt::logging::log"], [26, 3, 1, "_CPPv4N14torch_tensorrt7logging3logE5LevelNSt6stringE", "torch_tensorrt::logging::log::lvl"], [26, 3, 1, "_CPPv4N14torch_tensorrt7logging3logE5LevelNSt6stringE", "torch_tensorrt::logging::log::msg"], [27, 2, 1, "_CPPv4N14torch_tensorrt7logging24set_is_colored_output_onEb", "torch_tensorrt::logging::set_is_colored_output_on"], [27, 3, 1, "_CPPv4N14torch_tensorrt7logging24set_is_colored_output_onEb", "torch_tensorrt::logging::set_is_colored_output_on::colored_output_on"], [28, 2, 1, "_CPPv4N14torch_tensorrt7logging18set_logging_prefixENSt6stringE", "torch_tensorrt::logging::set_logging_prefix"], [28, 3, 1, "_CPPv4N14torch_tensorrt7logging18set_logging_prefixENSt6stringE", "torch_tensorrt::logging::set_logging_prefix::prefix"], [25, 2, 1, "_CPPv4N14torch_tensorrt7logging24set_reportable_log_levelE5Level", "torch_tensorrt::logging::set_reportable_log_level"], [25, 3, 1, "_CPPv4N14torch_tensorrt7logging24set_reportable_log_levelE5Level", "torch_tensorrt::logging::set_reportable_log_level::lvl"], [3, 1, 1, "_CPPv4I0EN14torch_tensorrt3ptq19Int8CacheCalibratorE", "torch_tensorrt::ptq::Int8CacheCalibrator"], [3, 7, 1, "_CPPv4I0EN14torch_tensorrt3ptq19Int8CacheCalibratorE", "torch_tensorrt::ptq::Int8CacheCalibrator::Algorithm"], [3, 2, 1, "_CPPv4N14torch_tensorrt3ptq19Int8CacheCalibrator19Int8CacheCalibratorERKNSt6stringE", "torch_tensorrt::ptq::Int8CacheCalibrator::Int8CacheCalibrator"], [3, 3, 1, "_CPPv4N14torch_tensorrt3ptq19Int8CacheCalibrator19Int8CacheCalibratorERKNSt6stringE", "torch_tensorrt::ptq::Int8CacheCalibrator::Int8CacheCalibrator::cache_file_path"], [3, 2, 1, "_CPPv4N14torch_tensorrt3ptq19Int8CacheCalibratorcvPN8nvinfer115IInt8CalibratorEEv", "torch_tensorrt::ptq::Int8CacheCalibrator::operator nvinfer1::IInt8Calibrator*"], [4, 1, 1, "_CPPv4I00EN14torch_tensorrt3ptq14Int8CalibratorE", "torch_tensorrt::ptq::Int8Calibrator"], [4, 7, 1, "_CPPv4I00EN14torch_tensorrt3ptq14Int8CalibratorE", "torch_tensorrt::ptq::Int8Calibrator::Algorithm"], [4, 7, 1, "_CPPv4I00EN14torch_tensorrt3ptq14Int8CalibratorE", "torch_tensorrt::ptq::Int8Calibrator::DataLoaderUniquePtr"], [4, 2, 1, "_CPPv4N14torch_tensorrt3ptq14Int8Calibrator14Int8CalibratorE19DataLoaderUniquePtrRKNSt6stringEb", "torch_tensorrt::ptq::Int8Calibrator::Int8Calibrator"], [4, 3, 1, "_CPPv4N14torch_tensorrt3ptq14Int8Calibrator14Int8CalibratorE19DataLoaderUniquePtrRKNSt6stringEb", "torch_tensorrt::ptq::Int8Calibrator::Int8Calibrator::cache_file_path"], [4, 3, 1, "_CPPv4N14torch_tensorrt3ptq14Int8Calibrator14Int8CalibratorE19DataLoaderUniquePtrRKNSt6stringEb", "torch_tensorrt::ptq::Int8Calibrator::Int8Calibrator::dataloader"], [4, 3, 1, "_CPPv4N14torch_tensorrt3ptq14Int8Calibrator14Int8CalibratorE19DataLoaderUniquePtrRKNSt6stringEb", "torch_tensorrt::ptq::Int8Calibrator::Int8Calibrator::use_cache"], [4, 2, 1, "_CPPv4N14torch_tensorrt3ptq14Int8CalibratorcvPN8nvinfer115IInt8CalibratorEEv", "torch_tensorrt::ptq::Int8Calibrator::operator nvinfer1::IInt8Calibrator*"], [29, 2, 1, "_CPPv4I0EN14torch_tensorrt3ptq26make_int8_cache_calibratorE19Int8CacheCalibratorI9AlgorithmERKNSt6stringE", "torch_tensorrt::ptq::make_int8_cache_calibrator"], [29, 7, 1, "_CPPv4I0EN14torch_tensorrt3ptq26make_int8_cache_calibratorE19Int8CacheCalibratorI9AlgorithmERKNSt6stringE", "torch_tensorrt::ptq::make_int8_cache_calibrator::Algorithm"], [29, 3, 1, "_CPPv4I0EN14torch_tensorrt3ptq26make_int8_cache_calibratorE19Int8CacheCalibratorI9AlgorithmERKNSt6stringE", "torch_tensorrt::ptq::make_int8_cache_calibrator::cache_file_path"], [30, 2, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator"], [30, 7, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::Algorithm"], [30, 7, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::DataLoader"], [30, 3, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::cache_file_path"], [30, 3, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::dataloader"], [30, 3, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::use_cache"], [36, 2, 1, "_CPPv4N14torch_tensorrt10set_deviceEKi", "torch_tensorrt::set_device"], [36, 3, 1, "_CPPv4N14torch_tensorrt10set_deviceEKi", "torch_tensorrt::set_device::gpu_id"], [49, 1, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpecE", "torch_tensorrt::torchscript::CompileSpec"], [49, 2, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecEN5torch3jit6IValueE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec"], [49, 2, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorI5InputEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec"], [49, 2, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorIN3c108ArrayRefI7int64_tEEEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec"], [49, 2, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorINSt6vectorI7int64_tEEEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec"], [49, 3, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorIN3c108ArrayRefI7int64_tEEEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec::fixed_sizes"], [49, 3, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorINSt6vectorI7int64_tEEEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec::fixed_sizes"], [49, 3, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecEN5torch3jit6IValueE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec::input_signature"], [49, 3, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorI5InputEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec::inputs"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec19allow_shape_tensorsE", "torch_tensorrt::torchscript::CompileSpec::allow_shape_tensors"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec10capabilityE", "torch_tensorrt::torchscript::CompileSpec::capability"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec5debugE", "torch_tensorrt::torchscript::CompileSpec::debug"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec6deviceE", "torch_tensorrt::torchscript::CompileSpec::device"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec12disable_tf32E", "torch_tensorrt::torchscript::CompileSpec::disable_tf32"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec20dla_global_dram_sizeE", "torch_tensorrt::torchscript::CompileSpec::dla_global_dram_size"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec19dla_local_dram_sizeE", "torch_tensorrt::torchscript::CompileSpec::dla_local_dram_size"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec13dla_sram_sizeE", "torch_tensorrt::torchscript::CompileSpec::dla_sram_size"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec18enabled_precisionsE", "torch_tensorrt::torchscript::CompileSpec::enabled_precisions"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec12graph_inputsE", "torch_tensorrt::torchscript::CompileSpec::graph_inputs"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec14min_block_sizeE", "torch_tensorrt::torchscript::CompileSpec::min_block_size"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec20num_avg_timing_itersE", "torch_tensorrt::torchscript::CompileSpec::num_avg_timing_iters"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec14ptq_calibratorE", "torch_tensorrt::torchscript::CompileSpec::ptq_calibrator"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec5refitE", "torch_tensorrt::torchscript::CompileSpec::refit"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec24require_full_compilationE", "torch_tensorrt::torchscript::CompileSpec::require_full_compilation"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec14sparse_weightsE", "torch_tensorrt::torchscript::CompileSpec::sparse_weights"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec22torch_executed_modulesE", "torch_tensorrt::torchscript::CompileSpec::torch_executed_modules"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec18torch_executed_opsE", "torch_tensorrt::torchscript::CompileSpec::torch_executed_ops"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec24truncate_long_and_doubleE", "torch_tensorrt::torchscript::CompileSpec::truncate_long_and_double"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec14workspace_sizeE", "torch_tensorrt::torchscript::CompileSpec::workspace_size"], [31, 2, 1, "_CPPv4N14torch_tensorrt11torchscript29check_method_operator_supportERKN5torch3jit6ModuleENSt6stringE", "torch_tensorrt::torchscript::check_method_operator_support"], [31, 3, 1, "_CPPv4N14torch_tensorrt11torchscript29check_method_operator_supportERKN5torch3jit6ModuleENSt6stringE", "torch_tensorrt::torchscript::check_method_operator_support::method_name"], [31, 3, 1, "_CPPv4N14torch_tensorrt11torchscript29check_method_operator_supportERKN5torch3jit6ModuleENSt6stringE", "torch_tensorrt::torchscript::check_method_operator_support::module"], [32, 2, 1, "_CPPv4N14torch_tensorrt11torchscript7compileERKN5torch3jit6ModuleE11CompileSpec", "torch_tensorrt::torchscript::compile"], [32, 3, 1, "_CPPv4N14torch_tensorrt11torchscript7compileERKN5torch3jit6ModuleE11CompileSpec", "torch_tensorrt::torchscript::compile::info"], [32, 3, 1, "_CPPv4N14torch_tensorrt11torchscript7compileERKN5torch3jit6ModuleE11CompileSpec", "torch_tensorrt::torchscript::compile::module"], [34, 2, 1, "_CPPv4N14torch_tensorrt11torchscript28convert_method_to_trt_engineERKN5torch3jit6ModuleENSt6stringE11CompileSpec", "torch_tensorrt::torchscript::convert_method_to_trt_engine"], [34, 3, 1, "_CPPv4N14torch_tensorrt11torchscript28convert_method_to_trt_engineERKN5torch3jit6ModuleENSt6stringE11CompileSpec", "torch_tensorrt::torchscript::convert_method_to_trt_engine::info"], [34, 3, 1, "_CPPv4N14torch_tensorrt11torchscript28convert_method_to_trt_engineERKN5torch3jit6ModuleENSt6stringE11CompileSpec", "torch_tensorrt::torchscript::convert_method_to_trt_engine::method_name"], [34, 3, 1, "_CPPv4N14torch_tensorrt11torchscript28convert_method_to_trt_engineERKN5torch3jit6ModuleENSt6stringE11CompileSpec", "torch_tensorrt::torchscript::convert_method_to_trt_engine::module"], [33, 2, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module"], [33, 3, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module::device"], [33, 3, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module::engine"], [33, 3, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module::input_binding_names"], [33, 3, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module::output_binding_names"], [75, 8, 0, "-", "torch_tensorrt"]], "torch_tensorrt": [[75, 9, 1, "", "Device"], [75, 9, 1, "", "DeviceType"], [75, 9, 1, "", "EngineCapability"], [75, 9, 1, "", "Input"], [75, 9, 1, "", "MutableTorchTensorRTModule"], [75, 12, 1, "", "compile"], [75, 12, 1, "", "convert_method_to_trt_engine"], [75, 9, 1, "", "dtype"], [111, 8, 0, "-", "dynamo"], [71, 8, 0, "-", "fx"], [75, 12, 1, "", "load"], [72, 8, 0, "-", "logging"], [75, 9, 1, "", "memory_format"], [74, 8, 0, "-", "runtime"], [75, 12, 1, "", "save"], [76, 8, 0, "-", "ts"]], "torch_tensorrt.Device": [[75, 10, 1, "", "__init__"], [75, 11, 1, "", "device_type"], [75, 11, 1, "", "dla_core"], [75, 11, 1, "", "gpu_id"]], "torch_tensorrt.DeviceType": [[75, 11, 1, "", "DLA"], [75, 11, 1, "", "GPU"], [75, 11, 1, "", "UNKNOWN"], [75, 10, 1, "", "to"], [75, 10, 1, "", "try_from"], [75, 10, 1, "", "try_to"]], "torch_tensorrt.EngineCapability": [[75, 11, 1, "", "DLA_STANDALONE"], [75, 11, 1, "", "SAFETY"], [75, 11, 1, "", "STANDARD"], [75, 10, 1, "", "to"], [75, 10, 1, "", "try_from"], [75, 10, 1, "", "try_to"]], "torch_tensorrt.Input": [[75, 10, 1, "", "__init__"], [75, 11, 1, "", "dtype"], [75, 10, 1, "", "example_tensor"], [75, 11, 1, "", "format"], [75, 10, 1, "", "from_tensor"], [75, 10, 1, "", "from_tensors"]], "torch_tensorrt.MutableTorchTensorRTModule": [[75, 10, 1, "", "__init__"], [75, 10, 1, "", "compile"], [75, 10, 1, "", "refit_gm"]], "torch_tensorrt.dtype": [[75, 11, 1, "", "b"], [75, 11, 1, "", "bf16"], [75, 11, 1, "", "f16"], [75, 11, 1, "", "f32"], [75, 11, 1, "", "f64"], [75, 11, 1, "", "f8"], [75, 11, 1, "", "i32"], [75, 11, 1, "", "i64"], [75, 11, 1, "", "i8"], [75, 10, 1, "", "to"], [75, 10, 1, "", "try_from"], [75, 10, 1, "", "try_to"], [75, 11, 1, "", "u8"], [75, 11, 1, "", "unknown"]], "torch_tensorrt.dynamo": [[70, 9, 1, "", "CompilationSettings"], [70, 12, 1, "", "compile"], [70, 12, 1, "", "export"], [70, 12, 1, "", "refit_module_weights"], [70, 12, 1, "", "trace"]], "torch_tensorrt.fx": [[71, 9, 1, "", "InputTensorSpec"], [71, 9, 1, "", "TRTInterpreter"], [71, 9, 1, "", "TRTInterpreterResult"], [71, 9, 1, "", "TRTModule"], [71, 12, 1, "", "compile"]], "torch_tensorrt.logging": [[72, 9, 1, "", "debug"], [72, 9, 1, "", "errors"], [72, 9, 1, "", "graphs"], [72, 9, 1, "", "info"], [72, 9, 1, "", "internal_errors"], [72, 9, 1, "", "warnings"]], "torch_tensorrt.memory_format": [[75, 11, 1, "", "cdhw32"], [75, 11, 1, "", "chw16"], [75, 11, 1, "", "chw2"], [75, 11, 1, "", "chw32"], [75, 11, 1, "", "chw4"], [75, 11, 1, "", "dhwc"], [75, 11, 1, "", "dhwc8"], [75, 11, 1, "", "dla_hwc4"], [75, 11, 1, "", "dla_linear"], [75, 11, 1, "", "hwc"], [75, 11, 1, "", "hwc16"], [75, 11, 1, "", "hwc8"], [75, 11, 1, "", "linear"], [75, 10, 1, "", "to"], [75, 10, 1, "", "try_from"], [75, 10, 1, "", "try_to"]], "torch_tensorrt.runtime": [[74, 9, 1, "", "PythonTorchTensorRTModule"], [74, 9, 1, "", "TorchTensorRTModule"], [74, 12, 1, "", "set_multi_device_safe_mode"]], "torch_tensorrt.runtime.PythonTorchTensorRTModule": [[74, 10, 1, "", "__init__"], [74, 10, 1, "", "cudagraphs_validate_shapes"], [74, 10, 1, "", "disable_profiling"], [74, 10, 1, "", "enable_profiling"], [74, 10, 1, "", "forward"], [74, 10, 1, "", "get_layer_info"]], "torch_tensorrt.runtime.TorchTensorRTModule": [[74, 10, 1, "", "__init__"], [74, 10, 1, "", "forward"], [74, 10, 1, "", "get_extra_state"], [74, 10, 1, "", "set_extra_state"]], "torch_tensorrt.ts": [[76, 12, 1, "", "TensorRTCompileSpec"], [76, 12, 1, "", "check_method_op_support"], [76, 12, 1, "", "compile"], [76, 12, 1, "", "convert_method_to_trt_engine"], [76, 12, 1, "", "embed_engine_in_new_module"], [73, 8, 0, "-", "ptq"]], "torch_tensorrt.ts.ptq": [[73, 9, 1, "", "CacheCalibrator"], [73, 9, 1, "", "CalibrationAlgo"], [73, 9, 1, "", "DataLoaderCalibrator"]], "torch_tensorrt.ts.ptq.CalibrationAlgo": [[73, 11, 1, "", "ENTROPY_CALIBRATION"], [73, 11, 1, "", "ENTROPY_CALIBRATION_2"], [73, 11, 1, "", "LEGACY_CALIBRATION"], [73, 11, 1, "", "MINMAX_CALIBRATION"]]}, "objtypes": {"0": "c:macro", "1": "cpp:class", "2": "cpp:function", "3": "cpp:functionParam", "4": "cpp:enum", "5": "cpp:enumerator", "6": "cpp:member", "7": "cpp:templateParam", "8": "py:module", "9": "py:class", "10": "py:method", "11": "py:attribute", "12": "py:function"}, "objnames": {"0": ["c", "macro", "C macro"], "1": ["cpp", "class", "C++ class"], "2": ["cpp", "function", "C++ function"], "3": ["cpp", "functionParam", "C++ function parameter"], "4": ["cpp", "enum", "C++ enum"], "5": ["cpp", "enumerator", "C++ enumerator"], "6": ["cpp", "member", "C++ member"], "7": ["cpp", "templateParam", "C++ template parameter"], "8": ["py", "module", "Python module"], "9": ["py", "class", "Python class"], "10": ["py", "method", "Python method"], "11": ["py", "attribute", "Python attribute"], "12": ["py", "function", "Python function"]}, "titleterms": {"class": [0, 1, 2, 3, 4, 20, 21, 38, 40, 41, 50, 70, 71, 73, 74, 75], "datatyp": 0, "document": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 16, 17, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 46, 47, 48, 49, 61, 68, 84, 85], "devic": [1, 46, 110], "devicetyp": 1, "nest": [1, 46], "relationship": [1, 3, 4, 46, 48], "tensorformat": 2, "templat": [3, 4, 29, 30], "int8cachecalibr": 3, "inherit": [3, 4, 48], "base": [3, 4, 48, 79], "type": [3, 4, 46, 48, 54], "int8calibr": 4, "defin": [5, 6, 7, 8, 9, 10, 11, 12, 19, 50, 105], "str": 5, "torch_tensorrt_patch_vers": 6, "torch_tensorrt_major_vers": 7, "torch_tensorrt_minor_vers": 8, "torchtrt_api": 9, "xstr": 10, "torchtrt_hidden": 11, "torch_tensorrt_vers": 12, "directori": [13, 14, 15, 51], "cpp": [13, 18, 19, 20, 21, 56], "subdirectori": [13, 14], "includ": [14, 18, 19, 20, 21], "torch_tensorrt": [15, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 45, 70, 71, 72, 73, 74, 75, 76, 99, 101, 102, 112], "file": [15, 18, 19, 20, 21, 42, 43, 44, 45, 50, 51], "enum": [16, 17, 38, 39, 50, 73, 75], "level": [16, 79, 81, 82], "enginecap": 17, "log": [18, 22, 23, 24, 25, 26, 27, 28, 39, 42, 72], "h": [18, 19, 20, 21, 42, 43, 44, 45, 56], "content": [18, 19, 20, 21, 38, 39, 40, 41, 79, 80, 81, 82, 83, 84], "definit": [18, 19, 20, 21, 82, 97, 98, 99, 100, 101, 102, 103, 104, 105], "By": [18, 19], "namespac": [18, 19, 20, 21, 38, 39, 40, 41, 50], "macro": [19, 43], "ptq": [20, 29, 30, 40, 44, 73, 90, 105], "function": [22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 50, 61, 70, 71, 74, 75, 76, 105], "get_logging_prefix": 22, "get_reportable_log_level": 23, "get_is_colored_output_on": 24, "set_reportable_log_level": 25, "set_is_colored_output_on": 27, "set_logging_prefix": 28, "make_int8_cache_calibr": 29, "make_int8_calibr": 30, "torchscript": [31, 32, 33, 34, 41, 60, 66, 68, 87, 88, 91, 111, 112], "check_method_operator_support": 31, "compil": [32, 57, 59, 63, 64, 66, 67, 68, 88, 94, 95, 97, 98, 99, 100, 101, 102, 103, 104, 106, 107, 109, 111, 112], "embed_engine_in_new_modul": 33, "convert_method_to_trt_engin": 34, "get_build_info": 35, "set_devic": 36, "dump_build_info": 37, "program": [42, 43, 44, 45, 63, 97, 110], "list": [42, 43, 44, 45, 82], "struct": [46, 47, 48, 49, 50], "graphinput": 47, "input": [48, 99, 101], "compilespec": 49, "torch": [50, 61, 63, 64, 65, 66, 67, 68, 88, 89, 91, 92, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112], "tensorrt": [50, 58, 61, 63, 64, 65, 66, 68, 88, 89, 91, 92, 96, 97, 99, 101, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112], "c": [50, 61, 66, 67, 68, 88, 90, 107], "api": [50, 51, 61, 66, 68], "hierarchi": 50, "full": [50, 51], "torchtrtc": [52, 88], "convers": [53, 57, 59, 60], "phase": [53, 55, 56, 57, 58, 59], "node": 53, "evalu": [53, 54, 69], "convert": [53, 54, 60, 65, 69, 88], "write": [54, 60, 62, 92], "dynamo": [54, 62, 68, 70, 95, 103, 104, 106, 111, 112], "implement": 54, "registr": 54, "capabl": 54, "valid": 54, "contract": [54, 60], "exampl": [54, 62, 81, 83, 107], "convolut": 54, "oper": [54, 64, 69, 88, 92], "decomposit": 54, "addmm": [54, 55], "lower": [55, 57, 59, 62], "pass": [55, 62], "us": [55, 61, 88, 89, 91, 92, 98, 99, 101, 102, 103, 104, 105, 107, 109], "eliminatecommonsubexpress": 55, "elimin": 55, "dead": 55, "code": [55, 68, 81], "except": 55, "Or": 55, "pattern": 55, "redund": 55, "guard": 55, "freez": 55, "modul": [55, 87, 88, 96, 112], "fuse": 55, "branch": 55, "linear": 55, "flatten": 55, "graph": [55, 58, 112], "tupl": 55, "fallback": [55, 56], "peephol": 55, "optim": [55, 67, 108], "remov": 55, "contigu": 55, "dropout": 55, "To": 55, "unpack": 55, "logsoftmax": 55, "unrol": 55, "loop": [55, 105], "replac": [55, 81], "tile": 55, "repeat": 55, "partit": [56, 57, 59], "partitoninfo": 56, "segmentedblock": 56, "shape_analysi": 56, "automat": 56, "depend": [56, 66], "awar": [56, 107], "runtim": [57, 58, 59, 74, 110], "background": [58, 60], "engin": [58, 65, 92, 93, 94], "executor": 58, "op": [58, 65, 92], "construct": 58, "result": 58, "serial": [58, 64, 67], "deseri": 58, "abi": [58, 66], "version": [58, 66], "format": [58, 112], "system": [59, 66], "overview": 59, "what": 60, "guarante": 60, "respons": 60, "context": [60, 79], "arg": [60, 80], "weight": [60, 97, 105], "other": 60, "advic": 60, "link": [61, 81], "develop": 61, "avail": 61, "layer": 61, "expect": 61, "dimens": 61, "python": [61, 66, 67, 68, 87, 89, 90], "sometim": 61, "easier": 61, "read": 61, "pytorch": [61, 65, 68, 91, 92, 103, 104, 107], "native_op": 61, "ir": [61, 111, 112], "aten": 62, "basic": 62, "requir": 62, "regist": [62, 88], "export": [63, 67, 102, 109], "customiz": [63, 64], "set": [63, 64, 96, 98, 102, 108], "under": [63, 88, 109], "hood": [63, 88, 109], "trace": 63, "backend": [64, 99, 103, 104], "kei": 64, "featur": 64, "custom": [64, 88, 92, 94, 98, 109], "usag": [64, 97, 98], "after": 64, "model": [64, 65, 92, 97, 98, 99, 100, 101, 102, 103, 104, 105, 107, 108, 111], "perform": 64, "coverag": 64, "feasibl": 64, "dynam": [64, 99, 107, 109], "shape": [64, 99, 107, 109], "support": [64, 69], "recompil": [64, 99], "condit": 64, "fx": [65, 68, 71, 107, 112], "frontend": [65, 66, 68, 91, 107, 112], "user": [65, 68], "guid": [65, 68], "acc": 65, "tracer": 65, "fx2trt": 65, "how": [65, 79, 90], "add": 65, "miss": 65, "instal": [66, 86], "precompil": 66, "binari": 66, "specif": 66, "cuda": [66, 98, 101], "nightli": 66, "build": [66, 79, 108], "onli": 66, "from": [66, 91], "sourc": 66, "linux": 66, "packag": [66, 110], "addit": 66, "option": [66, 67, 79, 80, 82, 99, 101, 112], "distribut": 66, "No": 66, "librari": [66, 110], "standalon": 66, "releas": 66, "debug": 66, "pre": [66, 105], "cxx11": 66, "choos": 66, "right": 66, "window": 66, "step": [66, 67, 108], "advanc": [66, 97, 98], "setup": 66, "troubleshoot": 66, "altern": 66, "cmake": 66, "nativ": 66, "aarch64": 66, "jetson": 66, "prerequisit": 66, "environ": 66, "cli": [66, 68], "quick": 67, "start": [67, 68], "1": [67, 83, 108], "2": [67, 83, 84, 108], "deploi": [67, 105, 107, 110], "deploy": 67, "In": [68, 97], "framework": 68, "infer": [68, 99, 100, 101, 102, 105, 108], "nvidia": 68, "gpu": 68, "get": 68, "tutori": [68, 106], "contributor": 68, "indic": 68, "legaci": [68, 112], "further": 68, "inform": 68, "current": 69, "through": 69, "ts": [73, 76, 112], "submodul": 75, "comput": 77, "time": [77, 112], "changelog": 78, "configur": 79, "project": 79, "wide": 79, "html": 79, "theme": [79, 85], "toc": 79, "page": 79, "tabl": [79, 80, 81, 82, 83, 84], "mod": 80, "test_py_modul": 80, "gener": [80, 103, 104], "index": 80, "paramet": 80, "data": 80, "paragraph": [81, 84], "markup": 81, "inlin": 81, "math": 81, "meta": 81, "block": 81, "liter": 81, "line": 81, "quot": 81, "doctest": 81, "emphas": 81, "number": [81, 82], "sidebar": 81, "ch": 81, "ien": 81, "The": [81, 88, 103, 104], "creativ": 81, "A": 81, "refer": 81, "footnot": 81, "citat": [81, 90], "glossari": 81, "target": 81, "direct": 81, "center": 81, "text": [81, 103, 104], "imag": [81, 82], "figur": 81, "admonit": 81, "And": 81, "wai": 81, "topic": 81, "rubric": 81, "titl": 81, "compound": 81, "download": [81, 86], "enumer": 82, "field": 82, "bullet": 82, "second": 82, "But": 82, "deeper": 82, "down": 82, "rabbit": 82, "hole": 82, "hlist": 82, "grid": 82, "giant": 82, "can": 82, "have": 82, "caption": [82, 85], "like": [82, 103, 104], "thi": [82, 85], "one": 82, "long": [83, 85], "sticki": 83, "nav": 83, "menu": [83, 85], "3": [83, 108], "4": 83, "5": 83, "6": 83, "7": 83, "8": 83, "9": 83, "10": 83, "11": 83, "12": 83, "13": 83, "14": 83, "15": 83, "16": 83, "17": 83, "18": 83, "19": 83, "20": 83, "submenu": 83, "subsubmenu": 83, "structur": 84, "element": 84, "section": 84, "subsect": 84, "subsubsect": 84, "demo": 85, "an": 85, "incred": 85, "via": 86, "git": 86, "creat": [87, 90], "work": [87, 88], "save": [87, 96, 111], "disk": 87, "quickstart": 88, "unsupport": 88, "post": 90, "train": [90, 105, 107], "quantiz": [90, 105, 107], "your": [90, 108], "own": 90, "applic": 90, "directli": 91, "kernel": 92, "within": 92, "test": 92, "our": 92, "wrap": 92, "insert": 92, "cach": [93, 94, 97], "bert": [93, 107], "jit": [94, 109], "aot": [94, 109], "mutabl": 96, "initi": 96, "make": [96, 97], "modif": 96, "stabl": [96, 100], "diffus": [96, 100], "huggingfac": 96, "refit": 97, "new": 97, "standard": 97, "workflow": 97, "import": [97, 98, 99, 100, 101, 102, 103, 104, 105], "pretrain": 97, "map": 97, "place": 97, "default": [98, 102], "cleanup": [98, 101], "driver": [98, 101], "error": [98, 101], "note": [98, 101], "resnet": 99, "argument": [99, 101], "avoid": 99, "specifi": 99, "befor": 99, "trt": 99, "transform": [101, 107], "cudagraph": [102, 110], "integr": 102, "gpt2": 103, "output": [103, 104], "decod": [103, 104], "sentenc": [103, 104], "should": [103, 104], "look": [103, 104], "i": [103, 104], "enjoi": [103, 104], "walk": [103, 104], "my": [103, 104], "cute": [103, 104], "dog": [103, 104], "m": [103, 104], "sure": [103, 104], "ll": [103, 104], "ever": [103, 104], "abl": [103, 104], "llama2": 104, "load": [105, 111], "dataset": 105, "loss": 105, "calibr": 105, "tune": 105, "fp8": 105, "notebook": 107, "citrinet": 107, "efficientnet": 107, "mask": 107, "languag": 107, "mlm": 107, "hug": 107, "face": 107, "acceler": 107, "serv": [107, 108], "resnet50": 107, "lenet": 107, "deep": 107, "learn": 107, "object": 107, "detect": 107, "ssd": 107, "int8": 107, "triton": 108, "up": 108, "server": 108, "client": 108, "queri": 108, "constraint": 109, "libtorchtrt": 110, "so": 110, "plugin": 110, "multi": 110, "safe": 110, "mode": 110, "exportedprogram": 111, "b": 111, "explain": 112, "just": 112, "accept": 112, "return": 112, "ahead": 112, "dla": 113}, "envversion": {"sphinx.domains.c": 2, "sphinx.domains.changeset": 1, "sphinx.domains.citation": 1, "sphinx.domains.cpp": 6, "sphinx.domains.index": 1, "sphinx.domains.javascript": 2, "sphinx.domains.math": 2, "sphinx.domains.python": 3, "sphinx.domains.rst": 2, "sphinx.domains.std": 2, "nbsphinx": 4, "sphinx.ext.intersphinx": 1, "sphinx.ext.todo": 2, "sphinx.ext.viewcode": 1, "sphinx": 56}}) \ No newline at end of file +Search.setIndex({"docnames": ["_cpp_api/classtorch__tensorrt_1_1DataType", "_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType", "_cpp_api/classtorch__tensorrt_1_1TensorFormat", "_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator", "_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator", "_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502", "_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268", "_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e", "_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827", "_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b", "_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da", "_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59", "_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883", "_cpp_api/dir_cpp", "_cpp_api/dir_cpp_include", "_cpp_api/dir_cpp_include_torch_tensorrt", "_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558", "_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb", "_cpp_api/file_cpp_include_torch_tensorrt_logging.h", "_cpp_api/file_cpp_include_torch_tensorrt_macros.h", "_cpp_api/file_cpp_include_torch_tensorrt_ptq.h", "_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc", "_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c", "_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2", "_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528", "_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384", "_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1", "_cpp_api/namespace_torch_tensorrt", "_cpp_api/namespace_torch_tensorrt__logging", "_cpp_api/namespace_torch_tensorrt__ptq", "_cpp_api/namespace_torch_tensorrt__torchscript", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h", "_cpp_api/structtorch__tensorrt_1_1Device", "_cpp_api/structtorch__tensorrt_1_1GraphInputs", "_cpp_api/structtorch__tensorrt_1_1Input", "_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec", "_cpp_api/torch_tensort_cpp", "_cpp_api/unabridged_orphan", "cli/torchtrtc", "contributors/conversion", "contributors/dynamo_converters", "contributors/lowering", "contributors/partitioning", "contributors/phases", "contributors/runtime", "contributors/system_overview", "contributors/ts_converters", "contributors/useful_links", "contributors/writing_dynamo_aten_lowering_passes", "dynamo/dynamo_export", "dynamo/torch_compile", "fx/getting_started_with_fx_path", "getting_started/installation", "getting_started/quick_start", "index", "indices/supported_ops", "py_api/dynamo", "py_api/fx", "py_api/logging", "py_api/ptq", "py_api/runtime", "py_api/torch_tensorrt", "py_api/ts", "sg_execution_times", "src/pytorch-sphinx-theme/docs/changelog", "src/pytorch-sphinx-theme/docs/configuring", "src/pytorch-sphinx-theme/docs/demo/api", "src/pytorch-sphinx-theme/docs/demo/demo", "src/pytorch-sphinx-theme/docs/demo/lists_tables", "src/pytorch-sphinx-theme/docs/demo/long", "src/pytorch-sphinx-theme/docs/demo/structure", "src/pytorch-sphinx-theme/docs/index", "src/pytorch-sphinx-theme/docs/installing", "ts/creating_torchscript_module_in_python", "ts/getting_started_with_cpp_api", "ts/getting_started_with_python_api", "ts/ptq", "ts/torchscript_frontend_from_pytorch", "tutorials/_rendered_examples/dynamo/custom_kernel_plugins", "tutorials/_rendered_examples/dynamo/engine_caching_bert_example", "tutorials/_rendered_examples/dynamo/engine_caching_example", "tutorials/_rendered_examples/dynamo/index", "tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example", "tutorials/_rendered_examples/dynamo/refit_engine_example", "tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage", "tutorials/_rendered_examples/dynamo/torch_compile_resnet_example", "tutorials/_rendered_examples/dynamo/torch_compile_stable_diffusion", "tutorials/_rendered_examples/dynamo/torch_compile_transformers_example", "tutorials/_rendered_examples/dynamo/torch_export_cudagraphs", "tutorials/_rendered_examples/dynamo/torch_export_gpt2", "tutorials/_rendered_examples/dynamo/torch_export_llama2", "tutorials/_rendered_examples/dynamo/vgg16_ptq", "tutorials/_rendered_examples/index", "tutorials/notebooks", "tutorials/serving_torch_tensorrt_with_triton", "user_guide/dynamic_shapes", "user_guide/runtime", "user_guide/saving_models", "user_guide/torch_tensorrt_explained", "user_guide/using_dla"], "filenames": ["_cpp_api/classtorch__tensorrt_1_1DataType.rst", "_cpp_api/classtorch__tensorrt_1_1Device_1_1DeviceType.rst", "_cpp_api/classtorch__tensorrt_1_1TensorFormat.rst", "_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8CacheCalibrator.rst", "_cpp_api/classtorch__tensorrt_1_1ptq_1_1Int8Calibrator.rst", "_cpp_api/define_macros_8h_1a18d295a837ac71add5578860b55e5502.rst", "_cpp_api/define_macros_8h_1a282fd3c0b1c3a215148ae372070e1268.rst", "_cpp_api/define_macros_8h_1a31398a6d4d27e28817afb0f0139e909e.rst", "_cpp_api/define_macros_8h_1a35703561b26b1a9d2738ad7d58b27827.rst", "_cpp_api/define_macros_8h_1abd1465eb38256d3f22cc1426b23d516b.rst", "_cpp_api/define_macros_8h_1abe87b341f562fd1cf40b7672e4d759da.rst", "_cpp_api/define_macros_8h_1ad19939408f7be171a74a89928b36eb59.rst", "_cpp_api/define_macros_8h_1adad592a7b1b7eed529cdf6acd584c883.rst", "_cpp_api/dir_cpp.rst", "_cpp_api/dir_cpp_include.rst", "_cpp_api/dir_cpp_include_torch_tensorrt.rst", "_cpp_api/enum_namespacetorch__tensorrt_1_1logging_1a130f65408ad8cbaee060f05e8db69558.rst", "_cpp_api/enum_namespacetorch__tensorrt_1a3fbe5d72e4fc624dbd038853079620eb.rst", "_cpp_api/file_cpp_include_torch_tensorrt_logging.h.rst", "_cpp_api/file_cpp_include_torch_tensorrt_macros.h.rst", "_cpp_api/file_cpp_include_torch_tensorrt_ptq.h.rst", "_cpp_api/file_cpp_include_torch_tensorrt_torch_tensorrt.h.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0593f776f469c20469e2f729fc7861a3.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a0c012cb374addd90eb1f42eaec570650.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a56e110feaaba2c3fd44bd201fd21a76a.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1a7cb50492421ea9de4e3db895819df6f2.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ac46ac0901cb97e3ae6e93b45f24e90b8.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1ad2efd47b6c3689e58ccc595680579ae5.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1logging_1af8f3443813315af7901903d25dd495cc.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a226e3c83379d1012cde8578c1c86b16c.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1ptq_1a6186e305f47c1d94b6130ef6c7f7e178.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a5b405fd3bf3c8fc2e2a54cbbab979797.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a6e19490a08fb1553c9dd347a5ae79db9.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1a81f9783517335dda877d8cfcf38987c9.rst", "_cpp_api/function_namespacetorch__tensorrt_1_1torchscript_1ae8d56472106eeef37fbe51ff7f40c9b2.rst", "_cpp_api/function_namespacetorch__tensorrt_1ac4ab8313ae72c2c899ea31548b528528.rst", "_cpp_api/function_namespacetorch__tensorrt_1ad1acd06eaeaffbbcf6e7ebf426891384.rst", "_cpp_api/function_namespacetorch__tensorrt_1ad6a4ee8ca6c8f6e5519eb1128ec7f4a1.rst", "_cpp_api/namespace_torch_tensorrt.rst", "_cpp_api/namespace_torch_tensorrt__logging.rst", "_cpp_api/namespace_torch_tensorrt__ptq.rst", "_cpp_api/namespace_torch_tensorrt__torchscript.rst", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_logging.h.rst", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_macros.h.rst", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_ptq.h.rst", "_cpp_api/program_listing_file_cpp_include_torch_tensorrt_torch_tensorrt.h.rst", "_cpp_api/structtorch__tensorrt_1_1Device.rst", "_cpp_api/structtorch__tensorrt_1_1GraphInputs.rst", "_cpp_api/structtorch__tensorrt_1_1Input.rst", "_cpp_api/structtorch__tensorrt_1_1torchscript_1_1CompileSpec.rst", "_cpp_api/torch_tensort_cpp.rst", "_cpp_api/unabridged_orphan.rst", "cli/torchtrtc.rst", "contributors/conversion.rst", "contributors/dynamo_converters.rst", "contributors/lowering.rst", "contributors/partitioning.rst", "contributors/phases.rst", "contributors/runtime.rst", "contributors/system_overview.rst", "contributors/ts_converters.rst", "contributors/useful_links.rst", "contributors/writing_dynamo_aten_lowering_passes.rst", "dynamo/dynamo_export.rst", "dynamo/torch_compile.rst", "fx/getting_started_with_fx_path.rst", "getting_started/installation.rst", "getting_started/quick_start.rst", "index.rst", "indices/supported_ops.rst", "py_api/dynamo.rst", "py_api/fx.rst", "py_api/logging.rst", "py_api/ptq.rst", "py_api/runtime.rst", "py_api/torch_tensorrt.rst", "py_api/ts.rst", "sg_execution_times.rst", "src/pytorch-sphinx-theme/docs/changelog.rst", "src/pytorch-sphinx-theme/docs/configuring.rst", "src/pytorch-sphinx-theme/docs/demo/api.rst", "src/pytorch-sphinx-theme/docs/demo/demo.rst", "src/pytorch-sphinx-theme/docs/demo/lists_tables.rst", "src/pytorch-sphinx-theme/docs/demo/long.rst", "src/pytorch-sphinx-theme/docs/demo/structure.rst", "src/pytorch-sphinx-theme/docs/index.rst", "src/pytorch-sphinx-theme/docs/installing.rst", "ts/creating_torchscript_module_in_python.rst", "ts/getting_started_with_cpp_api.rst", "ts/getting_started_with_python_api.rst", "ts/ptq.rst", "ts/torchscript_frontend_from_pytorch.rst", "tutorials/_rendered_examples/dynamo/custom_kernel_plugins.rst", "tutorials/_rendered_examples/dynamo/engine_caching_bert_example.rst", "tutorials/_rendered_examples/dynamo/engine_caching_example.rst", "tutorials/_rendered_examples/dynamo/index.rst", "tutorials/_rendered_examples/dynamo/mutable_torchtrt_module_example.rst", "tutorials/_rendered_examples/dynamo/refit_engine_example.rst", "tutorials/_rendered_examples/dynamo/torch_compile_advanced_usage.rst", "tutorials/_rendered_examples/dynamo/torch_compile_resnet_example.rst", "tutorials/_rendered_examples/dynamo/torch_compile_stable_diffusion.rst", "tutorials/_rendered_examples/dynamo/torch_compile_transformers_example.rst", "tutorials/_rendered_examples/dynamo/torch_export_cudagraphs.rst", "tutorials/_rendered_examples/dynamo/torch_export_gpt2.rst", "tutorials/_rendered_examples/dynamo/torch_export_llama2.rst", "tutorials/_rendered_examples/dynamo/vgg16_ptq.rst", "tutorials/_rendered_examples/index.rst", "tutorials/notebooks.rst", "tutorials/serving_torch_tensorrt_with_triton.rst", "user_guide/dynamic_shapes.rst", "user_guide/runtime.rst", "user_guide/saving_models.rst", "user_guide/torch_tensorrt_explained.rst", "user_guide/using_dla.rst"], "titles": ["Class DataType", "Class Device::DeviceType", "Class TensorFormat", "Template Class Int8CacheCalibrator", "Template Class Int8Calibrator", "Define STR", "Define TORCH_TENSORRT_PATCH_VERSION", "Define TORCH_TENSORRT_MAJOR_VERSION", "Define TORCH_TENSORRT_MINOR_VERSION", "Define TORCHTRT_API", "Define XSTR", "Define TORCHTRT_HIDDEN", "Define TORCH_TENSORRT_VERSION", "Directory cpp", "Directory include", "Directory torch_tensorrt", "Enum Level", "Enum EngineCapability", "File logging.h", "File macros.h", "File ptq.h", "File torch_tensorrt.h", "Function torch_tensorrt::logging::get_logging_prefix", "Function torch_tensorrt::logging::get_reportable_log_level", "Function torch_tensorrt::logging::get_is_colored_output_on", "Function torch_tensorrt::logging::set_reportable_log_level", "Function torch_tensorrt::logging::log", "Function torch_tensorrt::logging::set_is_colored_output_on", "Function torch_tensorrt::logging::set_logging_prefix", "Template Function torch_tensorrt::ptq::make_int8_cache_calibrator", "Template Function torch_tensorrt::ptq::make_int8_calibrator", "Function torch_tensorrt::torchscript::check_method_operator_support", "Function torch_tensorrt::torchscript::compile", "Function torch_tensorrt::torchscript::embed_engine_in_new_module", "Function torch_tensorrt::torchscript::convert_method_to_trt_engine", "Function torch_tensorrt::get_build_info", "Function torch_tensorrt::set_device", "Function torch_tensorrt::dump_build_info", "Namespace torch_tensorrt", "Namespace torch_tensorrt::logging", "Namespace torch_tensorrt::ptq", "Namespace torch_tensorrt::torchscript", "Program Listing for File logging.h", "Program Listing for File macros.h", "Program Listing for File ptq.h", "Program Listing for File torch_tensorrt.h", "Struct Device", "Struct GraphInputs", "Struct Input", "Struct CompileSpec", "Torch-TensorRT C++ API", "Full API", "torchtrtc", "Conversion Phase", "Writing Dynamo Converters", "Lowering Phase", "Partitioning Phase", "Compiler Phases", "Runtime Phase", "System Overview", "Writing TorchScript Converters", "Useful Links for Torch-TensorRT Development", "Writing Dynamo ATen Lowering Passes", "Compiling Exported Programs with Torch-TensorRT", "TensorRT Backend for torch.compile", "Torch-TensorRT (FX Frontend) User Guide", "Installation", "Quick Start", "Torch-TensorRT", "Operators Supported", "torch_tensorrt.dynamo", "torch_tensorrt.fx", "torch_tensorrt.logging", "torch_tensorrt.ts.ptq", "torch_tensorrt.runtime", "torch_tensorrt", "torch_tensorrt.ts", "Computation times", "Changelog", "Configuration", "5. :mod:`test_py_module`", "3. Paragraph Level Markup", "4. Lists & Tables", "1. Long Sticky Nav", "1. Structural Elements", "<no title>", "Installation", "Creating a TorchScript Module", "Using Torch-TensorRT in C++", "Using Torch-TensorRT in Python", "Post Training Quantization (PTQ)", "Using Torch-TensorRT TorchScript Frontend Directly From PyTorch", "Using Custom Kernels within TensorRT Engines with Torch-TensorRT", "Engine Caching (BERT)", "Engine Caching", "Dynamo / torch.compile", "Mutable Torch TensorRT Module", "Refitting Torch-TensorRT Programs with New Weights", "Torch Compile Advanced Usage", "Compiling ResNet using the Torch-TensorRT torch.compile Backend", "Torch Compile Stable Diffusion", "Compiling a Transformer using torch.compile and TensorRT", "Torch Export with Cudagraphs", "Compiling GPT2 using the Torch-TensorRT with dynamo backend", "Compiling Llama2 using the Torch-TensorRT with dynamo backend", "Deploy Quantized Models using Torch-TensorRT", "Torch-TensorRT Tutorials", "Example notebooks", "Serving a Torch-TensorRT model with Triton", "Dynamic shapes with Torch-TensorRT", "Deploying Torch-TensorRT Programs", "Saving models compiled with Torch-TensorRT", "Torch-TensorRT Explained", "DLA"], "terms": {"defin": [0, 1, 2, 3, 4, 33, 43, 46, 47, 48, 49, 51, 52, 54, 65, 67, 74, 75, 79, 87, 88, 89, 90, 92, 94, 98, 101, 102, 103, 104, 107], "file": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 46, 47, 48, 49, 52, 54, 56, 58, 59, 64, 65, 66, 67, 70, 71, 73, 75, 76, 77, 79, 80, 82, 86, 88, 90, 108, 109, 111], "torch_tensorrt": [0, 1, 2, 14, 16, 17, 42, 43, 44, 46, 47, 48, 49, 50, 51, 52, 54, 56, 62, 63, 64, 65, 67, 68, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 100, 103, 104, 105, 106, 107, 108, 109, 110, 111, 113], "h": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 15, 46, 47, 48, 49, 50, 51, 52, 55, 67, 75, 88, 90], "support": [0, 1, 2, 27, 31, 46, 48, 49, 52, 54, 56, 61, 63, 65, 67, 68, 71, 74, 75, 76, 79, 80, 87, 88, 89, 92, 97, 99, 101, 103, 104, 105, 108, 112, 113], "data": [0, 2, 3, 4, 29, 30, 44, 46, 48, 49, 52, 53, 56, 57, 59, 60, 64, 65, 69, 70, 71, 73, 75, 76, 81, 85, 89, 90, 92, 94, 105, 107], "type": [0, 1, 2, 30, 49, 50, 52, 53, 56, 58, 60, 62, 63, 64, 65, 70, 71, 73, 74, 75, 76, 81, 88, 89, 90, 92, 94, 105, 107, 111], "can": [0, 1, 4, 29, 30, 34, 46, 47, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 66, 67, 70, 73, 74, 75, 76, 79, 81, 87, 88, 89, 90, 91, 92, 94, 95, 96, 97, 98, 101, 102, 105, 106, 107, 108, 109, 110, 111, 112], "us": [0, 1, 2, 3, 4, 29, 30, 32, 34, 36, 43, 44, 45, 46, 48, 49, 52, 53, 54, 56, 58, 59, 60, 62, 63, 64, 65, 66, 67, 68, 70, 71, 73, 74, 75, 76, 77, 79, 80, 81, 82, 87, 90, 94, 95, 96, 97, 106, 108, 110, 111, 112, 113], "tensorrt": [0, 1, 3, 4, 29, 30, 31, 32, 33, 34, 37, 44, 45, 46, 48, 49, 52, 53, 54, 55, 56, 57, 59, 60, 62, 67, 70, 71, 73, 74, 75, 76, 77, 87, 90, 94, 95, 98, 100, 102], "engin": [0, 1, 17, 32, 33, 34, 45, 46, 48, 49, 52, 53, 56, 57, 59, 62, 63, 64, 70, 71, 74, 75, 76, 79, 88, 89, 90, 91, 95, 97, 99, 101, 106, 109, 110, 112, 113], "thi": [0, 1, 2, 29, 30, 42, 43, 44, 45, 46, 47, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 66, 67, 70, 71, 74, 75, 76, 79, 80, 81, 83, 84, 87, 88, 90, 91, 92, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112], "compat": [0, 1, 46, 55, 58, 64, 65, 70, 74, 75, 76, 112], "c10": [0, 1, 45, 46, 48, 49, 88, 90], "check": [0, 1, 31, 46, 52, 55, 60, 65, 70, 74, 76, 88, 92, 96, 97, 108, 110], "trt": [0, 1, 3, 4, 46, 48, 53, 55, 58, 60, 62, 64, 65, 67, 69, 70, 74, 75, 88, 92, 101, 103, 104, 109, 110, 111], "so": [0, 44, 52, 53, 54, 55, 58, 59, 60, 62, 64, 65, 66, 71, 74, 75, 80, 81, 82, 88, 90, 92, 94, 98, 99, 101, 103, 104, 109], "should": [0, 3, 4, 29, 45, 49, 52, 53, 54, 55, 56, 57, 59, 60, 63, 64, 65, 70, 74, 75, 76, 79, 81, 84, 90, 92, 93, 94, 97, 102, 108], "reason": [0, 65, 87, 92, 94, 112], "you": [0, 1, 2, 29, 30, 46, 48, 49, 52, 53, 54, 55, 56, 58, 59, 60, 63, 65, 66, 67, 70, 74, 75, 76, 79, 81, 82, 83, 87, 88, 89, 90, 91, 92, 94, 95, 96, 97, 105, 106, 107, 108, 109, 110, 111, 112], "need": [0, 1, 2, 25, 29, 43, 46, 53, 54, 55, 60, 65, 66, 70, 71, 74, 75, 81, 88, 89, 90, 92, 93, 94, 96, 97, 107, 108, 109, 110], "explictli": 0, "public": [0, 1, 2, 3, 4, 44, 45, 46, 47, 48, 49, 82, 90], "enum": [0, 1, 2, 42, 45, 46, 70, 76, 90], "valu": [0, 1, 2, 16, 17, 45, 46, 48, 53, 56, 58, 60, 63, 69, 70, 73, 75, 79, 88, 96, 98, 99, 101, 107], "underli": [0, 1, 2, 46, 60], "In": [0, 1, 2, 46, 53, 54, 56, 57, 58, 59, 60, 64, 65, 66, 74, 75, 81, 82, 84, 89, 90, 92, 95, 96, 106, 107, 108, 109, 110, 111], "case": [0, 1, 2, 46, 49, 53, 54, 56, 58, 60, 62, 64, 65, 66, 74, 75, 90, 92, 96, 97, 109, 110], "itself": [0, 1, 2, 46, 52, 55, 91, 108], "interfac": [0, 1, 2, 46, 58, 59, 60, 64, 68, 90], "vs": [0, 1, 2, 46, 55, 66, 70, 75, 76, 91], "normal": [0, 1, 2, 46, 65, 81, 87, 88, 90, 96, 97, 102, 105, 108, 113], "instatin": [0, 1, 2, 46], "ex": [0, 1, 2, 33, 46, 76, 82, 84], "kfloat": [0, 45, 49], "enumer": [0, 1, 2, 16, 17, 46], "klong": [0, 45], "int64": [0, 75, 76], "kdoubl": [0, 45], "fp64": [0, 75], "fp32": [0, 48, 49, 52, 64, 65, 70, 75, 76, 90, 107, 108], "khalf": [0, 45, 88], "fp16": [0, 48, 49, 52, 64, 65, 71, 75, 88, 89, 96, 100, 113], "kchar": [0, 45], "int8": [0, 44, 48, 49, 52, 64, 70, 75, 76, 90, 105, 113], "kint": [0, 45], "int": [0, 3, 4, 36, 44, 45, 49, 52, 54, 56, 63, 64, 69, 70, 71, 75, 76, 79, 88, 92, 105], "kbool": [0, 45], "bool": [0, 1, 2, 3, 4, 24, 27, 30, 31, 42, 44, 45, 46, 49, 55, 60, 64, 69, 70, 71, 73, 74, 75, 76, 79, 88, 90, 92], "kunknown": [0, 2, 45], "sentinel": [0, 2, 75], "function": [0, 1, 2, 3, 4, 46, 48, 49, 54, 55, 56, 58, 60, 62, 64, 65, 66, 87, 88, 90, 91, 92, 97, 98, 101, 102, 103, 104, 107, 108, 109, 110, 112, 113], "default": [0, 1, 2, 3, 4, 16, 29, 30, 33, 43, 45, 46, 48, 49, 52, 54, 56, 62, 64, 65, 66, 70, 71, 74, 75, 76, 79, 80, 81, 88, 89, 90, 91, 92, 94, 105, 109, 110, 111, 112], "construct": [0, 1, 2, 3, 4, 46, 48, 49, 53, 54, 55, 57, 59, 60, 65, 73, 74, 75, 81, 82, 88, 90, 92, 94, 109], "new": [0, 1, 2, 3, 4, 32, 33, 46, 48, 49, 56, 58, 59, 60, 62, 64, 65, 67, 70, 76, 81, 88, 94, 95, 96, 99, 101, 102, 106, 108, 110], "object": [0, 1, 2, 3, 4, 46, 48, 49, 52, 58, 60, 62, 63, 64, 70, 74, 75, 76, 90, 91, 109, 111], "inlin": [0, 1, 2, 3, 4, 29, 30, 44, 46, 48, 55, 82, 85, 88], "constexpr": [0, 1, 2, 45, 46, 92], "t": [0, 1, 2, 45, 46, 55, 60, 65, 66, 69, 75, 79, 81, 82, 87, 88, 90, 92, 105, 108, 109], "constructor": [0, 2, 46, 48, 49, 58, 87], "from": [0, 1, 2, 3, 4, 29, 30, 44, 46, 48, 49, 52, 53, 55, 56, 57, 58, 59, 60, 63, 64, 65, 68, 70, 71, 74, 75, 76, 77, 79, 80, 81, 82, 87, 88, 90, 92, 93, 94, 96, 97, 100, 101, 103, 104, 105, 107, 108, 110, 111, 112], "torchtrt_api": [0, 2, 19, 22, 23, 24, 25, 26, 27, 28, 31, 32, 33, 34, 35, 36, 37, 42, 43, 44, 45, 48, 49, 50], "scalartyp": [0, 45, 69], "torch": [0, 1, 2, 4, 20, 21, 29, 30, 31, 32, 33, 34, 37, 44, 45, 46, 47, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 70, 71, 73, 74, 75, 76, 77, 87, 90, 93, 94, 113], "paramet": [0, 1, 2, 3, 4, 25, 26, 27, 29, 30, 31, 32, 33, 34, 36, 46, 48, 49, 53, 54, 55, 60, 64, 65, 70, 71, 73, 74, 75, 76, 85, 87, 88, 97, 103, 104], "oper": [0, 1, 2, 3, 4, 31, 44, 45, 46, 49, 52, 53, 55, 56, 57, 58, 59, 60, 62, 63, 65, 68, 70, 75, 76, 89, 90, 97, 99, 101, 112, 113], "const": [0, 1, 2, 3, 4, 29, 30, 31, 32, 33, 34, 36, 44, 45, 46, 55, 60, 69, 88, 90], "get": [0, 1, 2, 3, 4, 23, 35, 44, 46, 55, 56, 60, 62, 63, 65, 74, 75, 88, 90, 94, 103, 104, 107, 108], "return": [0, 1, 2, 3, 4, 23, 24, 29, 30, 31, 32, 33, 34, 35, 42, 43, 44, 45, 46, 54, 55, 56, 57, 58, 59, 60, 62, 64, 65, 70, 71, 74, 75, 76, 87, 88, 89, 90, 92, 94, 97, 98, 105, 108, 109], "explicit": [0, 1, 2, 3, 4, 45, 46, 55, 65, 71, 74, 81, 90, 112], "delet": [0, 1, 2, 45, 46, 55], "other": [0, 1, 2, 45, 46, 52, 53, 55, 58, 62, 64, 65, 66, 69, 70, 74, 75, 80, 81, 88, 89, 110], "comparis": [0, 2], "true": [0, 1, 2, 4, 46, 49, 55, 56, 60, 62, 64, 65, 69, 70, 71, 74, 75, 76, 79, 82, 88, 90, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 108, 110, 113], "fals": [0, 1, 2, 3, 4, 44, 45, 46, 49, 54, 62, 64, 65, 69, 70, 71, 74, 75, 76, 79, 80, 81, 82, 88, 90, 91, 92, 93, 94, 96, 97, 98, 100, 101, 102, 103, 104, 105, 110], "struct": [1, 21, 38, 41, 45, 54, 90], "onli": [1, 3, 4, 16, 29, 44, 46, 48, 52, 54, 55, 56, 59, 60, 64, 65, 67, 70, 71, 74, 75, 81, 90, 92, 96, 97, 110, 113], "applic": [1, 29, 46, 52, 55, 59, 64, 70, 74, 75, 88, 89, 91, 110, 113], "kcuda": [1, 46, 56, 88], "which": [1, 2, 29, 32, 34, 46, 49, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 66, 70, 71, 73, 74, 75, 76, 79, 81, 82, 87, 88, 89, 90, 91, 92, 94, 98, 99, 102, 103, 104, 107, 108, 109, 110, 111, 112], "map": [1, 46, 53, 54, 55, 57, 59, 60, 65, 75, 88, 90, 91, 94, 98, 107, 108], "kgpu": [1, 45, 46], "To": [1, 46, 52, 54, 56, 64, 66, 70, 79, 87, 88, 89, 91, 92, 97, 108], "datatyp": [1, 21, 38, 45, 46, 48, 49, 50, 70, 75, 76, 89, 92, 108], "target": [1, 33, 45, 46, 48, 49, 52, 54, 56, 58, 59, 64, 65, 66, 68, 70, 74, 75, 76, 89, 90, 91, 92, 97, 112, 113], "gpu": [1, 32, 34, 36, 45, 46, 52, 64, 65, 70, 74, 75, 76, 88, 90, 91, 92, 103, 104, 108, 110, 112, 113], "run": [1, 34, 46, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 64, 65, 66, 67, 70, 71, 74, 75, 76, 81, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 107, 108, 109, 110, 111, 112, 113], "kdla": [1, 45, 46, 113], "dla": [1, 45, 46, 49, 52, 64, 68, 70, 75, 76], "intern": [1, 16, 46, 60, 63, 72, 74, 81, 88], "note": [1, 46, 48, 54, 60, 62, 65, 66, 74, 75, 79, 81, 88, 92, 97, 109, 113], "The": [1, 46, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 66, 70, 74, 75, 76, 79, 82, 87, 89, 90, 91, 92, 93, 94, 96, 97, 99, 102, 106, 107, 108, 109, 111, 112], "valid": [1, 46, 56, 60, 62, 70, 74, 75], "kcpu": [1, 46], "comparison": [1, 46], "an": [2, 3, 4, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 62, 64, 65, 66, 67, 70, 71, 73, 74, 75, 76, 79, 81, 82, 87, 88, 89, 90, 92, 94, 97, 98, 102, 103, 104, 107, 108, 109, 110, 111, 112], "memeori": 2, "layout": [2, 48, 69, 70, 75, 76], "store": [2, 4, 49, 52, 53, 58, 60, 64, 65, 70, 74, 75, 76, 87, 88, 92, 94, 97], "tensor": [2, 33, 44, 45, 48, 49, 52, 53, 54, 55, 56, 58, 60, 62, 63, 64, 65, 69, 70, 71, 74, 75, 76, 87, 88, 89, 90, 92, 98, 107], "kcontigu": [2, 45, 48], "contigu": [2, 48, 49, 52, 70, 75, 76], "nchw": [2, 70, 75, 76], "linear": [2, 56, 69, 75, 87, 92, 105], "kchannelslast": [2, 45], "channel": [2, 75, 80], "last": [2, 55, 65, 75, 105], "nhwc": [2, 52], "memoryformat": [2, 45], "ptq": [3, 4, 15, 18, 19, 38, 50, 51, 52, 68, 70, 75, 76, 95, 106], "privat": [3, 4, 44, 45, 90], "algorithm": [3, 4, 29, 30, 44, 65, 73, 90], "typenam": [3, 4, 29, 30, 44], "gener": [3, 4, 29, 52, 55, 58, 59, 60, 62, 64, 65, 66, 70, 71, 79, 81, 82, 85, 87, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 105, 106, 110], "int8calibr": [3, 20, 30, 40, 44, 50], "implement": [3, 4, 55, 56, 58, 63, 65, 74, 80, 88, 90, 92, 94, 110], "specifi": [3, 4, 33, 52, 54, 60, 64, 65, 66, 70, 75, 76, 79, 81, 89, 91, 108, 109, 111, 112], "calibr": [3, 4, 29, 30, 44, 49, 52, 70, 73, 75, 76, 88, 90], "read": [3, 4, 29, 30, 44, 79, 81, 90], "nvinfer1": [3, 4, 29, 30, 44, 45, 49, 60, 90], "iint8calibr": [3, 4, 29, 30, 44, 45, 49, 70, 75, 76, 90], "iint8entropycalibrator2": [3, 4, 29, 30, 44, 90], "std": [3, 4, 22, 26, 28, 29, 30, 31, 33, 34, 35, 42, 44, 45, 47, 48, 49, 56, 88, 90, 108, 113], "string": [3, 4, 18, 20, 21, 22, 26, 28, 29, 30, 31, 33, 34, 35, 42, 44, 45, 49, 54, 56, 58, 60, 64, 70, 75, 79, 88, 90], "cache_file_path": [3, 4, 29, 30, 44], "8": [3, 52, 55, 63, 64, 66, 74, 75, 81, 82, 85, 88, 92, 99, 102, 108, 109], "cach": [3, 4, 29, 30, 44, 52, 64, 65, 70, 71, 73, 75, 88, 90, 95, 106, 110], "getbatchs": [3, 4, 44], "noexceptoverrid": [3, 4], "batch": [3, 4, 44, 64, 65, 71, 74, 90, 94, 99, 101, 105, 108, 109, 113], "size": [3, 4, 44, 48, 49, 52, 55, 56, 64, 65, 69, 70, 71, 75, 76, 79, 88, 90, 92, 94, 99, 101, 105, 107, 109], "next": [3, 4, 53, 54, 58, 63, 71, 75, 79, 81, 82, 90, 98, 102, 105, 108], "alwai": [3, 4, 27, 52, 75, 81, 97], "1": [3, 4, 33, 44, 45, 48, 49, 52, 54, 55, 56, 58, 60, 62, 63, 64, 65, 66, 69, 70, 71, 73, 74, 75, 76, 78, 79, 81, 82, 85, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 99, 101, 102, 104, 105, 107, 109, 111, 113], "due": [3, 4, 66, 80, 81, 105], "issu": [3, 4, 64, 70, 75, 88, 98, 101], "getbatch": [3, 4, 44], "void": [3, 4, 25, 26, 27, 28, 36, 37, 42, 44, 45], "bind": [3, 4, 33, 44, 74, 76, 81], "char": [3, 4, 44, 52, 88], "name": [3, 4, 31, 33, 34, 44, 54, 56, 58, 60, 65, 66, 71, 73, 74, 75, 76, 81, 82, 87, 88, 91, 92, 97, 102, 105, 108], "nbbind": [3, 4, 44], "Not": 3, "arrai": [3, 4, 33, 53, 54, 75, 76, 92], "pointer": [3, 4, 90], "fed": [3, 4, 48], "buffer": [3, 4, 65, 92], "each": [3, 4, 49, 53, 55, 56, 58, 60, 64, 65, 66, 70, 71, 74, 79, 81, 88, 97, 110], "input": [3, 4, 21, 29, 33, 38, 44, 45, 47, 49, 50, 52, 53, 54, 55, 56, 58, 60, 62, 63, 64, 65, 67, 69, 70, 71, 72, 74, 75, 76, 82, 87, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 102, 103, 104, 105, 107, 108, 109, 110, 111, 112, 113], "number": [3, 4, 49, 52, 54, 55, 56, 60, 63, 64, 65, 70, 71, 75, 76, 79, 88, 89, 92, 95, 97, 99, 101, 106, 107, 112], "readcalibrationcach": [3, 4, 44], "size_t": [3, 4, 44, 90], "length": [3, 4, 44, 65, 69, 82], "how": [3, 4, 66, 81, 83, 85, 87, 91, 92, 94, 96, 98, 105, 107, 108, 109, 110], "enabl": [3, 4, 24, 49, 52, 54, 56, 57, 59, 64, 65, 66, 70, 71, 73, 74, 75, 76, 79, 94, 96, 97, 99, 101, 102, 110], "use_cach": [3, 4, 30, 44, 73, 90, 103, 104], "set": [3, 4, 16, 21, 25, 27, 29, 32, 34, 36, 45, 46, 48, 49, 52, 53, 54, 55, 56, 57, 58, 59, 65, 66, 70, 71, 74, 75, 76, 79, 83, 86, 87, 88, 89, 90, 92, 97, 105, 107, 109, 110, 112, 113], "writecalibrationcach": [3, 4, 44], "write": [3, 4, 29, 30, 44, 65, 68, 81, 88, 90, 108], "provid": [3, 4, 49, 52, 54, 56, 58, 60, 62, 64, 65, 66, 67, 70, 71, 74, 75, 76, 81, 88, 89, 90, 91, 94, 95, 97, 98, 102, 106, 108, 109, 110, 111, 112], "cast": [3, 4, 55], "convienc": [3, 4, 49], "convert": [3, 4, 31, 32, 34, 52, 55, 56, 57, 59, 63, 64, 68, 70, 75, 76, 89, 91, 92, 99, 101, 107, 110], "easili": [3, 4, 96], "assign": [3, 4, 80], "ptq_calibr": [3, 4, 45, 49, 90], "field": [3, 4, 63, 71, 75, 90], "compilespec": [3, 4, 21, 32, 34, 41, 45, 50, 56, 76, 88, 90, 113], "dataloaderuniqueptr": [4, 44], "libtorch": [4, 37, 60, 66, 67, 88, 90, 112], "dataload": [4, 29, 30, 44, 49, 73, 90, 105], "unique_ptr": [4, 30], "unqiue_ptr": 4, "A": [4, 29, 30, 32, 33, 47, 48, 54, 55, 56, 60, 65, 66, 70, 71, 75, 76, 82, 90, 100, 108], "uniqu": [4, 89], "what": [4, 54, 55, 65, 67, 75, 81, 87, 88, 89, 104, 112], "make_data_load": [4, 90], "factori": [4, 29, 30, 64, 70, 90], "path": [4, 13, 14, 15, 29, 30, 52, 64, 65, 66, 70, 73, 75, 87, 88, 90, 94, 102, 105, 108, 112], "find": [4, 65, 66, 88, 92], "whether": [4, 52, 54, 64, 65, 70, 71, 75, 80, 90, 99, 101, 110], "exist": [4, 31, 32, 34, 54, 63, 64, 65, 70, 73, 75, 76, 90, 94, 107], "There": [4, 53, 54, 59, 60, 62, 63, 65, 66, 82, 87, 90, 97, 107, 108, 109, 110], "consum": [4, 53, 87], "macro": [5, 6, 7, 8, 9, 10, 11, 12, 15, 18, 20, 21, 42, 44, 45, 50, 51], "x": [5, 10, 33, 43, 55, 56, 66, 67, 74, 76, 82, 87, 88, 92, 94, 98, 102, 105, 109, 111], "includ": [13, 15, 16, 35, 37, 42, 43, 44, 45, 51, 52, 54, 56, 57, 58, 59, 62, 64, 65, 66, 67, 70, 71, 74, 75, 79, 81, 87, 88, 90, 92, 95, 106, 110], "parent": [14, 15, 18, 19, 20, 21], "cpp": [14, 15, 42, 43, 44, 45, 51, 55, 59, 66, 88, 90], "log": [15, 16, 19, 20, 38, 44, 50, 51, 55, 60, 64, 65, 68, 69, 70, 71, 75, 99, 101], "emum": [16, 17], "messag": [16, 25, 26, 52, 72], "sever": [16, 26, 72], "kinternal_error": [16, 42], "print": [16, 31, 44, 62, 64, 70, 76, 81, 88, 91, 92, 93, 94, 96, 97, 99, 101, 103, 104, 105, 108], "error": [16, 49, 52, 53, 55, 59, 64, 65, 70, 72, 75, 76, 81, 88, 109], "kerror": [16, 42], "all": [16, 42, 43, 44, 45, 49, 52, 54, 55, 56, 58, 62, 64, 65, 66, 70, 72, 74, 75, 77, 81, 82, 87, 88, 89, 90, 92, 106, 107, 108, 110, 112], "kwarn": [16, 42], "warn": [16, 44, 52, 60, 72, 74], "kinfo": [16, 42, 44], "info": [16, 32, 34, 45, 52, 60, 72, 74, 75], "kdebug": [16, 42, 44], "debug": [16, 27, 45, 49, 52, 60, 62, 64, 70, 72, 74, 75, 76, 91, 92, 93, 94, 96, 97, 98, 99, 101, 105], "kgraph": [16, 42, 55], "everyth": [16, 64, 70, 75], "intermedi": [16, 49, 52, 54, 64, 70, 72, 75, 76, 87, 112], "graph": [16, 31, 32, 34, 45, 49, 52, 53, 54, 56, 57, 59, 60, 62, 63, 64, 65, 70, 71, 72, 75, 76, 87, 88, 92, 94, 95, 96, 97, 99, 101, 102, 106, 107, 109, 110], "lower": [16, 54, 63, 65, 68, 70, 71, 72, 75, 82, 92, 94, 99, 101, 107, 112], "phase": [16, 60, 63, 88, 97, 109, 112], "select": [17, 29, 30, 34, 49, 52, 58, 64, 65, 66, 69, 70, 75, 76, 80, 83, 89, 90, 92, 112], "capabl": [17, 45, 49, 52, 58, 70, 75, 76, 91], "kstandard": [17, 45, 49], "ksafeti": [17, 45], "kdla_standalon": [17, 45], "directori": [18, 19, 20, 21, 42, 43, 44, 45, 50, 66, 70, 90, 94], "program": [18, 19, 20, 21, 29, 51, 52, 57, 58, 59, 68, 70, 87, 94, 95, 104, 106, 109], "list": [18, 19, 20, 21, 31, 49, 51, 53, 56, 58, 60, 62, 63, 65, 67, 69, 70, 71, 74, 75, 76, 85, 88, 89, 92, 108], "torchscript": [19, 21, 38, 43, 45, 49, 50, 52, 56, 57, 58, 59, 63, 67, 70, 71, 73, 74, 75, 76, 89, 107, 109, 113], "str": [19, 43, 44, 50, 54, 64, 65, 69, 70, 73, 74, 75, 76, 92, 94, 105], "torch_tensorrt_major_vers": [19, 43, 50], "torch_tensorrt_minor_vers": [19, 43, 50], "torch_tensorrt_patch_vers": [19, 43, 50], "torch_tensorrt_vers": [19, 43, 50], "torchtrt_hidden": [19, 43, 50], "xstr": [19, 43, 50], "nvinfer": [20, 44], "fstream": [20, 44], "iostream": [20, 21, 44, 45, 88], "iter": [20, 44, 49, 52, 53, 64, 70, 73, 75, 76, 93, 94, 105], "memori": [20, 21, 44, 45, 55, 60, 70, 75, 76, 88, 89, 92, 94, 103, 104], "sstream": [20, 44], "vector": [20, 21, 33, 44, 45, 47, 48, 49, 56, 58, 75, 88, 90, 113], "templat": [20, 40, 44, 45, 50, 79, 88], "int8cachecalibr": [20, 29, 40, 44, 50], "cuda_runtim": [21, 45], "custom_class": [21, 45], "devic": [21, 33, 36, 38, 45, 49, 50, 52, 58, 64, 69, 70, 71, 73, 74, 75, 76, 89, 90, 91, 92, 96, 100, 103, 104, 107, 113], "graphinput": [21, 38, 45, 49, 50], "devicetyp": [21, 38, 45, 46, 50, 74, 75, 76, 90, 91, 92, 113], "tensorformat": [21, 38, 45, 48, 50, 75, 92], "level": [23, 25, 26, 39, 42, 44, 50, 54, 55, 56, 59, 64, 65, 70, 75, 76, 85, 87, 92, 108, 112], "current": [23, 54, 56, 58, 60, 62, 63, 64, 65, 66, 70, 71, 74, 75, 76, 79, 92, 96, 103, 104, 105, 110], "report": [23, 44, 74], "Is": [24, 75], "color": [24, 27, 81], "output": [24, 27, 33, 49, 52, 53, 54, 55, 56, 58, 60, 62, 63, 64, 65, 66, 70, 72, 74, 75, 76, 79, 81, 82, 88, 92, 94, 96, 97, 100, 107, 108, 109, 111], "lvl": [25, 26, 42], "inform": [25, 33, 35, 37, 48, 52, 53, 56, 58, 62, 64, 65, 66, 70, 71, 72, 75, 81, 87, 88, 90, 91, 92, 94, 109], "ad": [25, 52, 53, 54, 56, 62, 65, 66, 92, 96], "abov": [25, 54, 56, 62, 65, 66, 72, 80, 81, 88, 92, 99, 101, 111], "msg": [26, 42], "add": [26, 53, 54, 55, 56, 60, 63, 66, 69, 79, 81, 86, 88, 89, 92], "global": [26, 52, 64, 70, 75, 88], "colored_output_on": [27, 42], "prefix": [27, 28, 42, 81], "help": [27, 52, 53, 60, 64, 65, 88, 94, 105, 107, 110], "when": [27, 44, 45, 46, 52, 53, 55, 56, 57, 58, 59, 60, 64, 65, 66, 70, 74, 75, 76, 79, 81, 83, 87, 88, 90, 92, 94, 96, 97, 107, 109, 110], "termin": [27, 52, 88], "If": [27, 33, 53, 54, 55, 56, 62, 63, 64, 65, 66, 67, 70, 71, 75, 79, 81, 88, 89, 90, 92, 94, 97, 98, 102, 108, 109, 110, 112, 113], "build": [29, 30, 35, 49, 52, 53, 57, 59, 60, 63, 64, 65, 70, 74, 75, 80, 85, 88, 90, 92, 99, 101, 109], "post": [29, 30, 49, 52, 63, 68, 88, 94], "train": [29, 30, 49, 52, 68, 69, 88, 89, 94], "quantiz": [29, 30, 52, 64, 68, 73, 75, 88, 95, 106], "creat": [29, 30, 33, 52, 53, 54, 56, 58, 60, 65, 68, 75, 76, 81, 88, 92, 95, 97, 106, 108], "previous": [29, 33, 88, 94, 97], "therefor": [29, 58, 65, 66, 74, 81, 88, 107, 110], "have": [29, 33, 44, 52, 53, 54, 55, 56, 60, 62, 63, 65, 66, 70, 71, 73, 74, 75, 76, 81, 87, 88, 89, 90, 92, 99, 101, 105, 107, 108, 109], "requir": [29, 49, 52, 53, 54, 55, 63, 64, 65, 66, 70, 75, 76, 79, 88, 90, 92, 105, 108, 109, 110], "dataset": [29, 73, 90, 107], "save": [29, 44, 52, 58, 64, 65, 67, 68, 70, 74, 75, 76, 88, 89, 93, 94, 97, 100, 107, 108, 110, 112], "later": [29, 70, 88, 97, 111, 112], "differ": [29, 55, 56, 59, 64, 65, 66, 75, 79, 87, 92, 94, 96, 110, 112], "scratch": [29, 94, 97], "depend": [29, 35, 53, 59, 64, 65, 67, 70, 88, 89, 108, 110], "howev": [29, 66, 79, 80, 88, 92, 94, 108, 109, 112], "network": [29, 30, 54, 60, 65, 75, 88, 90, 92, 107, 108, 113], "also": [29, 53, 54, 60, 62, 64, 66, 67, 79, 81, 82, 88, 89, 90, 94, 102, 105, 106, 107], "recalibr": 29, "its": [29, 53, 56, 58, 60, 66, 74, 75, 81, 92, 105, 108, 110, 112], "structur": [29, 46, 49, 56, 59, 60, 64, 70, 75, 79, 81, 85, 87, 92, 108], "chang": [29, 55, 56, 59, 62, 64, 65, 74, 75, 76, 79, 90, 94, 96, 97, 108, 110, 112], "respons": [29, 54, 58, 81, 110], "ensur": [29, 54, 55, 56, 62, 64, 66, 74], "By": [29, 30, 51, 56, 66, 79, 87, 94, 109], "entropi": [29, 30, 90], "v2": [29, 30, 81], "perform": [29, 30, 54, 62, 63, 70, 74, 75, 90, 92, 102, 107, 108, 110, 111, 112], "recommend": [29, 30, 65, 66, 75, 81, 88, 92, 108, 109], "feed": [29, 30, 88], "forward": [29, 30, 32, 33, 56, 58, 60, 64, 67, 70, 74, 75, 76, 87, 88, 89, 90, 91, 92, 98, 105, 109], "overrid": [29, 30, 44, 54, 65, 90], "minmax": [29, 30, 90], "recomend": [29, 30], "nlp": [29, 30, 90], "task": [29, 30, 65, 90, 107], "call": [29, 30, 32, 49, 54, 55, 58, 60, 65, 70, 71, 74, 75, 76, 81, 87, 88, 91, 92, 94, 96, 98, 101, 107, 109, 110, 112], "make_int8_calibr": [29, 40, 44, 50, 90], "class": [29, 30, 44, 45, 46, 51, 58, 60, 64, 65, 72, 76, 81, 82, 87, 88, 89, 90, 92, 94, 98, 105, 107, 109], "e": [29, 30, 52, 55, 60, 65, 66, 67, 71, 75, 87, 88, 90, 92, 94, 97], "g": [29, 30, 52, 55, 65, 66, 71, 75, 81, 90, 92, 97], "iint8minmaxcalibr": [29, 30, 90], "calibration_cache_fil": [29, 30, 90], "move": [30, 44, 55, 58, 76, 88, 90, 103, 104], "calibration_dataload": [30, 90], "contain": [30, 31, 52, 53, 54, 55, 56, 60, 65, 66, 71, 74, 75, 81, 82, 87, 88, 90, 92, 94, 108, 110], "jit": [31, 32, 33, 34, 45, 47, 49, 52, 53, 55, 56, 57, 58, 59, 60, 61, 64, 67, 68, 70, 74, 75, 76, 87, 88, 89, 91, 92, 97, 108, 111, 112], "modul": [31, 32, 33, 34, 45, 49, 52, 56, 57, 58, 59, 60, 64, 65, 66, 67, 68, 70, 71, 73, 74, 75, 76, 80, 81, 82, 89, 90, 91, 92, 95, 97, 98, 105, 106, 107, 109, 111, 113], "method_nam": [31, 34, 45, 52, 75, 76, 88], "see": [31, 55, 56, 58, 62, 64, 65, 66, 75, 76, 81, 87, 88, 89, 92, 94, 97, 98], "fulli": [31, 52, 55, 64, 70, 74, 75, 76, 88, 90, 92, 113], "compil": [31, 34, 41, 45, 49, 50, 52, 54, 55, 56, 58, 60, 62, 65, 70, 71, 72, 74, 75, 76, 77, 79, 87, 89, 90, 91, 92, 93, 96, 105, 108, 110, 113], "take": [31, 32, 33, 34, 53, 54, 57, 58, 59, 60, 62, 65, 70, 71, 74, 75, 76, 79, 81, 88, 90, 91, 92, 98, 107, 109], "method": [31, 32, 33, 34, 48, 52, 55, 60, 66, 70, 75, 76, 81, 87, 88, 91, 94, 107], "pure": [31, 70, 75], "Will": 31, "out": [31, 44, 53, 55, 56, 57, 59, 60, 64, 66, 70, 75, 76, 81, 88, 92, 96, 105, 108, 109], "unsupport": [31, 49, 54, 64, 75, 92, 112], "script": [31, 55, 56, 67, 75, 76, 87, 88, 89, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 110, 112], "nvidia": [32, 34, 42, 43, 44, 45, 52, 61, 64, 65, 66, 70, 75, 76, 88, 98, 101, 108, 112, 113], "configur": [32, 34, 48, 62, 66, 70, 74, 75, 76, 85, 88, 90, 92, 108, 109], "equival": [32, 57, 59, 60, 70, 75, 76, 87, 88, 90, 92, 99, 101], "specif": [32, 49, 54, 55, 57, 59, 62, 64, 70, 75, 76, 81, 106, 107, 112], "traget": 32, "input_binding_nam": [33, 45, 74, 76], "output_binding_nam": [33, 45, 74, 76], "emb": [33, 52, 63, 76, 82], "pre": [33, 55, 73, 76, 90, 94, 110], "built": [33, 52, 58, 59, 64, 66, 70, 74, 75, 76, 94, 97], "serial": [33, 34, 52, 57, 59, 66, 70, 74, 75, 76, 88, 92, 94, 112], "regist": [33, 54, 58, 60, 65, 74, 76, 92], "execut": [33, 49, 52, 55, 57, 58, 59, 63, 64, 65, 66, 68, 70, 71, 74, 75, 76, 77, 87, 88, 90, 92, 108], "must": [33, 48, 49, 52, 54, 55, 56, 60, 62, 65, 66, 70, 71, 75, 76, 81, 82, 88, 94, 109, 110, 112], "follow": [33, 52, 54, 56, 58, 62, 63, 64, 65, 66, 76, 79, 81, 82, 86, 87, 88, 90, 92, 94, 95, 99, 106, 107, 108, 109, 110], "format": [33, 45, 48, 49, 52, 69, 70, 75, 76, 81, 82, 89, 92, 94, 105, 107, 108, 111], "symbol": [33, 65, 66, 76, 81, 110], "index": [33, 61, 62, 66, 68, 69, 76, 79, 85, 90, 92], "0": [33, 43, 44, 45, 49, 52, 54, 56, 59, 60, 62, 64, 65, 66, 68, 69, 70, 71, 73, 74, 75, 76, 77, 78, 80, 81, 88, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 108, 109, 113], "2": [33, 43, 54, 56, 60, 63, 64, 65, 66, 68, 69, 70, 73, 74, 75, 76, 79, 81, 82, 85, 87, 88, 90, 92, 93, 94, 95, 96, 97, 98, 99, 101, 104, 105, 106, 109, 111], "y": [33, 56, 76, 82, 92, 98], "compilesepc": 33, "order": [33, 49, 54, 56, 60, 62, 65, 66, 70, 71, 74, 75, 76, 88, 89, 94], "pass": [33, 53, 54, 56, 57, 58, 59, 60, 63, 64, 65, 66, 68, 72, 73, 74, 75, 76, 87, 88, 90, 92, 94, 97], "origin": [33, 65, 71, 75, 92, 94, 96, 112], "pytorch": [33, 48, 49, 52, 54, 55, 56, 57, 58, 59, 60, 63, 64, 66, 67, 70, 73, 74, 75, 76, 87, 88, 89, 90, 94, 95, 96, 97, 105, 106, 108, 109, 110, 111, 112], "assum": [33, 74, 91, 92], "convent": 33, "below": [33, 56, 60, 62, 63, 64, 65, 66, 81, 88, 89, 94, 100, 108], "equivil": 34, "librari": [35, 42, 43, 44, 45, 52, 54, 57, 58, 59, 60, 75, 88, 92], "version": [35, 37, 59, 62, 64, 65, 70, 74, 75, 79, 82, 92, 107, 108, 111], "gpu_id": [36, 45, 46, 52, 74, 75, 76, 90, 91, 92, 113], "id": [36, 45, 52, 75, 79, 80, 84, 113], "cudasetdevic": 36, "dump": [37, 52, 92], "base": [37, 50, 58, 63, 64, 66, 70, 71, 75, 81, 87, 89, 90, 93, 97, 101, 107, 112], "stdout": [37, 74], "enginecap": [38, 45, 49, 50, 64, 70, 74, 75, 76, 91, 92], "dump_build_info": [38, 45, 50], "get_build_info": [38, 45, 50], "set_devic": [38, 45, 50, 110], "get_is_colored_output_on": [39, 42, 50], "get_logging_prefix": [39, 42, 50], "get_reportable_log_level": [39, 42, 50], "set_is_colored_output_on": [39, 42, 50], "set_logging_prefix": [39, 42, 50], "set_reportable_log_level": [39, 42, 50], "make_int8_cache_calibr": [40, 44, 50, 90], "check_method_operator_support": [41, 45, 50], "convert_method_to_trt_engin": [41, 45, 50, 75, 76, 88, 91], "embed_engine_in_new_modul": [41, 45, 50, 76], "document": [42, 43, 44, 45, 50, 59, 79, 81, 82, 86, 87, 88, 90, 91, 108, 109, 110], "copyright": [42, 43, 44, 45, 82, 88], "c": [42, 43, 44, 45, 52, 59, 64, 69, 70, 71, 74, 75, 82, 89, 92, 96, 108, 110, 113], "corpor": [42, 43, 44, 45], "right": [42, 43, 44, 45, 55, 59, 60, 81], "reserv": [42, 43, 44, 45, 103, 104], "licens": [42, 43, 44, 45, 88], "under": [42, 43, 44, 45, 59, 65, 81, 99, 112], "bsd": [42, 43, 44, 45], "style": [42, 43, 44, 45, 64, 67, 79, 81, 82], "found": [42, 43, 44, 45, 63, 66, 74, 81, 88, 90, 92, 94, 110], "root": [42, 43, 44, 45, 66, 79, 90, 105], "sourc": [42, 43, 44, 45, 54, 59, 64, 65, 70, 71, 72, 73, 74, 75, 76, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "tree": [42, 43, 44, 45, 79, 90, 105, 110], "pragma": [42, 43, 44, 45, 90], "onc": [42, 43, 44, 45, 53, 55, 56, 58, 64, 65, 66, 75, 90, 92, 108, 110], "namespac": [42, 43, 44, 45, 51, 55, 68, 75, 90, 92], "ar": [42, 46, 49, 52, 53, 54, 55, 56, 58, 59, 60, 62, 63, 64, 65, 66, 70, 73, 74, 75, 76, 79, 81, 82, 83, 87, 88, 90, 91, 92, 93, 94, 96, 97, 99, 107, 108, 109, 110, 111, 112], "ones": [42, 56, 57, 59, 66, 81, 88, 92, 112], "necessari": [42, 62, 64, 66, 74, 97, 110], "user": [42, 48, 54, 56, 57, 58, 59, 62, 63, 64, 66, 70, 81, 82, 88, 89, 90, 94, 97, 106, 108, 109, 110, 112], "dont": 42, "know": [42, 60, 79, 81, 92], "we": [42, 44, 53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 64, 65, 71, 74, 79, 81, 87, 88, 90, 92, 94, 95, 96, 97, 98, 99, 101, 102, 103, 104, 105, 106, 107, 108, 109, 111, 112], "want": [42, 56, 65, 66, 67, 71, 87, 88, 90, 91, 92, 97, 98, 108], "use_cmake_generated_export_head": 43, "torch_tensorrt_export": 43, "els": [43, 44, 48, 76, 81, 82, 93, 94, 105], "__gnuc__": 43, "__attribute__": 43, "__visibility__": 43, "hidden": [43, 79], "endif": [43, 44, 45], "doe": [43, 44, 55, 56, 60, 62, 65, 66, 75, 81, 90, 92, 99, 101], "gaurd": 43, "someth": [43, 55, 81, 108], "5": [43, 52, 56, 58, 59, 64, 65, 66, 70, 74, 75, 81, 82, 85, 87, 88, 92, 96, 98, 108], "setup": [43, 90, 108], "alias": 43, "eas": 43, "ts": [43, 52, 56, 67, 68, 75, 87, 88, 89, 91, 109, 111], "torchtrt": [43, 56, 92, 105], "ifndef": [44, 45], "doxygen_should_skip_thi": [44, 45], "get_batch_impl": 44, "element_typ": 44, "super": [44, 87, 92, 98, 105, 109], "batchtyp": 44, "dataloader_": 44, "cache_file_path_": 44, "use_cache_": 44, "auto": [44, 56, 60, 64, 67, 70, 81, 82, 88, 90, 103, 104, 113], "batched_data_": 44, "push_back": [44, 56], "it_": 44, "begin": [44, 65, 66, 81, 98, 102], "noexcept": [44, 90], "hack": 44, "explict": 44, "work": [44, 55, 59, 60, 64, 65, 67, 70, 73, 74, 75, 81, 82, 90, 92, 97, 98, 102, 109], "here": [44, 53, 54, 56, 58, 63, 64, 65, 66, 67, 79, 81, 82, 87, 88, 90, 92, 102, 103, 104, 105, 108, 109, 110, 111], "explic": 44, "just": [44, 45, 55, 56, 64, 65, 68, 72, 74, 81, 83, 87, 88, 89, 91, 92, 94, 96, 107, 110], "still": [44, 56, 65, 66, 90, 98, 112], "static_cast": 44, "option": [44, 48, 52, 56, 57, 59, 62, 63, 64, 65, 70, 74, 75, 76, 81, 85, 90, 92, 93, 94, 98, 100, 110, 111, 113], "batch_siz": [44, 90, 105], "end": [44, 52, 60, 62, 69, 70, 75, 76, 81, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "statu": [44, 82], "reset": [44, 93, 94, 98, 101, 110], "incas": 44, "go": [44, 55, 56, 65, 67, 87, 88, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 107, 108, 112], "again": [44, 58, 60, 81, 92, 96], "stringstream": 44, "ss": 44, "cache_": 44, "clear": 44, "ifstream": 44, "io": [44, 108], "binari": [44, 90], "noskipw": 44, "good": [44, 60, 65, 81, 94], "copi": [44, 60, 65, 69, 73, 82, 108], "istream_iter": 44, "back_insert": 44, "nullptr": [44, 45, 49], "ofstream": [44, 88], "cache_fil": [44, 73, 90], "reinterpret_cast": 44, "cache_size_": 44, "int8_t": 45, "arrayref": [45, 48, 49], "friend": 45, "ostream": 45, "os": [45, 94], "dtype": [45, 48, 49, 52, 63, 64, 65, 69, 70, 71, 74, 75, 76, 89, 92, 93, 99, 101, 102, 107, 109], "device_typ": [45, 46, 75, 90, 91, 113], "int64_t": [45, 46, 48, 49, 90, 113], "core": [45, 52, 55, 56, 59, 64, 70, 75, 88, 112, 113], "agx": 45, "platform": [45, 52, 59, 66, 108, 113], "xavier": [45, 113], "dla_cor": [45, 46, 52, 75, 90, 91, 113], "allow_gpu_fallback": [45, 46, 70, 75, 76, 90, 91, 113], "customclasshold": [45, 48], "min_shap": [45, 48, 63, 65, 70, 75, 76, 89, 99, 102, 107, 109], "opt_shap": [45, 48, 63, 70, 75, 76, 89, 99, 102, 107, 109], "max_shap": [45, 48, 63, 65, 70, 75, 76, 89, 99, 102, 107, 109], "shape": [45, 47, 48, 49, 52, 56, 60, 63, 65, 68, 69, 70, 71, 74, 75, 76, 89, 92, 102, 105, 108, 110, 113], "doubl": [45, 48, 49, 52, 63, 70, 75, 76, 81, 110], "tensor_domain": [45, 48, 75], "input_is_dynam": 45, "ivalu": [45, 47, 49, 53, 58, 60, 88], "input_signatur": [45, 47, 49, 76, 89], "nest": [45, 49, 50, 81, 82], "full": [45, 49, 52, 60, 64, 70, 72, 75, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 108, 110, 113], "spec": [45, 48, 49, 52, 72, 75, 76, 91, 94], "flatten": [45, 47, 69, 87, 88, 105], "fixed_s": [45, 49], "reflect": [45, 75], "builderconfig": 45, "graph_input": [45, 49], "enabled_precis": [45, 49, 63, 64, 70, 74, 75, 76, 88, 89, 90, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 103, 104, 105, 108, 113], "disable_tf32": [45, 49, 64, 70, 74, 75, 76, 90, 92, 103, 104], "sparse_weight": [45, 49, 64, 65, 70, 74, 75, 76, 92], "refit": [45, 49, 64, 70, 75, 76, 91, 92, 94, 95, 96, 106], "truncate_long_and_doubl": [45, 49, 63, 64, 76, 100], "allow_shape_tensor": [45, 49, 76], "uint64_t": [45, 49], "num_avg_timing_it": [45, 49, 64, 70, 74, 75, 76, 91, 92], "workspace_s": [45, 49, 52, 64, 70, 74, 75, 76, 92, 97, 99, 101], "dla_sram_s": [45, 49, 52, 64, 70, 74, 75, 76, 92], "1048576": [45, 49, 64, 70, 74, 75, 76, 92], "dla_local_dram_s": [45, 49, 52, 64, 70, 74, 75, 76, 92], "1073741824": [45, 49, 64, 70, 74, 75, 76, 92], "dla_global_dram_s": [45, 49, 52, 64, 70, 74, 75, 76, 92], "536870912": [45, 49, 64, 70, 74, 75, 76, 92], "require_full_compil": [45, 49, 64, 70, 74, 75, 76, 92], "min_block_s": [45, 49, 56, 63, 64, 70, 74, 75, 76, 92, 93, 94, 97, 98, 99, 101, 104, 105], "3": [45, 49, 52, 55, 56, 58, 63, 64, 65, 67, 69, 70, 73, 75, 76, 81, 82, 85, 87, 88, 90, 91, 92, 93, 94, 96, 97, 99, 102, 105, 107, 109, 111, 113], "torch_executed_op": [45, 49, 56, 63, 64, 70, 74, 75, 76, 92, 97, 98, 99, 101], "torch_executed_modul": [45, 49, 56, 70, 75, 76], "member": [46, 47, 48, 49], "hold": [46, 47, 48, 53, 60, 75, 90], "relat": [46, 81, 98, 101], "let": [46, 52, 55, 60, 65, 70, 75, 76, 79, 81, 107, 108, 112], "layer": [46, 49, 52, 53, 55, 60, 62, 64, 65, 70, 74, 75, 76, 88, 90, 92, 105, 107, 108, 109, 112, 113], "thei": [46, 52, 53, 54, 55, 58, 60, 65, 73, 74, 75, 79, 81, 89, 94], "complex": [47, 49, 64, 66, 87, 89, 96], "either": [47, 48, 52, 60, 62, 70, 75, 76, 79, 81, 87, 88, 89, 92, 94, 111], "one": [47, 54, 55, 60, 64, 65, 70, 74, 75, 81, 87, 88, 89, 92, 98, 101, 103, 104, 108], "rang": [48, 49, 52, 65, 75, 92, 93, 94, 99, 107, 109], "optim": [48, 52, 63, 64, 65, 68, 70, 71, 73, 75, 87, 88, 89, 97, 99, 100, 101, 107, 109, 112], "profil": [48, 71, 74], "singl": [48, 52, 55, 56, 65, 75, 81, 87, 88, 90, 110], "repres": [48, 49, 54, 60, 65, 67, 81], "signifi": [48, 55], "static": [48, 49, 53, 60, 63, 64, 70, 75, 76, 79, 88, 105, 109], "three": [48, 57, 59, 65, 71, 75, 81, 82, 107, 108], "min": [48, 52, 60, 69, 75, 94, 99, 109], "optimin": 48, "max": [48, 52, 60, 69, 75, 79, 94, 99, 105, 109], "allow": [48, 49, 52, 53, 54, 55, 56, 62, 64, 65, 66, 70, 75, 76, 79, 92, 94, 97, 99, 101, 110], "argument": [48, 52, 54, 55, 58, 60, 62, 64, 65, 70, 74, 75, 76, 81, 82, 88, 89, 92, 109], "expect": [48, 54, 55, 60, 75, 88, 89, 107], "tradit": [48, 70, 75, 76, 90], "convect": 48, "produc": [48, 53, 54, 58, 60, 63, 75, 81, 88, 107], "low": [48, 65, 96], "high": [48, 55, 56, 79, 92, 112], "weight": [48, 49, 52, 53, 64, 65, 69, 70, 75, 76, 81, 88, 94, 95, 96, 100, 106, 107], "first": [48, 53, 54, 55, 65, 67, 81, 82, 88, 89, 90, 92, 94, 96, 98, 108, 109, 111, 112], "calcul": [48, 53, 56, 88, 92], "detect": [48, 58, 75], "float32": [48, 49, 52, 63, 64, 65, 70, 75, 76, 92, 96, 100, 103, 104, 109], "dynam": [48, 49, 63, 65, 68, 70, 71, 75, 76, 94, 98, 100, 101, 104, 110], "opt": [48, 66, 74, 75, 102], "minimum": [48, 49, 52, 56, 63, 64, 70, 75, 76, 92], "maximum": [48, 49, 52, 64, 65, 70, 71, 75, 76, 99, 101, 108], "accept": [48, 52, 54, 58, 60, 66, 75, 88, 89, 98, 111], "exampl": [48, 56, 58, 59, 60, 65, 66, 68, 70, 72, 74, 75, 76, 77, 79, 80, 82, 85, 87, 88, 89, 90, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 108, 109, 110, 111], "s": [48, 49, 53, 56, 58, 60, 63, 65, 66, 68, 70, 71, 74, 75, 79, 81, 82, 87, 88, 90, 92, 94, 107, 108, 109, 110, 111], "cannot": [48, 55, 56, 65, 66, 70, 74, 75, 76, 80, 87, 92], "through": [48, 53, 54, 55, 56, 58, 64, 65, 70, 72, 73, 81, 88, 89, 92, 96, 97, 107, 112], "altern": [48, 56, 62, 63, 75, 89, 102, 107, 111], "refer": [48, 54, 57, 59, 65, 80, 85, 88, 90, 92, 105, 108, 109, 111], "given": [48, 49, 52, 54, 55, 65, 70, 71, 73, 75, 76, 87, 88, 89, 91, 109], "kernel": [48, 49, 52, 60, 64, 65, 70, 75, 76, 95, 106, 110], "ani": [48, 52, 53, 54, 60, 62, 64, 65, 69, 70, 73, 74, 75, 76, 79, 81, 88, 89, 90, 92, 99, 109], "event": [48, 64, 93, 94], "place": [48, 55, 62, 65, 81, 82, 83, 90, 92, 105], "variabl": [48, 65, 74, 75], "dimens": [48, 55, 65, 71, 75, 99, 107, 109], "domain": [48, 75, 82, 90], "convien": 49, "fix": [49, 65, 81, 92, 110, 113], "describ": [49, 56, 60, 75, 87, 91, 95, 106, 108], "entri": [49, 60, 94], "okai": 49, "ha": [49, 53, 54, 55, 56, 57, 59, 60, 62, 64, 65, 66, 70, 71, 75, 81, 82, 87, 88, 90, 94, 97, 105, 107, 109, 112], "flaten": 49, "precis": [49, 52, 63, 64, 65, 70, 75, 88, 89, 90, 99, 101, 113], "dure": [49, 52, 54, 56, 60, 63, 64, 70, 73, 75, 90, 107, 109, 110], "prevent": [49, 52, 54, 56], "tf32": [49, 52, 64, 70], "comput": [49, 64, 65, 66, 70, 74, 81, 90, 107], "inner": [49, 82, 107], "product": [49, 75], "round": [49, 70, 75, 76, 92], "10": [49, 66, 70, 71, 75, 76, 85, 87, 88, 90, 105, 107, 108, 109], "bit": [49, 60, 65, 66, 70, 75, 76, 88], "mantissa": [49, 70, 75, 76], "befor": [49, 54, 55, 56, 59, 60, 65, 70, 75, 76, 88, 108, 109], "multipli": [49, 70, 75, 76], "accumul": [49, 70, 75, 76], "sum": [49, 65, 69, 70, 75, 76, 92, 105], "23": [49, 55, 70, 75, 76, 82], "behavior": [49, 56, 65, 70, 75, 76, 109, 110, 111], "sparsiti": [49, 52, 65, 70, 75, 76], "conv": [49, 52, 88, 92], "fc": [49, 52, 55], "truncat": [49, 52, 63, 64, 70, 75, 76], "long": [49, 52, 53, 63, 75, 81, 82], "float": [49, 52, 63, 64, 69, 75, 87, 88, 89, 90, 91, 92, 93, 94, 97, 98, 101, 102], "ishap": 49, "restrict": [49, 64, 70, 75, 76, 109], "cuda": [49, 58, 63, 65, 67, 70, 71, 74, 75, 88, 89, 90, 91, 92, 93, 94, 96, 97, 99, 100, 102, 103, 104, 105, 108, 109, 110, 111], "safeti": [49, 52, 75], "averag": [49, 52, 64, 70, 75, 76, 92], "time": [49, 52, 53, 54, 55, 56, 57, 58, 59, 60, 64, 65, 66, 67, 68, 70, 71, 74, 75, 76, 79, 81, 88, 90, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "workspac": [49, 52, 64, 65, 66, 70, 71, 75, 76, 92, 98, 99, 101], "fast": [49, 52, 64, 67, 70, 75, 76], "softwar": [49, 52, 64, 70, 75, 76, 81], "manag": [49, 52, 53, 55, 57, 59, 60, 64, 70, 72, 74, 75, 76, 88, 102, 110], "ram": [49, 52, 64, 70, 75, 76], "commun": [49, 52, 64, 70, 75, 76, 88], "within": [49, 52, 57, 59, 64, 70, 74, 75, 76, 79, 81, 95, 106], "host": [49, 52, 64, 66, 70, 75, 76, 92, 108], "share": [49, 52, 64, 66, 70, 74, 75, 76, 94], "across": [49, 52, 55, 56, 64, 70, 75, 76, 79], "metadata": [49, 52, 54, 58, 60, 64, 70, 75, 76, 79, 97, 109], "quantizatiom": 49, "instead": [49, 52, 53, 54, 55, 66, 70, 74, 75, 88, 97, 105, 110], "potenti": [49, 70, 75, 84], "subgraph": [49, 52, 53, 54, 55, 60, 62, 88, 92, 94, 112], "aten": [49, 54, 55, 56, 60, 61, 64, 68, 69, 70, 75, 76, 88, 98, 112], "thrown": [49, 70, 75, 76], "empti": [49, 70, 71, 75, 76, 82, 87, 92], "torch_tensorrtnamespac": 50, "loggingenum": 50, "levelnamespac": 50, "ptqtemplat": 50, "int8cachecalibratortempl": 50, "int8calibratornamespac": 50, "torchscriptstruct": 50, "compilespecstruct": 50, "deviceclass": 50, "devicetypestruct": 50, "graphinputsstruct": 50, "inputclass": 50, "datatypeclass": 50, "tensorformatenum": 50, "cppdirectori": 50, "includedirectori": 50, "torch_tensorrtfil": 50, "hfile": 50, "relationship": 50, "inherit": [50, 65, 70, 90], "subdirectori": 51, "definit": [51, 54, 60, 81], "cli": [52, 89], "It": [52, 54, 55, 56, 57, 59, 60, 65, 66, 68, 75, 79, 81, 92, 107, 110, 112], "serv": [52, 58, 65, 68, 70, 75], "easi": [52, 53, 55, 88, 90], "wai": [52, 64, 65, 66, 87, 88, 90, 92, 94, 95, 97, 106, 107, 110, 111], "command": [52, 64, 66, 81, 82, 87, 88, 108], "line": [52, 66, 82, 88, 96], "quickli": [52, 88, 90], "part": [52, 56, 59, 65, 74, 79, 80, 81, 92, 94], "deploy": [52, 74, 88, 89, 90, 107, 108, 110, 113], "pipelin": [52, 88, 96, 100, 113], "basic": [52, 56, 65, 82, 106, 108], "featur": [52, 56, 65, 66, 88, 90, 91, 100, 105, 107, 112], "though": [52, 59, 60, 87, 88, 112], "alreadi": [52, 53, 54, 55, 88, 90, 92, 109], "two": [52, 55, 60, 62, 64, 65, 66, 75, 81, 82, 86, 87, 89, 90, 94, 108, 109], "embed": [52, 54, 58, 69, 76, 81, 113], "plan": [52, 59, 63, 64, 70], "after": [52, 53, 55, 56, 62, 65, 70, 74, 75, 87, 88, 89, 98, 101, 108, 110], "link": [52, 53, 62, 68, 79, 80, 85, 88, 92, 110], "against": [52, 88], "libtorchtrt": [52, 66, 88], "python": [52, 56, 59, 62, 64, 65, 70, 71, 74, 75, 76, 81, 82, 88, 91, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 110, 113], "import": [52, 55, 56, 63, 64, 65, 66, 67, 74, 79, 81, 87, 88, 89, 91, 92, 93, 94, 96, 108, 109, 110, 111], "packag": [52, 55, 64, 88], "aspect": 52, "ident": [52, 62, 70, 75, 97], "standard": [52, 58, 66, 68, 70, 74, 75, 76, 81, 91, 92, 96, 107, 110], "load": [52, 56, 58, 64, 65, 67, 70, 73, 74, 75, 76, 88, 89, 90, 91, 92, 93, 94, 96, 97, 107, 108, 110, 112], "like": [52, 53, 55, 58, 60, 65, 66, 67, 75, 80, 81, 87, 88, 89, 90, 92, 94, 96, 97, 108, 110], "would": [52, 54, 60, 64, 65, 66, 74, 88, 89, 91, 92, 108, 110], "input_file_path": [52, 113], "output_file_path": [52, 113], "input_spec": [52, 65, 71], "displai": [52, 62, 64, 72, 79, 110], "menu": [52, 79, 81], "verbios": 52, "v": [52, 82, 105, 108], "verbos": [52, 64, 65, 70, 71, 82, 99, 101], "about": [52, 53, 58, 60, 66, 74, 79, 88, 108, 109], "process": [52, 56, 64, 75, 80, 81, 87, 90, 91, 97, 98, 102, 107, 108, 110], "onto": [52, 58], "consol": 52, "w": [52, 66, 75], "disabl": [52, 64, 66, 70, 74, 79, 80, 94, 110], "i": [52, 55, 60, 66, 67, 69, 81, 82, 87, 88, 90, 92, 93, 94, 105], "debugg": [52, 70, 75, 76], "fallback": [52, 57, 59, 60, 97, 113], "model": [52, 56, 58, 63, 67, 68, 70, 71, 72, 73, 75, 87, 88, 89, 90, 91, 93, 94, 95, 96, 106, 109, 110, 112], "throw": [52, 55, 75, 88], "spars": [52, 54, 64, 69, 70], "p": [52, 69, 88, 108, 113], "repeat": [52, 69], "f32": [52, 70, 74, 75, 92], "half": [52, 64, 75, 81, 88, 89, 90, 91, 92, 98, 99, 108, 113], "float16": [52, 75, 92, 96, 100], "f16": [52, 75, 88, 113], "i8": [52, 75], "d": [52, 75, 81, 82, 88, 113], "multi": [52, 74], "dlacor": 52, "avail": [52, 54, 60, 62, 64, 65, 66, 70, 74, 75, 79, 92, 112, 113], "dla_standalon": [52, 75], "file_path": [52, 75, 111], "teo": 52, "op_nam": 52, "op": [52, 53, 54, 55, 56, 57, 59, 60, 62, 63, 64, 74, 75, 88, 98, 110, 112], "partial": [52, 81], "tem": 52, "module_nam": 52, "mod": [52, 56, 65, 70, 85, 88, 90], "mb": [52, 77], "num_op": 52, "block": [52, 53, 55, 56, 64, 70, 85, 112], "treat": 52, "num": 52, "avg": 52, "num_it": 52, "sram": 52, "local": [52, 55, 66, 79, 88], "dram": 52, "atol": 52, "absolut": [52, 66], "toler": 52, "threshold": 52, "numer": [52, 65, 82], "deviat": 52, "1e": [52, 96, 97], "rtol": 52, "rel": [52, 56], "skip": 52, "complianc": 52, "64bit": 52, "32bit": 52, "custom": [52, 62, 63, 65, 66, 95, 103, 104, 106], "dll": 52, "n": [52, 60, 62, 75, 88, 90, 92, 93], "min_n": 52, "min_c": 52, "min_h": 52, "min_w": 52, "opt_n": 52, "opt_c": 52, "opt_h": 52, "opt_w": 52, "max_n": 52, "max_c": 52, "max_h": 52, "max_w": 52, "32": [52, 75, 87, 88, 89, 90, 103, 104, 105, 113], "flag": [52, 56, 57, 59, 66, 73, 75, 89, 102, 110, 111], "forc": [52, 63, 65, 70, 75, 76, 79], "posit": [52, 54, 65, 75, 79], "test": [52, 56, 59, 65, 66, 70, 75, 81, 82, 90, 105, 107, 108], "ssd_trace": 52, "pt": [52, 65, 88, 103, 104, 108], "ssd_trt": 52, "300": [52, 91], "512": [52, 70, 75, 76, 105, 107], "1024": [52, 70, 75, 76, 103, 107], "simplifi": [53, 92], "form": [53, 74, 75, 81, 89, 108], "up": [53, 55, 56, 57, 58, 59, 62, 65, 66, 70, 75, 81, 87, 92, 94, 95, 97, 98, 101, 106, 107], "context": [53, 57, 58, 59, 64, 72, 74, 102, 110], "inetworkdefinit": [53, 54], "record": [53, 87, 93, 94, 102, 110], "togeth": [53, 60, 88], "start": [53, 56, 65, 69, 73, 75, 82, 88, 91, 92, 93, 94, 107], "look": [53, 54, 55, 67, 70, 75, 87, 90, 91, 94, 108, 109], "assembl": [53, 62, 88], "resourc": [53, 90, 92], "coupl": [53, 59, 65, 110], "state": [53, 54, 60, 62, 74, 88, 96], "been": [53, 60, 64, 66, 73, 82, 88, 94, 97, 112], "evaluated_value_map": [53, 60], "stage": [53, 65], "arg": [53, 54, 62, 65, 70, 73, 74, 75, 85, 88, 92, 94, 105, 107], "itensor": [53, 54, 60, 65, 88, 92], "value_tensor_map": [53, 60], "typic": [53, 60, 75, 108], "abl": [53, 55, 60, 62, 65, 90, 91, 92, 97], "system": [53, 60, 62, 64, 68, 70, 74, 75, 76, 92, 94, 97, 112], "registri": [53, 54, 88, 92], "enter": [53, 75], "recurs": 53, "resolv": [53, 55, 57, 59, 98, 101], "until": [53, 56, 59, 60, 66, 70, 75, 112], "final": [53, 56, 57, 59, 66, 92, 98, 101, 107], "some": [53, 54, 55, 56, 57, 58, 59, 60, 62, 63, 65, 66, 75, 80, 81, 88, 90, 92, 94, 109, 112], "These": [53, 54, 56, 58, 62, 64, 66, 70, 73, 74, 75, 79, 81, 90, 108, 112], "those": [53, 54, 62, 64, 81], "do": [53, 54, 55, 56, 60, 63, 65, 80, 82, 87, 88, 89, 90, 92, 113], "theori": [53, 81], "kind": [53, 65], "common": [53, 55, 65, 71, 81, 94], "prim": [53, 55, 56, 58, 69, 87, 88], "constant": [53, 54, 55, 56, 88, 92], "emit": 53, "listconstruct": [53, 56, 58, 88], "make": [53, 54, 65, 66, 70, 75, 81, 83, 88, 89, 90, 92, 94, 95, 106, 107, 108, 113], "associ": [53, 60, 88, 94, 110], "where": [53, 54, 55, 60, 62, 64, 65, 70, 74, 75, 76, 82, 88, 90, 97], "result": [53, 55, 56, 66, 67, 70, 72, 74, 75, 76, 79, 87, 89, 92, 96, 97, 108, 112], "done": [53, 56, 59, 92, 97, 108, 111], "mai": [53, 54, 56, 58, 59, 65, 66, 70, 74, 75, 76, 81, 82, 87, 88, 89, 90, 92, 97, 98, 101, 108, 110], "For": [53, 56, 62, 63, 64, 65, 66, 67, 71, 75, 79, 81, 82, 87, 88, 90, 91, 92, 96, 98, 105, 107, 108, 110, 111], "more": [53, 64, 65, 66, 68, 70, 75, 79, 82, 87, 88, 89, 90, 91, 92, 94, 96, 99, 101, 108, 110], "writing_convert": [53, 88], "locat": [54, 62, 66, 90, 92], "py": [54, 55, 59, 62, 65, 66, 77, 79, 81, 86, 87, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 109], "convers": [54, 55, 56, 58, 63, 64, 65, 70, 75, 76, 88, 92, 109], "decror": 54, "dynamo_tensorrt_convert": [54, 92], "signatur": [54, 76], "leaky_relu": [54, 69], "def": [54, 62, 65, 81, 87, 89, 92, 93, 94, 98, 105, 108, 109], "leaky_relu_convert": 54, "ctx": [54, 60, 88, 92], "conversionctx": [54, 60, 88], "tupl": [54, 58, 63, 65, 70, 71, 74, 75, 76, 89, 92, 94, 97, 109], "kwarg": [54, 65, 70, 73, 74, 75, 92, 107], "dict": [54, 70, 74, 75, 76, 92, 94], "union": [54, 60, 64, 70, 74, 75, 76, 88], "sequenc": [54, 62, 65, 70, 71, 74, 75, 76, 81, 92, 107], "decor": [54, 62, 65], "kei": [54, 81, 87, 94, 108, 109], "node": [54, 55, 56, 57, 59, 60, 62, 65, 71, 88, 92, 105, 107, 109], "capability_valid": 54, "lambda": [54, 60, 81, 88, 108], "fx": [54, 62, 63, 70, 74, 75, 88, 89, 92, 97, 111], "determin": [54, 55, 64, 65, 75, 109, 110], "properli": [54, 66], "handl": [54, 55, 56, 58, 64, 65, 74, 75, 92], "partition": [54, 70, 75, 92], "sure": [54, 66, 88, 89, 108, 113], "prioriti": 54, "develop": [54, 65, 66, 68, 81, 82, 88, 92], "bodi": [54, 81, 82], "nativ": [54, 59, 61, 88, 92, 97], "numpi": [54, 75, 92, 93, 94, 96, 97, 108], "frozen": 54, "attribut": [54, 55, 56, 58, 65, 75, 81, 88], "previou": [54, 79, 98], "correspond": [54, 60, 65, 66, 74, 75, 94, 96, 105, 110], "edg": [54, 81], "well": [54, 63, 66, 68, 72, 74, 81, 88, 90, 94, 102, 111], "being": [54, 65, 66, 70, 88, 92, 97], "truth": 54, "http": [54, 61, 64, 66, 79, 81, 87, 88, 90, 92, 96, 98, 101, 105, 107, 108, 109, 110], "github": [54, 61, 64, 66, 79, 88, 90, 98, 101, 105, 108, 110], "com": [54, 61, 64, 66, 88, 90, 96, 98, 101, 105, 108, 110], "blob": [54, 61, 66, 79, 90, 94], "main": [54, 55, 56, 57, 58, 59, 60, 63, 65, 66, 79, 81, 83, 88, 92, 105], "src": [54, 58, 61, 69], "native_funct": [54, 61], "yaml": [54, 61], "sinc": [54, 55, 64, 65, 74, 81, 87, 88, 90, 93, 94, 97], "mani": [54, 56, 64, 65, 79, 81, 82, 94, 97, 112], "composit": [54, 88], "raw": [54, 79], "impl": 54, "subpackag": 54, "chain": [54, 60], "primarili": [54, 59, 66, 88], "manipul": [54, 62, 75], "net": [54, 60, 81, 82, 88, 92], "addit": [54, 55, 64, 65, 74, 75, 88, 92, 94, 97, 107, 109], "call_modul": 54, "call_funct": [54, 62, 65], "eg": [54, 108], "aten_": 54, "_leaky_relu": 54, "opoverloadpacket": 54, "while": [54, 56, 66, 74, 90, 96, 107, 108, 110, 112], "opoverload": 54, "particular": [54, 64, 94], "collect": [54, 56, 64, 70, 75, 76, 88, 89, 105], "trtinterpret": [54, 65, 71], "along": [54, 75], "match": [54, 55, 97], "special": [54, 56], "account": [54, 108], "illustr": [54, 65, 99, 107], "scale_grad_by_freq": [54, 69], "embedding_param_valid": 54, "establish": 54, "subset": [54, 64, 70, 75, 90, 107], "converter_util": [54, 92], "enforce_tensor_typ": 54, "dictionari": [54, 75, 76, 91, 98], "between": [54, 55, 56, 60, 66, 75, 81, 82, 90, 94, 96], "possibl": [54, 66, 81, 92, 94, 107, 108], "prefer": [54, 64, 66, 88], "keyword": [54, 62, 70, 74, 75, 76, 98, 101], "both": [54, 56, 64, 66, 68, 70, 71, 74, 75, 79, 81, 87, 90, 92, 94], "enforc": [54, 88], "situat": 54, "partit": [54, 55, 63, 64, 70, 75, 112], "greater": [54, 70, 72, 75], "than": [54, 55, 64, 66, 70, 75, 80, 81, 93, 94, 96, 107, 110], "3d": [54, 65], "autocast": 54, "therebi": [54, 58, 92, 107], "limit": [54, 55, 72, 80, 90, 94, 112], "author": [54, 82], "conv_nod": 54, "7": [54, 56, 58, 59, 74, 75, 85, 88, 92, 98, 99, 101, 105, 109], "ignor": [54, 70, 74, 75, 92], "misc": [54, 92], "trttensor": 54, "np": [54, 92, 93, 94, 96, 97, 108], "ndarrai": [54, 92], "aten_ops_convolut": 54, "conversioncontext": [54, 92], "side": [54, 55, 79, 88], "effect": [54, 55, 64, 65, 70, 79, 88, 90, 92, 107], "term": [54, 75, 81, 82, 90, 92, 107], "getitem": 54, "categor": 54, "modif": [54, 62, 75], "op_evalu": 54, "capbility_valid": 54, "opcod": 54, "decompos": 54, "suboper": 54, "separ": [54, 56, 57, 59, 66], "Such": 54, "via": [54, 64, 65, 68, 70, 74, 75, 76, 79, 85, 89, 90, 98, 99, 101, 107, 109, 110, 111, 112], "register_torch_trt_decomposit": 54, "addmm_replac": 54, "replac": [54, 56, 62, 66, 73, 92, 105, 112], "input_": 54, "mat1": 54, "mat2": [54, 69], "beta": [54, 65, 69, 76], "alpha": [54, 65, 69, 82], "mul": [54, 56, 69], "matmul": [54, 55, 69, 88, 109], "modifi": [54, 56, 62, 65, 82, 95, 96, 106, 109], "edit": [54, 66, 79], "torch_enabled_decomposit": 54, "torch_disabled_decomposit": 54, "disjoint": 54, "preced": [54, 81], "over": [54, 57, 59, 65, 81, 105, 108, 112], "much": [54, 60, 79, 81, 90], "significantli": [54, 55, 79, 94], "easier": [54, 57, 59, 60, 65, 70, 74, 75, 88, 90, 92, 96], "tri": 54, "made": [55, 57, 59, 75, 81], "represent": [55, 60, 65, 87, 107, 112], "instanc": [55, 62, 64, 66, 70, 73, 74, 87, 88, 107, 110], "idea": [55, 81], "reduc": [55, 56, 57, 59, 65, 70, 75, 90, 92, 94, 107, 110], "actual": [55, 58, 60, 65, 87, 88, 92], "aim": [55, 112], "closer": 55, "scope": [55, 92, 98, 101], "csrc": [55, 61], "common_subexpression_elimin": 55, "subexpress": 55, "dead_code_elimin": 55, "exception_elimin": 55, "wa": [55, 58, 62, 64, 65, 70, 74, 75, 81, 88, 112], "1013": 55, "ne": [55, 69], "1012": 55, "24": 55, "lib": [55, 66, 88], "python3": [55, 66, 88], "6": [55, 56, 58, 66, 69, 85, 87, 88, 92], "site": [55, 66, 81, 88], "nn": [55, 61, 65, 70, 71, 74, 75, 76, 87, 88, 89, 92, 98, 105, 109, 112], "batchnorm": 55, "248": 55, "11": [55, 66, 81, 85, 88, 108], "block0": 55, "raiseexcept": 55, "249": 55, "12": [55, 56, 81, 85, 87, 88, 99, 108, 109], "block1": 55, "guard_elimin": 55, "whose": [55, 65, 99], "freeze_modul": 55, "propag": 55, "fuse_addmm_branch": 55, "variant": [55, 110], "caught": 55, "ret": 55, "622": 55, "self": [55, 58, 60, 69, 74, 75, 87, 88, 89, 92, 94, 98, 105, 107, 109, 113], "bia": [55, 69, 88, 105], "x9": 55, "3677": 55, "output0": 55, "add_": [55, 69, 88], "fuse_linear": 55, "back": [55, 56, 58, 59, 74, 75, 81, 87, 88, 92, 112], "fuse_flatten_linear": 55, "implicitli": [55, 75], "connect": [55, 70, 75, 76, 81, 96, 108, 113], "higher": [55, 64, 70, 75, 79, 81, 87], "1d": 55, "lower_graph": 55, "access": [55, 60, 65, 79, 88, 91, 112], "rather": 55, "getattr": [55, 58, 87, 88], "trainabl": 55, "remain": [55, 75, 90, 112], "lower_tupl": 55, "lowersimpletupl": 55, "tupleconstruct": [55, 58], "tupleunpack": 55, "leav": [55, 62, 64, 70], "statement": [55, 81], "loweralltupl": 55, "_all_": 55, "rais": [55, 65, 75], "onnx": 55, "module_fallback": 55, "consist": [55, 65, 81, 92, 110, 112], "pair": [55, 60, 66, 81, 90, 107], "delimit": 55, "around": [55, 58, 60, 66, 74, 81, 84, 87, 92], "second": [55, 65, 81, 89, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "mark": [55, 56, 79, 94], "notatemoduleforfallback": 55, "marknodesforfallback": 55, "tell": [55, 56, 57, 58, 59, 60, 81, 112], "them": [55, 56, 58, 63, 64, 65, 66, 70, 74, 79, 88, 92, 94, 107, 109, 112], "peephole_optimz": 55, "intent": [55, 81], "catch": [55, 75, 88], "small": [55, 92, 93, 108], "might": [55, 66, 79, 97, 109], "interest": [55, 81], "now": [55, 56, 59, 60, 65, 66, 75, 81, 88, 91, 92, 94, 97, 110], "expand": [55, 69], "simpli": [55, 98, 107], "remove_contigu": 55, "remove_dropout": 55, "infer": [55, 64, 65, 70, 75, 76, 88, 90, 95, 97, 98, 106, 107, 109, 110, 111, 112], "remove_to": 55, "unpack_addmm": 55, "reus": [55, 65, 90, 94], "dedic": [55, 82], "unpack_log_softmax": 55, "softmax": [55, 65, 69, 105], "loop_unrol": 55, "suffici": [55, 66, 75], "short": [55, 64, 70, 81, 82, 97], "tile_to_repeat": 55, "instruct": [56, 57, 59, 65, 66, 88, 108], "criteria": [56, 57, 59, 64], "lack": [56, 57, 59, 65, 92], "explicitli": [56, 57, 59, 66, 76, 89, 90, 91], "On": 56, "segment": [56, 63, 92, 99, 101, 107], "verifi": [56, 70, 92, 97], "Then": [56, 90, 91, 97], "roughli": 56, "analysi": 56, "everi": [56, 71, 74, 75, 88, 110], "complet": [56, 63, 70, 75, 87, 88], "mean": [56, 60, 65, 69, 71, 98, 108, 112], "trace": [56, 65, 70, 74, 76, 87, 88, 109, 111, 112], "tensorlist": [56, 60], "figur": [56, 82, 84], "our": [56, 59, 63, 87, 88, 108], "stitch": [56, 88], "altogeth": [56, 79], "brief": 56, "descript": [56, 82, 105], "partitioninfo": 56, "api": [56, 59, 60, 62, 63, 64, 65, 74, 75, 76, 80, 88, 89, 90, 91, 92, 95, 98, 99, 102, 106, 107, 108, 109, 110, 111], "maintain": [56, 58, 60, 75, 96, 112], "code": [56, 59, 62, 64, 65, 66, 80, 82, 87, 88, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 109], "mymodel": [56, 63, 67, 89, 92, 109, 111], "ts_model": [56, 88], "trt_model": [56, 91, 92, 99, 103, 104, 105, 108, 111], "off": [56, 58, 102], "consecut": [56, 63], "satisfi": [56, 62, 65], "forced_fallback_op": 56, "randn": [56, 63, 67, 70, 75, 76, 88, 91, 94, 99, 102, 109, 111], "224": [56, 63, 67, 70, 71, 75, 76, 88, 94, 96, 97, 99, 102, 107, 108, 109, 111], "trt_ts_modul": [56, 89], "input_s": 56, "inputrang": 56, "cfg": [56, 88], "relu": [56, 69, 87, 88, 98, 105], "trt_mod": [56, 67, 88, 90, 113], "consid": [56, 76, 88, 92], "segmentmodelwithdependencyawar": 56, "test_segment": 56, "20": [56, 85, 97, 99, 101], "x_lgamma": 56, "lgamma": 56, "y_lgamma": 56, "div": [56, 69], "div_lgamma": 56, "27": [56, 88], "cat": [56, 66, 69, 105], "greedi": [56, 103, 104], "strategi": [56, 75], "travers": [56, 57, 59, 64], "gather": 56, "same": [56, 58, 62, 64, 65, 66, 70, 75, 79, 81, 87, 88, 91, 92, 94, 97, 99, 101, 108, 109, 110, 111], "encount": [56, 64, 66, 98, 101], "4": [56, 58, 63, 64, 65, 66, 69, 75, 77, 79, 81, 82, 85, 88, 92, 98, 100, 101, 102, 105, 109], "suboptim": 56, "arithmet": 56, "split": [56, 65, 69], "own": [56, 60, 64, 66, 70, 81, 88, 94, 105, 108], "could": [56, 64, 65, 92, 99, 101, 110], "rewrit": [56, 62], "portion": [56, 81, 92, 100], "without": [56, 60, 67, 70, 79, 81, 88, 90, 92, 93, 94, 97, 110], "reorder": 56, "seri": 56, "cleanli": 56, "approach": [56, 94], "achiev": [56, 107], "hit": 56, "larger": [56, 70, 75, 79, 107], "boundari": [56, 73, 75], "guarante": [56, 74], "trigger": [56, 64, 65, 75, 88, 94, 96, 97, 112], "appear": [56, 81], "adjac": [56, 70, 75, 81], "As": [56, 65, 66, 75, 88, 92, 94, 97, 112], "clean": [56, 62, 81, 98, 101], "step": [56, 65, 69, 75, 90, 92, 97, 107], "consolid": [56, 87], "further": [56, 64, 65, 110, 112], "merg": 56, "identifi": 56, "do_not_merg": 56, "combin": [56, 64, 65], "condit": [56, 81, 112], "loop": [56, 64, 65, 103, 104], "ir": [57, 59, 60, 63, 64, 67, 70, 75, 87, 88, 89, 95, 98, 99, 101, 102, 106, 109], "larg": [57, 59, 79, 81, 88, 90, 97, 107], "opset": [57, 59], "compon": [57, 59, 66, 73, 87, 110, 112], "evalu": [57, 58, 59, 105], "deploi": [57, 59, 68, 88, 90, 95, 106, 108], "instanti": [57, 58, 59, 60, 88, 100], "wrap": [57, 58, 59, 65, 81, 84, 88, 91, 98, 101], "extend": [57, 59, 60, 69, 88, 94, 107], "providi": [57, 59], "stand": [58, 81], "interpret": [58, 65, 81], "execute_engin": [58, 74, 88], "stack": [58, 69, 90, 105, 112], "machin": [58, 66, 90, 108], "pop": 58, "push": 58, "element": [58, 65, 81, 82, 85], "realiz": 58, "abstract": [58, 60, 82], "__torch__": [58, 87, 88], "portabl": [58, 66, 76], "serializ": [58, 64, 87, 112], "instnanti": 58, "whatev": [58, 65, 92], "self_1": [58, 88], "torchvis": [58, 90, 91, 94, 96, 97, 99, 102, 105, 108], "resnet": [58, 77, 95, 96, 106, 107, 108], "___torch_mangle_4847": 58, "resnet_trt": 58, "input_0": [58, 88], "__torch___torchvision_models_resnet____torch_mangle_4847_resnet_trt_engin": 58, "listunpack": [58, 88], "multipl": [58, 66, 70, 74, 75, 81, 82, 90, 108, 110], "repack": 58, "ssd": 58, "ssd300_trt": 58, "__torch___pytorch_detection_ssd_src_model_ssd300_trt_engin": 58, "holder": [58, 83], "torchbind": 58, "pickler": 58, "seril": 58, "zip": [58, 66, 96, 97, 106], "depickl": 58, "encod": [58, 107], "sm": 58, "correct": [58, 66, 79, 96, 97, 105], "bazel": [59, 66], "linux": [59, 88], "x86_64": [59, 66], "aarch64": 59, "gcc": [59, 88], "untest": 59, "try": [59, 75, 81, 82, 88, 91, 92, 94, 112], "older": 59, "repositori": [59, 66, 79, 86, 108], "notebook": [59, 68, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "doc": [59, 61, 66, 79, 80, 81, 86, 92, 109], "docsrc": 59, "third_parti": [59, 66], "toolchain": [59, 66], "unstabl": 59, "subject": [59, 62, 112], "matur": 59, "most": [59, 65, 66, 71, 92, 97, 108, 110, 112], "hood": [59, 99, 112], "major": [59, 65, 75], "top": [59, 79, 83], "coordin": [59, 75], "ingest": 59, "flow": [60, 65, 81, 87, 107], "ilay": 60, "analogu": 60, "goal": [60, 64, 94], "registernodeconversionpattern": [60, 88], "helper": 60, "pattern": [60, 75, 88], "schema": [60, 88, 92], "caus": [60, 64, 79, 98, 99, 101, 110], "acthardtanh": 60, "torchtrt_unus": 60, "hardtanh": [60, 69], "scalar": [60, 69], "min_val": [60, 69], "max_val": [60, 69], "unwraptodoubl": 60, "new_lay": 60, "addactiv": 60, "activationtyp": [60, 65], "kclip": 60, "torchtrt_check": 60, "unabl": [60, 88, 92], "setalpha": 60, "setbeta": 60, "setnam": [60, 88], "util": [60, 62, 73, 76, 88, 90, 95, 98, 101, 103, 104, 105, 106, 107, 108, 112], "node_info": [60, 88], "c_str": [60, 88], "out_tensor": [60, 88], "associatevalueandtensor": [60, 88], "getoutput": [60, 88], "log_debug": 60, "getdimens": [60, 88], "accord": [60, 64, 76], "unwrap": 60, "tool": [60, 64, 65, 66, 88, 94, 107], "don": [60, 65, 79, 81, 82, 90, 105, 108, 109], "annot": [60, 88], "your": [60, 63, 64, 66, 67, 74, 79, 81, 82, 86, 87, 88, 89, 91, 94, 109, 110], "Its": [60, 81], "track": [60, 90], "sort": [60, 69, 91], "live": [60, 81], "directli": [60, 62, 63, 66, 68, 73, 75, 90, 92, 95, 98, 106, 111], "associatevalueandivalu": 60, "inspect": [60, 87, 88], "dataflow": [60, 88], "mechan": [60, 64, 65, 92, 97, 107], "safe": [60, 64, 70, 74, 75, 76], "unsur": 60, "deep": [60, 64, 68, 79, 90, 92, 113], "straight": 60, "chanc": 60, "none": [60, 64, 65, 69, 70, 71, 73, 74, 75, 76, 79, 81, 92, 94, 98, 105], "wrapper": [60, 65, 111], "similar": [60, 63, 64, 65, 66, 88, 91, 92, 103, 104], "tocustomclass": 60, "tensorcontain": 60, "istensor": 60, "iscustomclass": 60, "lot": [60, 63], "singular": 60, "becaus": [60, 65, 66, 71, 87, 88, 92, 93, 94], "alloc": 60, "freed": 60, "destructor": 60, "destroi": [60, 82], "realli": 60, "think": [60, 81], "becom": [60, 66, 96], "benefit": [60, 88, 94], "deal": [60, 94], "quit": [60, 66, 88, 107], "effici": 60, "batch_norm": [60, 69], "fusion": [60, 62, 65], "deeplearn": [61, 65], "sdk": [61, 112], "matrix": 61, "html": [61, 66, 81, 87, 90, 92, 109], "c_api": 61, "python_api": 61, "org": [61, 66, 79, 81, 87, 88, 90, 92, 109, 110], "stabl": [61, 76, 77, 79, 95, 106, 109], "master": [61, 66, 90, 110], "overview": [61, 68, 98, 102], "md": 61, "appli": [62, 63, 90, 97], "desir": [62, 70, 82, 90, 94], "coalesc": 62, "insert": [62, 88, 90, 94, 97], "graphmodul": [62, 63, 70, 71, 75, 88, 89, 92, 97, 111, 112], "caller": 62, "invok": [62, 64, 65, 87, 88, 110], "lint": 62, "recompil": [62, 70, 75, 94, 97, 101, 109, 112], "repair": 62, "disallow": 62, "repair_input_as_output": 62, "gm": [62, 70], "sample_input": [62, 65, 98], "scenario": [62, 64, 96], "clone": [62, 66, 69, 92], "modified_graph": 62, "extract": [62, 88, 107], "placehold": 62, "isinst": [62, 65, 92, 105], "issubclass": 62, "direct": [62, 85, 97, 110], "len": [62, 69, 92], "direct_output": 62, "inserting_aft": 62, "cloned_placehold": 62, "replace_input_with": 62, "date": [62, 82, 112], "eliminate_dead_cod": 62, "logger": [62, 72], "f": [62, 64, 65, 75, 81, 87, 92, 105], "__init__": [62, 74, 75, 81, 87, 92, 94, 98, 105, 109], "pass_manag": 62, "passmanag": 62, "backend": [62, 67, 76, 77, 80, 91, 93, 94, 95, 98, 100, 101, 105, 106, 109], "offer": [62, 64], "registr": [62, 65], "conveni": [62, 90, 101, 107, 110, 112], "control": [62, 65, 87, 97, 110], "_aten_lowering_pass": 62, "my_custom_pass": 62, "front": [62, 70], "passlist": 62, "arbitrari": [62, 74], "remov": [62, 63, 70, 79, 93, 94, 105], "dump_lowering_pass": 62, "apply_lowering_pass": 62, "graph_modul": [62, 70], "_remove_lowering_pass": 62, "evolv": 62, "introduc": [63, 65, 107], "exportedprogram": [63, 67, 70, 75, 97, 103, 104, 109, 112], "dynamo": [63, 64, 66, 67, 73, 74, 75, 77, 88, 92, 93, 94, 97, 98, 99, 101, 102, 105, 109], "frontend": [63, 70, 73, 89, 92, 95, 99, 101, 105, 106, 109], "simpl": [63, 64, 65, 81, 82, 87, 107, 108, 109], "usag": [63, 65, 73, 77, 81, 88, 95, 106, 109, 111], "eval": [63, 67, 88, 89, 93, 94, 96, 97, 98, 99, 101, 102, 103, 104, 105, 108, 109, 111], "exp_program": [63, 94, 97, 105, 109], "trt_gm": [63, 67, 94, 97, 109, 111], "interact": [63, 81, 96, 98, 99, 100, 101, 102, 103, 104], "ideal": 63, "discuss": [63, 64, 108], "section": [63, 65, 79, 81, 82, 83, 85, 88, 90, 108, 111], "frequent": 63, "builder": [63, 64, 65, 70], "respect": [63, 66, 75], "releas": [63, 64, 81, 95, 106], "insid": [63, 81, 92, 95, 106, 108], "decomposit": [63, 64, 70, 75, 92], "downstream": [63, 107], "constraint": 63, "guid": [64, 80, 106], "present": [64, 97], "learn": [64, 66, 68, 88, 90, 92, 108, 113], "acceler": [64, 71, 75, 95, 106, 110, 112, 113], "workflow": [64, 65, 67, 68, 70, 71, 75, 88, 91, 94, 96, 99, 100, 101, 103, 104, 107], "wide": [64, 75, 85], "varieti": [64, 108], "primari": [64, 94, 111], "simplic": 64, "optimized_model": [64, 67, 93, 98, 99, 101], "depth": [64, 79, 107], "challeng": [64, 96, 108], "addition": [64, 92], "fit": [64, 81], "compilationset": [64, 70, 74, 92, 98], "_enum": [64, 70], "callabl": [64, 70, 75], "pass_through_build_failur": [64, 70, 74, 75, 92], "max_aux_stream": [64, 70, 74, 75, 92], "version_compat": [64, 70, 74, 75, 92], "optimization_level": [64, 70, 74, 75, 92, 98], "use_python_runtim": [64, 70, 74, 75, 92, 93, 94, 96, 97, 98], "truncate_doubl": [64, 70, 74, 75, 92, 93, 103, 104], "use_fast_partition": [64, 70, 74, 75, 92], "enable_experimental_decomposit": [64, 70, 74, 75, 92], "_devic": [64, 70], "assume_dynamic_shape_support": [64, 70, 74, 75], "make_refit": [64, 70, 74, 75, 93, 94, 96, 97], "engine_cap": [64, 70, 74, 75, 92], "dryrun": [64, 70, 74, 75, 92], "hardware_compat": [64, 70, 74, 75, 92], "timing_cache_path": [64, 70, 74, 75, 94], "tmp": [64, 70, 74, 75, 88, 93], "torch_tensorrt_engine_cach": [64, 70, 74, 75], "timing_cach": [64, 65, 70, 74, 75], "bin": [64, 66, 70, 74, 75], "lazy_engine_init": [64, 70, 74, 75], "cache_built_engin": [64, 70, 74, 93, 94], "reuse_cached_engin": [64, 70, 74, 93, 94], "dpython": [64, 70, 75, 76], "per": [64, 70, 92, 110], "regardless": [64, 70, 82, 99, 101], "fail": [64, 70, 75, 88, 96, 97, 105, 113], "auxiliari": [64, 70], "stream": [64, 70, 75, 92], "impli": [64, 70], "longer": [64, 66, 70, 75, 79, 110], "search": [64, 68, 70, 75, 79], "strictli": [64, 70], "runtim": [64, 66, 67, 68, 70, 75, 88, 96, 98, 101, 102, 112], "presenc": [64, 70], "preferenti": [64, 70], "choos": [64, 65, 70, 87], "float64": [64, 70, 75, 76], "refitt": [64, 70], "toggl": [64, 70, 75], "mode": [64, 65, 70, 74, 75, 89, 90, 102, 105], "detail": [64, 65, 70, 87, 88, 92, 94, 108, 110], "natur": [64, 70, 81], "architectur": [64, 66, 68, 70, 75, 94, 107], "amper": [64, 70, 75], "newer": [64, 66, 70, 75], "storag": [64, 70, 90], "sub": [64, 69, 81, 87, 98], "slate": 64, "futur": [64, 65, 70, 75, 76, 110], "occur": 64, "first_output": 64, "subsequ": [64, 94], "second_output": 64, "session": [64, 67, 81, 94, 102], "point": [64, 66, 75, 79, 80, 81, 88, 105, 108], "cover": [64, 106, 107], "benchmark": [64, 69], "automat": [64, 75, 81, 88, 97, 109, 112], "vari": [64, 71, 109], "distribut": [64, 88, 90, 110], "inf": 64, "dynamo_convers": 64, "contribut": 64, "demonstr": [64, 81, 82, 83, 90, 92, 94, 95, 96, 105, 106, 107, 108], "break": [64, 65, 70, 74, 75, 81, 92], "successfulli": [64, 96, 97], "_dynamo": [64, 93, 94, 98, 99, 101, 109], "explain": [64, 65, 68], "veri": [64, 65, 82, 83, 90, 91, 103, 104, 108], "explan": [64, 65], "graph_break_count": 64, "furthermor": 64, "durat": [64, 81], "latter": [64, 74], "logic": [64, 65], "guard": 64, "compos": [65, 87, 90, 105, 108], "variou": [65, 113], "etc": [65, 79, 81, 92, 113], "environ": [65, 67, 108], "research": 65, "few": [65, 66, 75], "nightli": 65, "lower_exampl": 65, "welcom": [65, 88], "finish": 65, "converison": 65, "pleas": [65, 75, 81, 88, 105, 108, 109], "max_batch_s": [65, 71, 108], "2048": [65, 71], "max_workspace_s": [65, 71], "33554432": [65, 71], "explicit_batch_dimens": [65, 71], "lower_precis": [65, 71], "lowerprecis": [65, 71], "verbose_log": [65, 71], "timing_cache_prefix": [65, 71], "save_timing_cach": [65, 71], "cuda_graph_batch_s": [65, 71], "dynamic_batch": [65, 71], "turn": [65, 71, 102], "trtmodul": [65, 71], "otherwis": [65, 66, 71, 94, 110], "implicit": [65, 69, 71, 81], "config": [65, 66, 71, 108], "updat": [65, 66, 70, 71, 75, 92, 95, 97, 106], "dim": [65, 69, 71, 92, 94, 105, 108, 109], "fx2trt_exampl": 65, "acc_trac": 65, "come": [65, 66, 80, 92, 96, 108], "my_pytorch_model": 65, "build_model": 65, "prepar": [65, 108], "acc_mod": 65, "earli": [65, 97], "deprec": [65, 69], "continu": [65, 81, 110], "backward": [65, 74, 92, 112], "vision": [65, 108], "activ": [65, 74, 76, 81, 88, 90, 107, 110, 113], "except": [65, 70, 75], "permut": [65, 69], "transpos": [65, 69, 109], "ll": [65, 94], "inputtensorspec": [65, 71, 75], "experiment": [65, 75, 76], "dataclass": [65, 98], "re": [65, 75, 81, 94, 96, 102, 110], "manual": [65, 75, 80, 81, 97], "sampl": [65, 70, 81, 89, 90, 96, 97, 98, 99, 100, 101, 102, 103, 104, 108], "rand": [65, 88, 94, 96, 97, 98], "from_tensor": [65, 75], "slightli": [65, 92], "promis": 65, "optimize_target_shap": 65, "input_tensor_spec": 65, "shape_rang": [65, 71], "100": [65, 71, 92, 94, 105], "accordingli": [65, 79, 109, 110], "trtinterpreterresult": [65, 71], "namedtupl": 65, "input_nam": [65, 71], "output_nam": [65, 71], "serialized_cach": [65, 71], "bytearrai": [65, 74, 76], "afford": 65, "temporari": [65, 94], "best": [65, 70, 75, 81, 96], "perforamnc": 65, "examin": 65, "suitabl": 65, "force_fp32_output": 65, "strict_type_constraint": 65, "usual": [65, 66, 79], "unless": 65, "certain": [65, 66, 98, 110], "algorithm_selector": 65, "profiling_verbos": 65, "trt_interpreter_result": 65, "64": [65, 75, 89, 104, 105, 109], "25": [65, 71, 88], "runtimeerror": [65, 105], "xxx": 65, "One": [65, 81, 82, 88, 107, 110], "reload_trt_mod": 65, "reload_model_output": 65, "far": [65, 81], "give": [65, 79, 81], "convtert": 65, "scheme": [65, 70, 75], "action": [65, 81], "tensort": [65, 112], "thing": [65, 66, 81], "compar": [65, 70, 75, 89, 97], "vanilla": 65, "mainli": 65, "builtin": 65, "purpos": [65, 107, 108], "acc_op": 65, "leverag": [65, 90, 95, 106], "power": [65, 81, 88, 107], "goe": [65, 81], "whole": 65, "sigmoid": [65, 69], "tensorrt_convert": 65, "acc_ops_sigmoid": 65, "rest": [65, 81, 82], "input_v": 65, "receiv": 65, "region": 65, "add_activ": 65, "get_output": [65, 92], "wherev": 65, "rememb": [65, 66], "mapper": 65, "todo": [65, 79], "logist": 65, "down": [65, 66, 79], "happen": [65, 87, 96, 99, 109], "acc_norm": 65, "foo": [65, 81, 82], "register_acc_op": 65, "register_acc_op_map": 65, "this_arg_is_opt": 65, "op_and_target": 65, "arg_replacement_tupl": 65, "rule": [65, 66, 76], "third": [65, 82], "boolean": [65, 75], "matter": [65, 92], "register_custom_acc_mapper_fn": 65, "design": [65, 73, 96, 107, 113], "redund": 65, "throught": 65, "custom_mapp": 65, "_": [65, 81, 92, 105], "foo_kwarg": 65, "inserting_befor": 65, "foo_nod": 65, "meta": [65, 85, 104], "children": 65, "unit": [65, 75], "test_acc_trac": 65, "acc_op_convert": 65, "essenti": 65, "plugin": [65, 92, 95, 106], "yet": [65, 107], "folder": 65, "center": 66, "pypi": 66, "m": [66, 82, 105], "pip": [66, 108], "upload": [66, 108], "x86": [66, 110], "extra": [66, 74, 88, 92, 96], "url": [66, 79, 108], "download": [66, 85, 90, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 108], "whl": 66, "cu118": 66, "cu124": 66, "tarbal": [66, 88, 90], "easiest": [66, 92], "bazelisk": 66, "bazelbuild": 66, "export": [66, 68, 70, 75, 94, 95, 97, 99, 103, 104, 105, 106, 111, 112], "bazel_vers": 66, "path_to_torchtrt_root": 66, "bazelvers": 66, "mkdir": 66, "cd": [66, 108], "curl": [66, 81], "fssl": 66, "o": [66, 81, 108], "dist": 66, "unzip": 66, "bash": 66, "sh": 66, "cp": [66, 92], "usr": 66, "driver": 66, "branch": 66, "4e5b0f6e860910eb510fa70a76ee3eb9825e7a4d": 66, "l46": 66, "pull": [66, 94, 108], "latest": [66, 79], "l53c1": 66, "fact": 66, "reproduc": 66, "l71": 66, "http_archiv": 66, "build_fil": 66, "archiv": 66, "sha256": 66, "strip_prefix": 66, "OR": 66, "TO": [66, 88], "gnu": 66, "tar": [66, 81, 90], "gz": [66, 81, 82, 90], "ld_library_path": 66, "comment": [66, 81], "uncom": 66, "l114c1": 66, "l124c3": 66, "wheel": 66, "dep": 66, "lighter": 66, "executor": 66, "avoid": [66, 92, 97, 109], "implic": 66, "python_onli": 66, "legaci": [66, 73], "mainten": 66, "torchdynamo": [66, 109, 112], "technolog": [66, 112], "project": [66, 80, 85], "exclud": [66, 92], "speed": [66, 94, 95, 97, 106], "no_torchscript": 66, "dbg": 66, "pre_cxx11_abi": 66, "complic": 66, "incompat": 66, "popular": [66, 80, 107], "ngc": [66, 108], "tabl": [66, 85], "bdist_wheel": 66, "preinstal": 66, "forum": 66, "correctli": [66, 92], "declar": 66, "intend": [66, 98, 99, 100, 101, 102, 103, 104], "microsoft": 66, "2022": [66, 68], "open": [66, 107, 108], "app": 66, "x64": 66, "prompt": [66, 96, 100, 103, 104], "admin": 66, "privileg": 66, "launcher": 66, "chocolatei": 66, "navig": [66, 79], "ninja": 66, "setuptool": 66, "r": [66, 81], "txt": 66, "distutils_use_sdk": 66, "cuda_win": 66, "libtorch_win": 66, "tensorrt_win": 66, "non": [66, 75, 82, 84, 110], "similarli": [66, 94, 102, 110], "ci_workspac": 66, "win": 66, "tmpl": 66, "torchtrtc": [66, 68, 113], "websit": 66, "finder": 66, "dcmake_module_path": 66, "doesn": [66, 81, 87, 88], "dtorch_dir": 66, "dtensorrt_root": 66, "choic": [66, 73], "b": [66, 69, 75, 82, 108], "dcmake_build_typ": 66, "72048": 66, "jp_workspac": 66, "new_local_repositori": 66, "sudo": 66, "home": 66, "unlik": [66, 91], "libtorch_pre_cxx11_abi": 66, "shift": [66, 69, 81], "jetpack": 66, "jetpack_x": 66, "jetpack_5": 66, "drop": [66, 79, 105], "anywher": 67, "ahead": [67, 68, 88, 96], "ep": [67, 69, 97, 111], "output_format": [67, 75, 111], "input_tensor": [67, 92, 105], "fill": 67, "aot": [68, 88, 96, 97, 112], "integr": [68, 95, 96, 98, 106], "seamlessli": [68, 75], "ecosystem": [68, 112], "hybrid": [68, 70, 75, 76, 112], "instal": [68, 85, 88, 108, 110], "triton": [68, 92], "page": [68, 83, 85, 108], "introductori": 68, "blog": [68, 110], "gtc": 68, "2020": [68, 88], "talk": 68, "fall": [68, 75, 92], "2021": 68, "dai": 68, "confer": 68, "_convolut": [69, 88], "stride": [69, 75, 92, 105], "pad": [69, 75, 92, 105], "dilat": 69, "output_pad": 69, "group": [69, 81, 82], "determinist": 69, "cudnn_en": 69, "allow_tf32": 69, "ab": 69, "aco": 69, "acosh": 69, "adaptive_avg_pool1d": 69, "output_s": 69, "adaptive_avg_pool2d": 69, "adaptive_avg_pool3d": 69, "adaptive_max_pool1d": 69, "adaptive_max_pool2d": 69, "adaptive_max_pool3d": 69, "argmax": 69, "keepdim": 69, "argmin": 69, "asin": 69, "asinh": 69, "atan": 69, "atanh": 69, "avg_pool1d": 69, "kernel_s": [69, 92, 105], "ceil_mod": 69, "count_include_pad": 69, "avg_pool2d": 69, "divisor_overrid": 69, "avg_pool3d": 69, "gamma": 69, "var": 69, "momentum": 69, "bitwise_not": 69, "bmm": 69, "ceil": 69, "clamp": 69, "clamp_max": 69, "clamp_min": 69, "constant_pad_nd": 69, "co": [69, 82, 107], "cosh": 69, "cumsum": 69, "tensor_mod": 69, "rounding_mod": 69, "div_": 69, "elu": 69, "scale": [69, 90, 107], "input_scal": 69, "indic": [69, 79, 81, 97, 99, 109], "padding_idx": 69, "eq": [69, 81], "erf": 69, "exp": 69, "expand_a": 69, "fake_quantize_per_channel_affin": 69, "zero_point": 69, "axi": [69, 75], "quant_min": 69, "quant_max": 69, "fake_quantize_per_tensor_affin": 69, "using_int": [69, 88], "start_dim": [69, 88], "end_dim": [69, 88], "floor": 69, "floor_divid": 69, "ge": 69, "gru_cel": 69, "hx": 69, "w_ih": 69, "w_hh": 69, "b_ih": 69, "b_hh": 69, "gt": 69, "hardtanh_": 69, "instance_norm": 69, "running_mean": 69, "running_var": 69, "use_input_stat": 69, "layer_norm": 69, "normalized_shap": 69, "le": 69, "negative_slop": 69, "01": [69, 82, 88, 105], "leaky_relu_": 69, "lstm_cell": 69, "lt": 69, "masked_fil": 69, "mask": [69, 92], "max_pool1d": 69, "max_pool2d": [69, 87, 88], "max_pool3d": 69, "mul_": 69, "narrow": 69, "neg": [69, 96], "norm": 69, "scalaropt_dim": 69, "pixel_shuffl": 69, "upscale_factor": 69, "pow": 69, "tensor_scalar": 69, "expon": 69, "tensor_tensor": 69, "prelu": 69, "prod": [69, 92], "dim_int": 69, "reciproc": 69, "reflection_pad1d": 69, "reflection_pad2d": 69, "relu_": 69, "repeat_interleav": 69, "self_int": 69, "replication_pad1d": 69, "replication_pad2d": 69, "replication_pad3d": 69, "reshap": [69, 92, 108], "roll": 69, "rsub": 69, "scatter": 69, "sigmoid_": 69, "sin": [69, 81], "sinh": 69, "slice": 69, "split_siz": 69, "split_with_s": 69, "sqrt": 69, "squar": 69, "squeez": [69, 107], "sub_": 69, "dim_intlist": 69, "tan": 69, "tanh": 69, "tanh_": 69, "non_block": [69, 105], "memory_format": [69, 75], "prim_devic": 69, "topk": 69, "k": [69, 90, 105], "largest": 69, "dim0": [69, 94], "dim1": 69, "unbind": 69, "unsqueez": 69, "upsample_bilinear2d": 69, "align_corn": 69, "scales_h": 69, "scales_w": 69, "vec": 69, "scale_factor": 69, "upsample_linear1d": 69, "upsample_nearest1d": 69, "upsample_nearest2d": 69, "upsample_nearest3d": 69, "scales_d": 69, "upsample_trilinear3d": 69, "view": [69, 79], "__and__": 69, "__derive_index": 69, "idx": 69, "__getitem__": 69, "__is__": 69, "t1": 69, "t2": 69, "obj": 69, "__isnot__": 69, "__not__": 69, "__or__": 69, "__range_length": 69, "lo": 69, "hi": [69, 81, 82], "__round_to_zero_floordiv": 69, "__xor__": 69, "append": [69, 93, 94, 105], "el": 69, "arang": [69, 92], "pin_memori": 69, "start_step": 69, "copy_": 69, "float_int": 69, "int_float": 69, "floordiv": 69, "is_floating_point": 69, "numel": 69, "l": [69, 105], "9223372036854775807": 69, "requires_grad": 69, "tupleindex": 69, "tup": 69, "exported_program": [70, 75, 111], "arg_input": [70, 75, 97], "kwarg_input": [70, 75, 97], "engine_cache_dir": [70, 93, 94], "engine_cache_s": [70, 93, 94], "custom_engine_cach": [70, 94], "baseenginecach": [70, 94], "int32": [70, 75, 76, 92, 93, 101, 107], "channel_last": [70, 75, 76, 107], "244": [70, 75, 76], "alia": [70, 75], "better": [70, 75, 87, 107, 112], "understand": [70, 75, 109], "convolut": [70, 75, 76, 90, 92, 113], "_c": [70, 75, 76, 91], "oppos": [70, 75, 76], "lean": [70, 75], "spend": [70, 75], "integ": [70, 75, 84], "faster": [70, 75, 93, 94, 107], "parition": [70, 75], "increas": [70, 75, 94], "amount": [70, 75], "defer": [70, 75, 112], "lead": [70, 75, 81, 110], "oversubscript": [70, 75], "hard": [70, 97], "disk": [70, 75, 94], "space": [70, 81, 82, 90], "byte": [70, 74, 75, 76, 92, 94, 107], "1gb": [70, 93, 94], "exce": 70, "oldest": 70, "gear": [70, 90], "toward": [70, 90], "refit_module_weight": [70, 97], "compiled_modul": [70, 97], "new_weight_modul": [70, 97], "verify_output": [70, 97], "use_weight_map_cach": [70, 97], "in_plac": [70, 97], "compmil": 70, "coverag": [70, 92], "min_acc_module_s": 71, "is_aten": 71, "use_experimental_fx_rt": 71, "correctness_atol": 71, "correctness_rtol": 71, "minim": [71, 90, 92], "submodul": [71, 87, 92], "fx2trt": 71, "cpu": [71, 103, 104], "has_batch_dim": 71, "dtyep": 71, "prop": 71, "min_input_shap": 71, "optimized_input_shap": 71, "max_input_shap": 71, "popul": 71, "225": [71, 108], "explicit_precis": 71, "logger_level": 71, "model_trt": 72, "model_torchtrt": 72, "internal_error": 72, "toolkit": 73, "dataloadercalibr": [73, 90], "preprocess": [73, 90, 108], "algo_typ": [73, 90], "calibrationalgo": [73, 90], "cachecalibr": [73, 90], "qualnam": [73, 75], "entropy_calibr": 73, "entropy_calibration_2": [73, 90], "legacy_calibr": 73, "minmax_calibr": 73, "set_multi_device_safe_mod": [74, 110], "_multidevicesafemodecontextmanag": 74, "impact": 74, "suppress": 74, "unsaf": 74, "trt_compiled_modul": 74, "torchtensorrtmodul": [74, 92], "encompass": [74, 76], "simpili": 74, "de": 74, "initi": [74, 75, 81, 97, 98, 99, 101, 102, 103, 104], "scriptmodul": [74, 75, 76, 88, 89, 111, 112], "overridden": [74, 75], "subclass": 74, "although": [74, 81], "recip": [74, 90], "afterward": 74, "former": 74, "care": 74, "hook": 74, "silent": 74, "get_extra_st": 74, "state_dict": [74, 75, 96], "set_extra_st": 74, "picklabl": 74, "pickl": [74, 92, 94], "load_state_dict": [74, 96, 105], "pythontorchtensorrtmodul": 74, "serialized_engin": [74, 76], "_set": [74, 98], "weight_name_map": 74, "trt_modul": 74, "engine_str": 74, "my_modul": 74, "current_devic": 74, "cudagraphs_validate_shap": 74, "versu": 74, "disable_profil": 74, "enable_profil": 74, "iprofil": 74, "spent": 74, "get_layer_info": 74, "request": [75, 88, 108], "decid": 75, "deseri": [75, 76, 88, 92], "retrac": 75, "strict": [75, 110], "valueerror": 75, "mutabletorchtensorrtmodul": [75, 95, 96, 106], "pytorch_model": 75, "regular": 75, "whenev": 75, "refit_gm": 75, "shape_mod": 75, "_shapemod": 75, "interv": 75, "notat": 75, "bound": 75, "torch_tensor": 75, "tracer": 75, "example_tensor": 75, "optimization_profile_field": 75, "classmethod": 75, "disable_memory_format_check": 75, "core_id": 75, "schedul": [75, 108], "use_default": 75, "try_to": 75, "anoth": [75, 81, 82, 87, 89, 97], "typeerror": 75, "unknown": 75, "succe": 75, "float_dtyp": 75, "failur": 75, "bf16": 75, "try_from": [75, 92], "complex128": 75, "16": [75, 85, 87, 88, 89, 99, 102], "brain": 75, "bfloat16": 75, "f64": 75, "f8": 75, "fp8": [75, 95, 106], "float8": 75, "i32": 75, "sign": [75, 108], "i64": 75, "u8": 75, "unsign": 75, "uint8": 75, "trt_dla": 75, "torchtrt_dla": 75, "_from": 75, "torchtrt_dla_ec": 75, "torchtrt_safety_ec": 75, "saefti": 75, "trt_dla_ec": 75, "standalon": [75, 81], "certifi": 75, "tf": 75, "torchtrt_linear": 75, "cdhw32": 75, "thirti": 75, "row": [75, 82], "spatial": 75, "31": [75, 88], "subscript": [75, 81], "chw16": 75, "sixteen": 75, "15": [75, 81, 85], "chw2": 75, "chw32": 75, "chw4": 75, "four": [75, 81, 82], "dhwc": 75, "equivi": 75, "channels_last_3d": 75, "dhwc8": 75, "eight": 75, "dla_hwc4": 75, "imag": [75, 90, 92, 96, 100, 105, 108], "roundup": 75, "elements": 75, "dla_linear": 75, "planar": 75, "hwc": 75, "channels_last": 75, "hwc16": 75, "hwc8": 75, "least": [75, 81, 82], "ishapelay": 76, "check_method_op_support": 76, "seriali": 76, "put_binding_nam": 76, "tensorrtcompilespec": [76, 91], "scriptclass": 76, "0x7fad0f3e7bf0": 76, "_jit_to_tensorrt": 76, "00": 77, "000": [77, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "total": [77, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "galleri": [77, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "mem": 77, "advanc": [77, 82, 90, 95, 106], "torch_compile_advanced_usag": [77, 98], "torch_compile_resnet_exampl": [77, 99], "diffus": [77, 95, 106], "torch_compile_stable_diffus": [77, 100], "transform": [77, 88, 90, 93, 95, 97, 103, 104, 105, 106, 108, 111], "torch_compile_transformers_exampl": [77, 101], "v0": [78, 108], "pytorch_sphinx_them": [79, 86], "conf": [79, 86], "html_theme_opt": 79, "canonical_url": 79, "analytics_id": 79, "logo_onli": 79, "display_vers": 79, "prev_next_buttons_loc": 79, "bottom": 79, "style_external_link": 79, "vcs_pageview_mod": 79, "collapse_navig": 79, "sticky_navig": [79, 83], "navigation_depth": 79, "includehidden": 79, "titles_onli": 79, "canon": 79, "rank": 79, "trail": 79, "slash": 79, "googl": 79, "analyt": 79, "With": [79, 81, 88, 90, 94, 108], "isn": [79, 81, 92], "shown": [79, 81, 88], "sidebar": [79, 85], "button": [79, 81], "icon": [79, 81], "extern": [79, 81], "display_github": 79, "display_gitlab": 79, "gitlab": 79, "bitbucket": 79, "bar": [79, 81], "www": [79, 81, 88, 90, 108], "sphinx": [79, 80, 81, 82, 86, 92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "en": 79, "toctre": 79, "lose": 79, "scroll": [79, 83], "unlimit": 79, "header": [79, 81, 82, 88, 108], "render": 79, "github_url": 79, "bitbucket_url": 79, "gitlab_url": 79, "left": [79, 81], "upon": [79, 98, 101], "rst": [79, 81], "visitor": 79, "revert": 79, "misbuild": 79, "show": [79, 81, 94, 100, 107], "properti": [79, 92], "stick": 79, "screen": 79, "vertic": [79, 81], "too": [79, 81, 82], "sticki": [79, 85], "nav": [79, 85], "At": [80, 97], "django": 80, "payment": 80, "dotpai": 80, "dotpayprovid": 80, "seller_id": 80, "pin": 80, "lock": 80, "lang": 80, "pl": 80, "polish": 80, "gatewai": 80, "transfer": 80, "purchas": 80, "item": [80, 82, 105], "param": 80, "seller": 80, "consult": 80, "ui": 80, "languag": [80, 81, 82, 87, 92, 108], "data_item_1": 80, "emphasi": 81, "strong": 81, "hyperlink": 81, "cross": 81, "uri": 81, "web": 81, "anonym": 81, "label": [81, 90, 105, 107, 108], "substitut": 81, "charact": 81, "exceedingli": 81, "ugli": 81, "problem": 81, "problemat": 81, "ext": [81, 82], "autodoc": [81, 82], "demo": [81, 90], "test_py_modul": [81, 85], "my": 81, "role": 81, "pep": 81, "287": 81, "rfc": 81, "2822": 81, "superscript": 81, "gui": 81, "taken": 81, "height": 81, "interfer": 81, "press": 81, "keyboard": 81, "mous": 81, "mmb": 81, "menuselect": 81, "seen": [81, 82], "whitespac": 81, "signific": [81, 92], "strang": 81, "hyphen": 81, "word": [81, 107], "adjust": 81, "width": [81, 107], "browser": 81, "window": 81, "sentenc": [81, 107], "suppli": [81, 97], "258": 81, "equat": 81, "x_": 81, "x_0": 81, "x_1": 81, "x_2": 81, "x_3": 81, "x_4": 81, "nabla": 81, "frac": 81, "theta": 81, "phi": 81, "restructuredtext": [81, 82], "parser": [81, 105], "colon": 81, "indent": 81, "literal_block": 81, "spaces_and_linebreak": 81, "preserv": [81, 87, 90], "markup_process": 81, "Or": 81, "great": [81, 88, 92, 94, 112], "why": [81, 110], "didn": 81, "blank": 81, "align": 81, "permit": 81, "awai": 81, "eric": 81, "orchestra": 81, "leader": 81, "bee": 81, "philosoph": 81, "ipso": 81, "facto": 81, "But": [81, 88, 97], "got": [81, 88], "vi": 81, "entiti": 81, "said": 81, "entir": [81, 112], "ancient": 81, "injuri": 81, "sing": 81, "elk": 81, "bracket": 81, "miss": [81, 88], "brontosaurus": 81, "thin": 81, "thicker": 81, "middl": 81, "That": [81, 88], "mine": 81, "belong": 81, "me": [81, 82], "ann": 81, "begun": 81, "cut": [81, 97], "past": 81, "pars": [81, 88], "someurl": 81, "dev0": 81, "a4a9419": 81, "caption": [81, 84], "pane": 81, "shell_command": 81, "echo": 81, "did": 81, "window_nam": 81, "session_nam": 81, "shorthand": 81, "some_funct": 81, "highlight": 81, "THE": 81, "heaven": 81, "hexagram": 81, "six": 81, "unbroken": 81, "primal": 81, "light": [81, 111], "spirit": 81, "weak": 81, "essenc": 81, "energi": 81, "unrestrict": 81, "conceiv": 81, "motion": 81, "regard": [81, 112], "basi": 81, "thu": 81, "persist": 81, "dual": 81, "sens": [81, 88], "univers": 81, "world": 81, "men": 81, "express": 81, "deiti": 81, "human": 81, "denot": [81, 92], "holi": 81, "man": [81, 82], "sage": 81, "ruler": 81, "who": 81, "awaken": 81, "utf": [81, 82], "sphinx_rtd_them": [81, 82], "docstr": [81, 82, 89], "dl": 81, "dt": 81, "tag": [81, 108], "tt": 81, "descnam": 81, "descclassnam": 81, "wrote": 81, "anyth": [81, 82, 110], "programm": 81, "myclass": 81, "dothismethod": 81, "flush": 81, "meth": 81, "capit": 81, "flox": 81, "unreferenc": 81, "nonexist": 81, "extrem": 81, "stuff": 81, "mayb": 81, "bold": 81, "ital": 81, "heck": 81, "put": [81, 107], "13": [81, 85], "backlink": 81, "knowledg": 81, "mind": 81, "ey": 81, "thought": 81, "medium": 81, "peopl": 81, "subsect": 81, "interpol": 81, "indirect": 81, "phrase": 81, "docutil": [81, 82], "sourceforg": [81, 82], "ref": 81, "clickabl": 81, "legend": 81, "revis": [81, 82, 96, 100], "revisit": 81, "enhanc": 81, "structuredtext": 81, "wooden": 81, "nickel": 81, "mad": 81, "scientist": 81, "bigger": 81, "bread": 81, "box": [81, 109, 112], "wash": 81, "behind": 81, "ear": 81, "room": 81, "closet": 81, "bathroom": 81, "trash": 81, "sink": 81, "mother": 81, "g_": 81, "mu": 81, "nu": 81, "pi": 81, "t_": 81, "rho_": 81, "servic": 81, "thing1": 81, "thing2": 81, "thing3": 81, "prose": 81, "provok": 81, "mental": 81, "exert": 81, "reader": 81, "discret": 81, "strongli": 81, "advis": 81, "subtitl": 81, "outsid": 81, "often": 81, "besid": 81, "border": 81, "background": [81, 87], "ok": [81, 88], "transmit": 81, "disconnect": 81, "nonetheless": 81, "semant": 81, "blue": [81, 92], "white": 81, "arab": 82, "roman": 82, "upper": 82, "iii": 82, "iv": 82, "classifi": [82, 87, 88, 105, 107], "paragraph": [82, 85], "z": 82, "commonli": 82, "vm": 82, "david": 82, "goodger": 82, "address": [82, 92, 96], "123": 82, "street": 82, "canada": 82, "a1b": 82, "2c3": 82, "contact": 82, "myself": 82, "organ": 82, "humankind": 82, "2012": 82, "03": 82, "19": [82, 85], "53": 82, "0000": 82, "tue": 82, "jan": 82, "progress": 82, "7302": 82, "wish": 82, "redistribut": 82, "reattribut": 82, "sell": 82, "bui": 82, "rent": 82, "leas": 82, "improv": [82, 110], "quot": 82, "excerpt": 82, "incorpor": 82, "collat": 82, "fold": 82, "stapl": 82, "mutil": 82, "anyon": 82, "heart": 82, "bibliograph": 82, "markup": [82, 85], "literal": 82, "yahoo": 82, "oh": 82, "liter": 82, "heh": 82, "child": 82, "beat": 82, "text": [82, 84, 107], "hehe": 82, "kept": 82, "sai": [82, 107], "cackl": 82, "night": 82, "lone": 82, "guangzhou": 82, "destini": 82, "hope": 82, "dream": 82, "forth": 82, "fifth": 82, "sixth": 82, "lorem": [82, 84], "ipsum": [82, 84], "dolor": [82, 84], "sit": [82, 84], "amet": [82, 84], "consectetur": [82, 84], "adipisc": [82, 84], "elit": [82, 84], "donec": [82, 84], "porttitor": [82, 84], "odio": [82, 84], "posuer": [82, 84], "vita": [82, 84], "ornar": [82, 84], "libero": [82, 84], "matti": 82, "loborti": [82, 84], "justo": [82, 84], "vestibulum": [82, 84], "nibh": [82, 84], "aliquet": [82, 84], "sed": [82, 84], "feugiat": [82, 84], "sagitti": [82, 84], "nequ": [82, 84], "qui": [82, 84], "eleifend": 82, "dui": [82, 84], "rutrum": [82, 84], "lectu": [82, 84], "suscipit": [82, 84], "letter": [82, 107], "column": 82, "cell": 82, "span": 82, "nam": [82, 84], "mauri": [82, 84], "arcu": [82, 84], "stub": 82, "behav": 83, "area": 83, "interdum": 84, "nec": 84, "finibu": 84, "dictum": 84, "velit": 84, "ut": 84, "eu": 84, "efficitur": 84, "aliquam": 84, "erat": 84, "diam": 84, "gravida": 84, "imperdiet": 84, "tellu": 84, "nisl": 84, "praesent": 84, "eget": 84, "elementum": 84, "rhoncu": 84, "tincidunt": 84, "suspendiss": 84, "volutpat": 84, "scelerisqu": 84, "tristiqu": 84, "aenean": 84, "condimentum": 84, "risu": 84, "accumsan": 84, "laoreet": 84, "maximu": 84, "sapien": 84, "ligula": 84, "fringilla": 84, "commodo": 84, "proin": 84, "et": 84, "pharetra": 84, "etiam": 84, "turpi": 84, "ant": 84, "luctu": 84, "vel": 84, "malesuada": 84, "dignissim": 84, "mi": 84, "nunc": 84, "augu": 84, "sem": 84, "cursu": 84, "nulla": 84, "pellentesqu": 84, "habit": 84, "morbi": 84, "senectu": 84, "netu": 84, "fame": 84, "ac": 84, "egesta": 84, "placerat": 84, "tortor": 84, "iaculi": 84, "venenati": 84, "cra": 84, "puru": 84, "ero": 84, "vehicula": 84, "fusc": 84, "auctor": 84, "phasellu": 84, "est": 84, "viverra": 84, "conval": 84, "faucibu": 84, "vulput": 84, "feli": 84, "sodal": 84, "maecena": 84, "congu": 84, "semper": 84, "enim": 84, "blandit": 84, "sollicitudin": 84, "urna": 84, "orci": 84, "lacu": 84, "quisqu": 84, "facilisi": 84, "hendrerit": 84, "curabitur": 84, "variu": 84, "bibendum": 84, "massa": 84, "magna": 84, "tempu": 84, "metu": 84, "nisi": 84, "pretium": 84, "leo": 84, "euismod": 84, "ultric": 84, "dapibu": 84, "lacinia": 84, "vivamu": 84, "molesti": 84, "hac": 84, "habitass": 84, "platea": 84, "dictumst": 84, "git": 85, "content": [85, 90, 108], "changelog": 85, "math": 85, "9": [85, 88, 92, 108], "14": [85, 93, 101, 108], "17": 85, "18": [85, 88, 96], "submenu": 85, "symlink": 86, "subtre": 86, "_theme": 86, "html_theme": 86, "html_theme_path": 86, "optimiz": 87, "tutori": [87, 90, 92, 94, 96, 97], "beginn": 87, "intro_to_torchscript_tutori": 87, "briefli": 87, "lenet": [87, 88], "lenetfeatextractor": 87, "conv1": [87, 88], "conv2d": [87, 92, 105], "conv2": [87, 88], "lenetclassifi": 87, "fc1": [87, 88], "120": [87, 88], "fc2": [87, 88], "84": [87, 88], "fc3": [87, 88], "feat": [87, 88], "obvious": 87, "pathwai": 87, "input_data": [87, 89], "traced_model": 87, "pick": 87, "script_model": [87, 91], "perspect": 87, "___torch_mangle_10": 87, "129": 87, "___torch_mangle_9": 87, "119": 87, "___torch_mangle_5": 87, "137": 87, "callmethod": 87, "138": 87, "38": 87, "39": 87, "torch_script_modul": [87, 88], "in_tensor": 87, "fly": 87, "lenet_script": [87, 88], "haven": 88, "acquir": 88, "dyanmo": 88, "almost": [88, 112], "trt_lenet_script": 88, "apr": 88, "56": 88, "04": [88, 108], "credit": 88, "stop": 88, "argc": 88, "argv": 88, "cerr": 88, "cout": 88, "even": [88, 96], "cppdoc": 88, "pretti": 88, "fashion": [88, 107], "enable_precis": 88, "And": 88, "convertgraphtotrtengin": 88, "engine_converted_from_jit": 88, "close": 88, "saw": 88, "576": 88, "346": 88, "539": 88, "0464": 88, "0383": 88, "0678": 88, "0932": 88, "1045": 88, "0805": 88, "0435": 88, "0818": 88, "0208": 88, "0358": 88, "cudafloattyp": 88, "0530": 88, "1691": 88, "2802": 88, "1502": 88, "1056": 88, "1549": 88, "input0": [88, 89], "1063": 88, "input1": [88, 89], "input2": 88, "28": 88, "29": 88, "33": 88, "35": 88, "36": 88, "37": 88, "compilegraph": [88, 90], "laid": 88, "translat": [88, 97], "aren": 88, "techniqu": [88, 90, 110], "checkmethodoperatorsupport": 88, "modular": 88, "ship": [88, 110], "exhaust": 88, "109": 88, "addlay": 88, "yourself": 88, "question": 88, "outself": 88, "flatten_convert": 88, "unwraptoint": 88, "in_shap": 88, "tovec": 88, "out_shap": 88, "shuffl": [88, 90, 105], "addshuffl": 88, "setreshapedimens": 88, "todim": 88, "extens": [88, 112], "ctype": 88, "cdll": 88, "contributor": 88, "upstream": 88, "pr": 88, "usecas": [89, 106], "sole": [89, 90, 112], "individu": 89, "accuraci": [90, 107], "loss": [90, 107], "infrastructur": [90, 108], "streamlin": 90, "expos": [90, 92], "cpp_frontend": 90, "loading_data_recip": 90, "cifar10": [90, 105], "cstddef": 90, "ktrain": 90, "ktest": 90, "un": 90, "cs": 90, "toronto": 90, "edu": 90, "kriz": 90, "cifar": 90, "is_train": 90, "trim": 90, "use_subset": 90, "new_siz": 90, "mode_": 90, "images_": 90, "targets_": 90, "calibration_dataset": 90, "data_dir": 90, "320": 90, "4914": [90, 105], "4822": [90, 105], "4465": [90, 105], "2023": [90, 105], "1994": [90, 105], "2010": [90, 105], "dataloaderopt": 90, "worker": 90, "simpler": 90, "virtual": 90, "input_shap": [90, 113], "compile_spec": [90, 99, 113], "kf16": [90, 113], "ki8": 90, "vgg16": [90, 95, 105, 106], "testing_dataset": [90, 105], "totensor": [90, 105, 108], "testing_dataload": [90, 105], "num_work": [90, 105], "vgg": [90, 105], "test_ptq_dataloader_calibr": 90, "test_ptq_trt_calibr": 90, "krizhevski": 90, "hinton": 90, "2009": 90, "tini": 90, "simonyan": 90, "zisserman": 90, "2014": 90, "recognit": [90, 107], "arxiv": 90, "preprint": 90, "1409": 90, "1556": 90, "_jit_to_backend": 91, "mobilenet_v2": 91, "pretrain": [91, 94, 96, 99, 102, 107, 108], "cost": [92, 94, 97, 110], "perhap": 92, "overhead": [92, 110], "sake": 92, "circular": 92, "red": 92, "green": 92, "twice": 92, "written": 92, "openai": 92, "formal": 92, "tl": 92, "custom_op": 92, "circ_pad_kernel": 92, "all_pads_0": 92, "all_pads_2": 92, "all_pads_4": 92, "all_pads_6": 92, "orig_dims_0": 92, "orig_dims_1": 92, "orig_dims_2": 92, "orig_dims_3": 92, "y_shape_1": 92, "y_shape_2": 92, "y_shape_3": 92, "x_len": 92, "y_len": 92, "block_siz": 92, "pid": 92, "program_id": 92, "mask_i": 92, "i3": 92, "i2": 92, "i1": 92, "i0": 92, "j0": 92, "j1": 92, "j2": 92, "j3": 92, "load_idx": 92, "mask_x": 92, "launch": [92, 108], "torchtrt_ex": 92, "triton_circular_pad": 92, "mutates_arg": 92, "out_dim": 92, "tolist": 92, "all_pad": 92, "zero": 92, "orig_dim": 92, "blocksiz": 92, "256": [92, 105, 108], "numblock": 92, "ex_input": 92, "tracabl": 92, "prerequisit": 92, "fake": 92, "real": 92, "faketensor": 92, "register_fak": 92, "autograd": 92, "beyond": 92, "register_autograd": 92, "padded_x": 92, "my_model": 92, "2604": 92, "4232": 92, "3041": 92, "0833": 92, "2461": 92, "1270": 92, "2450": 92, "4079": 92, "2887": 92, "2828": 92, "0373": 92, "0332": 92, "3143": 92, "6344": 92, "5638": 92, "1867": 92, "5068": 92, "4363": 92, "7937": 92, "3488": 92, "1350": 92, "7966": 92, "3517": 92, "1379": 92, "5537": 92, "1088": 92, "8950": 92, "0550": 92, "6163": 92, "0109": 92, "5245": 92, "9632": 92, "5686": 92, "3775": 92, "8162": 92, "4216": 92, "4311": 92, "1649": 92, "2091": 92, "3668": 92, "1006": 92, "1447": 92, "0352": 92, "7689": 92, "8131": 92, "_run_on_gpu_0": 92, "_run_on_acc_1": 92, "dry": 92, "50": [92, 107], "count": 92, "__": 92, "were": [92, 97, 110], "aggreg": 92, "stat": 92, "latenc": [92, 110], "abstractli": 92, "pkl": [92, 96], "cupi": 92, "gap": 92, "prealloc": 92, "circularpaddingplugin": 92, "ipluginv2dynamicext": 92, "field_collect": 92, "pluginfieldcollect": 92, "x_shape": 92, "num_output": 92, "plugin_namespac": 92, "plugin_typ": 92, "plugin_vers": 92, "assert": [92, 96, 97], "get_output_datatyp": 92, "input_typ": 92, "get_output_dimens": 92, "output_index": 92, "dimsexpr": 92, "exprbuild": 92, "iexprbuild": 92, "output_dim": 92, "dimensionoper": 92, "configure_plugin": 92, "inp": 92, "dynamicplugintensordesc": 92, "x_dim": 92, "desc": 92, "supports_format_combin": 92, "po": 92, "in_out": 92, "plugintensordesc": 92, "num_input": 92, "enqueu": 92, "input_desc": 92, "output_desc": 92, "in_dtyp": 92, "a_mem": 92, "unownedmemori": 92, "items": 92, "c_mem": 92, "a_ptr": 92, "memorypoint": 92, "c_ptr": 92, "a_d": 92, "memptr": 92, "c_d": 92, "a_t": 92, "as_tensor": 92, "c_t": 92, "cloned_plugin": 92, "__dict__": 92, "circularpaddingplugincr": 92, "iplugincr": 92, "field_nam": 92, "pluginfield": 92, "pluginfieldtyp": 92, "create_plugin": 92, "pluginfieldcollection_": 92, "deserialize_plugin": 92, "pads_dict": 92, "creator": 92, "trt_plugin_registri": 92, "get_plugin_registri": 92, "register_cr": 92, "untyp": 92, "get_trt_tensor": 92, "set_layer_nam": 92, "recal": 92, "intlist": 92, "circular_padding_convert": 92, "retriev": 92, "elsewher": 92, "plugin_registri": 92, "plugin_cr": 92, "get_plugin_cr": 92, "field_config": 92, "eventu": 92, "freez": 92, "_input": 92, "add_plugin_v2": 92, "circular_padding_plugin": 92, "_run_on_acc_0": 92, "grad_fn": 92, "subbackward0": 92, "minut": [92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "custom_kernel_plugin": 92, "jupyt": [92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106], "ipynb": [92, 93, 94, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105], "engine_caching_exampl": [93, 94], "remove_timing_cach": [93, 94], "bertmodel": [93, 101], "random": [93, 94, 96, 97], "seed": [93, 94, 96, 97], "manual_se": [93, 94, 96, 97], "from_pretrain": [93, 96, 100, 101, 103, 104], "uncas": [93, 101, 107], "return_dict": 93, "randint": [93, 101], "compile_bert": 93, "enable_tim": [93, 94], "1st": [93, 94], "measur": [93, 94], "2nd": [93, 94], "3rd": [93, 94], "slower": [93, 94], "messur": [93, 94], "compilation_kwarg": [93, 101], "torch_trt_bert_engine_cach": 93, "30": [93, 94, 96, 97, 99, 101], "synchron": [93, 94], "elapsed_tim": [93, 94], "millisecond": 93, "__name__": [93, 98, 101], "__main__": [93, 98, 101], "engine_caching_bert_exampl": 93, "paid": 94, "upfront": 94, "invalid": 94, "repeatedli": 94, "mitig": 94, "explor": [94, 106], "resnet18": [94, 96, 97, 99, 102], "torch_trt": [94, 96, 97], "_default": 94, "_engine_cach": 94, "flexibl": [94, 112], "histor": 94, "barrier": 94, "reconstruct": 94, "prior": [94, 109, 110], "ti": 94, "hash": 94, "magnitud": 94, "make_refitt": 94, "torch_compil": [94, 98, 99, 101, 102, 109, 112], "compiled_model": 94, "ms": 94, "dynamo_compil": 94, "example_input": 94, "200": 94, "dynamic_shap": [94, 109], "remot": 94, "systen": 94, "agnost": 94, "implent": 94, "ramenginecach": 94, "held": 94, "engine_cach": 94, "torch_compile_my_cach": 94, "cudagraph": [95, 106], "mutabl": [95, 106], "vgg16_fp8_ptq": [95, 106], "bert": [95, 101, 106], "gpt2": [95, 106], "llama2": [95, 106], "straightforward": 96, "especi": 96, "hug": [96, 103, 104], "face": [96, 103, 104], "difficult": 96, "ever": 96, "walk": [96, 97], "lora": [96, 97], "use_python": 96, "mutable_modul": 96, "model2": [96, 97], "expected_output": [96, 97], "refitted_output": [96, 97], "allclos": [96, 97], "reload": [96, 112], "checkpoint": [96, 105], "civitai": 96, "12597": 96, "moxin": 96, "diffusionpipelin": [96, 100], "no_grad": [96, 103, 104, 105], "model_id": [96, 100], "runwayml": 96, "v1": [96, 100], "hous": 96, "forest": 96, "shuimobysim": 96, "wuchangshuo": 96, "qualiti": 96, "worst": 96, "lowr": 96, "focu": 96, "cloudi": 96, "watermark": 96, "pipe": [96, 100], "torch_dtyp": [96, 100], "unet": [96, 100], "negative_prompt": 96, "num_inference_step": 96, "without_lora_mut": 96, "jpg": [96, 108], "procedur": 96, "load_lora_weight": 96, "stablediffusionapi": 96, "load_lora_embed": 96, "weight_nam": 96, "safetensor": 96, "adapter_nam": 96, "lora1": 96, "set_adapt": 96, "adapter_weight": 96, "fuse_lora": 96, "unload_lora_weight": 96, "with_lora_mut": 96, "mutable_torchtrt_module_exampl": 96, "expens": 97, "involv": 97, "occasion": [97, 98, 101], "adapt": 97, "infeas": 97, "focus": 97, "mostli": 97, "recogn": 97, "behalf": 97, "init": [97, 105], "sett": 97, "randomli": 97, "exp_program2": 97, "compiled_trt_ep": 97, "new_trt_gm": 97, "accomplish": 97, "gaurente": 97, "attempt": [97, 105, 109], "rebuild": 97, "heurist": 97, "refit_engine_exampl": 97, "x_out": 98, "y_out": 98, "x_y_out": 98, "invoc": 98, "sample_inputs_half": 98, "model_half": 98, "backend_kwarg": 98, "optimized_model_custom": 98, "exit": [98, 101, 108], "2052": [98, 101], "compile_engine_and_inf": [98, 101], "new_input": [99, 101], "new_output": [99, 101], "new_batch_size_input": 99, "new_batch_size_output": 99, "inputs_bs8": 99, "mark_dynam": [99, 109], "outputs_bs8": 99, "No": [99, 109], "inputs_bs12": 99, "outputs_bs12": 99, "compvi": 100, "majest": 100, "castl": 100, "cloud": 100, "majestic_castl": 100, "png": 100, "enable_cudagraph": [102, 110], "out_trt": 102, "set_cudagraphs_mod": [102, 110], "inputs_2": 102, "inputs_3": 102, "out_trt_2": 102, "out_trt_3": 102, "torch_export_cudagraph": 102, "automodelforcausallm": [103, 104], "autotoken": [103, 104], "export_llm": [103, 104], "max_token": [103, 104], "kv_cach": [103, 104], "token": [103, 104, 107], "pad_token_id": 103, "eos_token_id": [103, 104], "attn_implement": [103, 104], "eager": [103, 104], "model_input": [103, 104], "return_tensor": [103, 104], "input_id": [103, 104], "regress": [103, 104], "huggingfac": [103, 104, 107], "pyt_gen_token": [103, 104], "gpt2_ep": 103, "max_seq_len": [103, 104], "trt_gen_token": [103, 104], "skip_special_token": [103, 104], "torch_export_gpt2": 103, "llama_path": 104, "llama": 104, "7b": 104, "chat": 104, "hf": 104, "llama2_ep": 104, "batch_decod": 104, "clean_up_tokenization_spac": 104, "torch_export_llama2": 104, "argpars": 105, "modelopt": 105, "mtq": 105, "export_torch_mod": 105, "layer_spec": 105, "num_class": 105, "1000": [105, 108], "init_weight": 105, "in_channel": 105, "pool": [105, 113], "maxpool2d": 105, "batchnorm2d": 105, "sequenti": 105, "avgpool": 105, "adaptiveavgpool2d": 105, "4096": 105, "dropout": 105, "_initialize_weight": 105, "kaiming_normal_": 105, "fan_out": 105, "nonlinear": 105, "constant_": 105, "elif": 105, "normal_": 105, "vgg16_cfg": 105, "128": 105, "argumentpars": 105, "add_argu": 105, "ckpt": 105, "parse_arg": 105, "model_state_dict": 105, "device_count": 105, "ordereddict": 105, "new_state_dict": 105, "forget": 105, "training_dataset": 105, "randomcrop": 105, "randomhorizontalflip": 105, "training_dataload": 105, "drop_last": 105, "crit": 105, "crossentropyloss": 105, "calibrate_loop": 105, "pred": 105, "5f": 105, "acc": 105, "2f": 105, "quantize_typ": 105, "quant_cfg": 105, "int8_default_cfg": 105, "fp8_default_cfg": 105, "forward_loop": 105, "qdq": 105, "incomplet": 105, "functionaltensor": 105, "functionaltensormod": 105, "_trace": 105, "_export": 105, "float8_e4m3fn": 105, "class_prob": 105, "class_pr": 105, "test_prob": 105, "test_pr": 105, "test_loss": 105, "test_acc": 105, "vgg16_ptq": 105, "concept": 106, "_rendered_examples_python": 106, "_rendered_examples_jupyt": 106, "acoust": 107, "speech": 107, "quartznet": 107, "contextnet": 107, "subword": 107, "piec": 107, "excit": 107, "se": 107, "smaller": 107, "audio": 107, "transcrib": 107, "speedup": 107, "obtain": [107, 111], "feedforward": 107, "cnn": 107, "uniformli": 107, "resolut": 107, "highli": [107, 108], "compound": 107, "coeffici": 107, "b0": 107, "corpu": 107, "english": 107, "supervis": 107, "walkthrough": 107, "overal": 107, "jetson": 107, "adopt": 107, "mobilenetv2": 107, "classif": 107, "imagenet": 107, "imagenett": 107, "qat": 107, "simul": 107, "hand": 108, "consider": 108, "concurr": 108, "grpc": 108, "solv": 108, "aforement": 108, "familiar": 108, "resnet50": 108, "torchhub": 108, "docker": 108, "login": 108, "xx": 108, "yy": 108, "mm": 108, "publish": 108, "22": 108, "pwd": 108, "scratch_spac": 108, "nvcr": 108, "py3": 108, "proce": 108, "hub": 108, "_validate_not_a_forked_repo": 108, "suggest": 108, "simplest": 108, "model_repositori": 108, "pbtxt": 108, "pytorch_libtorch": 108, "input__0": 108, "data_typ": 108, "type_fp32": 108, "output__0": 108, "exact": 108, "encourag": 108, "rm": 108, "8000": 108, "8001": 108, "8002": 108, "the_model_repositori": 108, "tritonserv": 108, "spin": 108, "proceed": 108, "flesh": 108, "wget": 108, "img1": 108, "hakaimagazin": 108, "wp": 108, "gulf": 108, "bird": 108, "attrdict": 108, "pyindex": 108, "tritoncli": 108, "jump": 108, "firstli": 108, "resiz": 108, "pil": 108, "httpclient": 108, "triton_to_np_dtyp": 108, "rn50_preprocess": 108, "img_path": 108, "img": 108, "centercrop": 108, "485": 108, "456": 108, "406": 108, "229": 108, "transformed_img": 108, "inferenceservercli": 108, "localhost": 108, "secondli": 108, "inferinput": 108, "set_data_from_numpi": 108, "binary_data": 108, "inferrequestedoutput": 108, "class_count": 108, "lastli": 108, "send": 108, "model_nam": 108, "inference_output": 108, "as_numpi": 108, "468750": 108, "90": 108, "523438": 108, "92": 108, "664062": 108, "429688": 108, "136": 108, "234375": 108, "confidence_scor": 108, "classification_index": 108, "eagerli": 109, "swap": 109, "exactli": 109, "_tracer": 109, "sometim": 109, "queri": 109, "attn_weight": 109, "seq_len": 109, "compiler_dynamic_shap": 109, "inputs_bs2": 109, "libtorchtrt_runtim": 110, "dl_open": 110, "ld_preload": 110, "load_librari": 110, "cxx11": 110, "abi": 110, "wl": 110, "ltorchtrt": 110, "torchtrt_runtime_exampl": 110, "libtorchtrt_plugin": 110, "neglig": 110, "thread": 110, "alert": 110, "switch": 110, "mismatch": 110, "crash": 110, "sacrif": 110, "incur": 110, "intens": 110, "trt_ep": 111, "stai": 111, "trt_t": 111, "ergonom": 112, "deleg": 112, "believ": 112, "amen": 112, "artifact": 112, "pack": 112, "year": 112, "superset": 112, "codebas": 112, "immedi": 112, "traceabl": 112, "scriptabl": 112, "hardwar": 113, "neural": 113, "deconvolut": 113, "scripted_model": 113}, "objects": {"": [[5, 0, 1, "c.STR", "STR"], [9, 0, 1, "c.TORCHTRT_API", "TORCHTRT_API"], [11, 0, 1, "c.TORCHTRT_HIDDEN", "TORCHTRT_HIDDEN"], [7, 0, 1, "c.TORCH_TENSORRT_MAJOR_VERSION", "TORCH_TENSORRT_MAJOR_VERSION"], [8, 0, 1, "c.TORCH_TENSORRT_MINOR_VERSION", "TORCH_TENSORRT_MINOR_VERSION"], [6, 0, 1, "c.TORCH_TENSORRT_PATCH_VERSION", "TORCH_TENSORRT_PATCH_VERSION"], [12, 0, 1, "c.TORCH_TENSORRT_VERSION", "TORCH_TENSORRT_VERSION"], [10, 0, 1, "c.XSTR", "XSTR"], [0, 1, 1, "_CPPv4N14torch_tensorrt8DataTypeE", "torch_tensorrt::DataType"], [0, 2, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeE5Value", "torch_tensorrt::DataType::DataType"], [0, 2, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeEN3c1010ScalarTypeE", "torch_tensorrt::DataType::DataType"], [0, 2, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeEv", "torch_tensorrt::DataType::DataType"], [0, 3, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeE5Value", "torch_tensorrt::DataType::DataType::t"], [0, 3, 1, "_CPPv4N14torch_tensorrt8DataType8DataTypeEN3c1010ScalarTypeE", "torch_tensorrt::DataType::DataType::t"], [0, 4, 1, "_CPPv4N14torch_tensorrt8DataType5ValueE", "torch_tensorrt::DataType::Value"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kBoolE", "torch_tensorrt::DataType::Value::kBool"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kCharE", "torch_tensorrt::DataType::Value::kChar"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value7kDoubleE", "torch_tensorrt::DataType::Value::kDouble"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value6kFloatE", "torch_tensorrt::DataType::Value::kFloat"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kHalfE", "torch_tensorrt::DataType::Value::kHalf"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value4kIntE", "torch_tensorrt::DataType::Value::kInt"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kLongE", "torch_tensorrt::DataType::Value::kLong"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value8kUnknownE", "torch_tensorrt::DataType::Value::kUnknown"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kBoolE", "torch_tensorrt::DataType::kBool"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kCharE", "torch_tensorrt::DataType::kChar"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value7kDoubleE", "torch_tensorrt::DataType::kDouble"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value6kFloatE", "torch_tensorrt::DataType::kFloat"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kHalfE", "torch_tensorrt::DataType::kHalf"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value4kIntE", "torch_tensorrt::DataType::kInt"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value5kLongE", "torch_tensorrt::DataType::kLong"], [0, 5, 1, "_CPPv4N14torch_tensorrt8DataType5Value8kUnknownE", "torch_tensorrt::DataType::kUnknown"], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypecv5ValueEv", "torch_tensorrt::DataType::operator Value"], [0, 2, 1, "_CPPv4N14torch_tensorrt8DataTypecvbEv", "torch_tensorrt::DataType::operator bool"], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypeneE8DataType", "torch_tensorrt::DataType::operator!="], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypeneEN8DataType5ValueE", "torch_tensorrt::DataType::operator!="], [0, 3, 1, "_CPPv4NK14torch_tensorrt8DataTypeneE8DataType", "torch_tensorrt::DataType::operator!=::other"], [0, 3, 1, "_CPPv4NK14torch_tensorrt8DataTypeneEN8DataType5ValueE", "torch_tensorrt::DataType::operator!=::other"], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypeeqE8DataType", "torch_tensorrt::DataType::operator=="], [0, 2, 1, "_CPPv4NK14torch_tensorrt8DataTypeeqEN8DataType5ValueE", "torch_tensorrt::DataType::operator=="], [0, 3, 1, "_CPPv4NK14torch_tensorrt8DataTypeeqE8DataType", "torch_tensorrt::DataType::operator==::other"], [0, 3, 1, "_CPPv4NK14torch_tensorrt8DataTypeeqEN8DataType5ValueE", "torch_tensorrt::DataType::operator==::other"], [46, 1, 1, "_CPPv4N14torch_tensorrt6DeviceE", "torch_tensorrt::Device"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device6DeviceEv", "torch_tensorrt::Device::Device"], [1, 1, 1, "_CPPv4N14torch_tensorrt6Device10DeviceTypeE", "torch_tensorrt::Device::DeviceType"], [46, 1, 1, "_CPPv4N14torch_tensorrt6Device10DeviceTypeE", "torch_tensorrt::Device::DeviceType"], [1, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeE5Value", "torch_tensorrt::Device::DeviceType::DeviceType"], [1, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEN3c1010DeviceTypeE", "torch_tensorrt::Device::DeviceType::DeviceType"], [1, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEv", "torch_tensorrt::Device::DeviceType::DeviceType"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeE5Value", "torch_tensorrt::Device::DeviceType::DeviceType"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEN3c1010DeviceTypeE", "torch_tensorrt::Device::DeviceType::DeviceType"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEv", "torch_tensorrt::Device::DeviceType::DeviceType"], [1, 3, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeE5Value", "torch_tensorrt::Device::DeviceType::DeviceType::t"], [1, 3, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEN3c1010DeviceTypeE", "torch_tensorrt::Device::DeviceType::DeviceType::t"], [46, 3, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeE5Value", "torch_tensorrt::Device::DeviceType::DeviceType::t"], [46, 3, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType10DeviceTypeEN3c1010DeviceTypeE", "torch_tensorrt::Device::DeviceType::DeviceType::t"], [1, 4, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5ValueE", "torch_tensorrt::Device::DeviceType::Value"], [46, 4, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5ValueE", "torch_tensorrt::Device::DeviceType::Value"], [1, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kDLAE", "torch_tensorrt::Device::DeviceType::Value::kDLA"], [46, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kDLAE", "torch_tensorrt::Device::DeviceType::Value::kDLA"], [1, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kGPUE", "torch_tensorrt::Device::DeviceType::Value::kGPU"], [46, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kGPUE", "torch_tensorrt::Device::DeviceType::Value::kGPU"], [1, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kDLAE", "torch_tensorrt::Device::DeviceType::kDLA"], [1, 5, 1, "_CPPv4N14torch_tensorrt6Device10DeviceType5Value4kGPUE", "torch_tensorrt::Device::DeviceType::kGPU"], [1, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypecv5ValueEv", "torch_tensorrt::Device::DeviceType::operator Value"], [46, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypecv5ValueEv", "torch_tensorrt::Device::DeviceType::operator Value"], [1, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceTypecvbEv", "torch_tensorrt::Device::DeviceType::operator bool"], [46, 2, 1, "_CPPv4N14torch_tensorrt6Device10DeviceTypecvbEv", "torch_tensorrt::Device::DeviceType::operator bool"], [1, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeneE10DeviceType", "torch_tensorrt::Device::DeviceType::operator!="], [46, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeneE10DeviceType", "torch_tensorrt::Device::DeviceType::operator!="], [1, 3, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeneE10DeviceType", "torch_tensorrt::Device::DeviceType::operator!=::other"], [46, 3, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeneE10DeviceType", "torch_tensorrt::Device::DeviceType::operator!=::other"], [1, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeeqE10DeviceType", "torch_tensorrt::Device::DeviceType::operator=="], [46, 2, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeeqE10DeviceType", "torch_tensorrt::Device::DeviceType::operator=="], [1, 3, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeeqE10DeviceType", "torch_tensorrt::Device::DeviceType::operator==::other"], [46, 3, 1, "_CPPv4NK14torch_tensorrt6Device10DeviceTypeeqE10DeviceType", "torch_tensorrt::Device::DeviceType::operator==::other"], [46, 6, 1, "_CPPv4N14torch_tensorrt6Device18allow_gpu_fallbackE", "torch_tensorrt::Device::allow_gpu_fallback"], [46, 6, 1, "_CPPv4N14torch_tensorrt6Device11device_typeE", "torch_tensorrt::Device::device_type"], [46, 6, 1, "_CPPv4N14torch_tensorrt6Device8dla_coreE", "torch_tensorrt::Device::dla_core"], [46, 6, 1, "_CPPv4N14torch_tensorrt6Device6gpu_idE", "torch_tensorrt::Device::gpu_id"], [17, 4, 1, "_CPPv4N14torch_tensorrt16EngineCapabilityE", "torch_tensorrt::EngineCapability"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability15kDLA_STANDALONEE", "torch_tensorrt::EngineCapability::kDLA_STANDALONE"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability7kSAFETYE", "torch_tensorrt::EngineCapability::kSAFETY"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability9kSTANDARDE", "torch_tensorrt::EngineCapability::kSTANDARD"], [47, 1, 1, "_CPPv4N14torch_tensorrt11GraphInputsE", "torch_tensorrt::GraphInputs"], [47, 6, 1, "_CPPv4N14torch_tensorrt11GraphInputs15input_signatureE", "torch_tensorrt::GraphInputs::input_signature"], [47, 6, 1, "_CPPv4N14torch_tensorrt11GraphInputs6inputsE", "torch_tensorrt::GraphInputs::inputs"], [48, 1, 1, "_CPPv4N14torch_tensorrt5InputE", "torch_tensorrt::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN2at6TensorE", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input"], [48, 2, 1, "_CPPv4N14torch_tensorrt5Input5InputEv", "torch_tensorrt::Input::Input"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::dtype"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::format"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::max_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::min_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::opt_shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataType12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::shape"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN2at6TensorE", "torch_tensorrt::Input::Input::tensor"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputEN3c108ArrayRefI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEE8DataTypeNSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 3, 1, "_CPPv4N14torch_tensorrt5Input5InputENSt6vectorI7int64_tEENSt6vectorIdEE12TensorFormat", "torch_tensorrt::Input::Input::tensor_domain"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input5dtypeE", "torch_tensorrt::Input::dtype"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input6formatE", "torch_tensorrt::Input::format"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input9max_shapeE", "torch_tensorrt::Input::max_shape"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input9min_shapeE", "torch_tensorrt::Input::min_shape"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input9opt_shapeE", "torch_tensorrt::Input::opt_shape"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input5shapeE", "torch_tensorrt::Input::shape"], [48, 6, 1, "_CPPv4N14torch_tensorrt5Input13tensor_domainE", "torch_tensorrt::Input::tensor_domain"], [2, 1, 1, "_CPPv4N14torch_tensorrt12TensorFormatE", "torch_tensorrt::TensorFormat"], [2, 2, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatE5Value", "torch_tensorrt::TensorFormat::TensorFormat"], [2, 2, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatEN2at12MemoryFormatE", "torch_tensorrt::TensorFormat::TensorFormat"], [2, 2, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatEv", "torch_tensorrt::TensorFormat::TensorFormat"], [2, 3, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatE5Value", "torch_tensorrt::TensorFormat::TensorFormat::t"], [2, 3, 1, "_CPPv4N14torch_tensorrt12TensorFormat12TensorFormatEN2at12MemoryFormatE", "torch_tensorrt::TensorFormat::TensorFormat::t"], [2, 4, 1, "_CPPv4N14torch_tensorrt12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::Value"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value13kChannelsLastE", "torch_tensorrt::TensorFormat::Value::kChannelsLast"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value11kContiguousE", "torch_tensorrt::TensorFormat::Value::kContiguous"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value8kUnknownE", "torch_tensorrt::TensorFormat::Value::kUnknown"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value13kChannelsLastE", "torch_tensorrt::TensorFormat::kChannelsLast"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value11kContiguousE", "torch_tensorrt::TensorFormat::kContiguous"], [2, 5, 1, "_CPPv4N14torch_tensorrt12TensorFormat5Value8kUnknownE", "torch_tensorrt::TensorFormat::kUnknown"], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormatcv5ValueEv", "torch_tensorrt::TensorFormat::operator Value"], [2, 2, 1, "_CPPv4N14torch_tensorrt12TensorFormatcvbEv", "torch_tensorrt::TensorFormat::operator bool"], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormatneE12TensorFormat", "torch_tensorrt::TensorFormat::operator!="], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormatneEN12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::operator!="], [2, 3, 1, "_CPPv4NK14torch_tensorrt12TensorFormatneE12TensorFormat", "torch_tensorrt::TensorFormat::operator!=::other"], [2, 3, 1, "_CPPv4NK14torch_tensorrt12TensorFormatneEN12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::operator!=::other"], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormateqE12TensorFormat", "torch_tensorrt::TensorFormat::operator=="], [2, 2, 1, "_CPPv4NK14torch_tensorrt12TensorFormateqEN12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::operator=="], [2, 3, 1, "_CPPv4NK14torch_tensorrt12TensorFormateqE12TensorFormat", "torch_tensorrt::TensorFormat::operator==::other"], [2, 3, 1, "_CPPv4NK14torch_tensorrt12TensorFormateqEN12TensorFormat5ValueE", "torch_tensorrt::TensorFormat::operator==::other"], [37, 2, 1, "_CPPv4N14torch_tensorrt15dump_build_infoEv", "torch_tensorrt::dump_build_info"], [35, 2, 1, "_CPPv4N14torch_tensorrt14get_build_infoEv", "torch_tensorrt::get_build_info"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability15kDLA_STANDALONEE", "torch_tensorrt::kDLA_STANDALONE"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability7kSAFETYE", "torch_tensorrt::kSAFETY"], [17, 5, 1, "_CPPv4N14torch_tensorrt16EngineCapability9kSTANDARDE", "torch_tensorrt::kSTANDARD"], [16, 4, 1, "_CPPv4N14torch_tensorrt7logging5LevelE", "torch_tensorrt::logging::Level"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kDEBUGE", "torch_tensorrt::logging::Level::kDEBUG"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kERRORE", "torch_tensorrt::logging::Level::kERROR"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kGRAPHE", "torch_tensorrt::logging::Level::kGRAPH"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level5kINFOE", "torch_tensorrt::logging::Level::kINFO"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level15kINTERNAL_ERRORE", "torch_tensorrt::logging::Level::kINTERNAL_ERROR"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level8kWARNINGE", "torch_tensorrt::logging::Level::kWARNING"], [24, 2, 1, "_CPPv4N14torch_tensorrt7logging24get_is_colored_output_onEv", "torch_tensorrt::logging::get_is_colored_output_on"], [22, 2, 1, "_CPPv4N14torch_tensorrt7logging18get_logging_prefixEv", "torch_tensorrt::logging::get_logging_prefix"], [23, 2, 1, "_CPPv4N14torch_tensorrt7logging24get_reportable_log_levelEv", "torch_tensorrt::logging::get_reportable_log_level"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kDEBUGE", "torch_tensorrt::logging::kDEBUG"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kERRORE", "torch_tensorrt::logging::kERROR"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level6kGRAPHE", "torch_tensorrt::logging::kGRAPH"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level5kINFOE", "torch_tensorrt::logging::kINFO"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level15kINTERNAL_ERRORE", "torch_tensorrt::logging::kINTERNAL_ERROR"], [16, 5, 1, "_CPPv4N14torch_tensorrt7logging5Level8kWARNINGE", "torch_tensorrt::logging::kWARNING"], [26, 2, 1, "_CPPv4N14torch_tensorrt7logging3logE5LevelNSt6stringE", "torch_tensorrt::logging::log"], [26, 3, 1, "_CPPv4N14torch_tensorrt7logging3logE5LevelNSt6stringE", "torch_tensorrt::logging::log::lvl"], [26, 3, 1, "_CPPv4N14torch_tensorrt7logging3logE5LevelNSt6stringE", "torch_tensorrt::logging::log::msg"], [27, 2, 1, "_CPPv4N14torch_tensorrt7logging24set_is_colored_output_onEb", "torch_tensorrt::logging::set_is_colored_output_on"], [27, 3, 1, "_CPPv4N14torch_tensorrt7logging24set_is_colored_output_onEb", "torch_tensorrt::logging::set_is_colored_output_on::colored_output_on"], [28, 2, 1, "_CPPv4N14torch_tensorrt7logging18set_logging_prefixENSt6stringE", "torch_tensorrt::logging::set_logging_prefix"], [28, 3, 1, "_CPPv4N14torch_tensorrt7logging18set_logging_prefixENSt6stringE", "torch_tensorrt::logging::set_logging_prefix::prefix"], [25, 2, 1, "_CPPv4N14torch_tensorrt7logging24set_reportable_log_levelE5Level", "torch_tensorrt::logging::set_reportable_log_level"], [25, 3, 1, "_CPPv4N14torch_tensorrt7logging24set_reportable_log_levelE5Level", "torch_tensorrt::logging::set_reportable_log_level::lvl"], [3, 1, 1, "_CPPv4I0EN14torch_tensorrt3ptq19Int8CacheCalibratorE", "torch_tensorrt::ptq::Int8CacheCalibrator"], [3, 7, 1, "_CPPv4I0EN14torch_tensorrt3ptq19Int8CacheCalibratorE", "torch_tensorrt::ptq::Int8CacheCalibrator::Algorithm"], [3, 2, 1, "_CPPv4N14torch_tensorrt3ptq19Int8CacheCalibrator19Int8CacheCalibratorERKNSt6stringE", "torch_tensorrt::ptq::Int8CacheCalibrator::Int8CacheCalibrator"], [3, 3, 1, "_CPPv4N14torch_tensorrt3ptq19Int8CacheCalibrator19Int8CacheCalibratorERKNSt6stringE", "torch_tensorrt::ptq::Int8CacheCalibrator::Int8CacheCalibrator::cache_file_path"], [3, 2, 1, "_CPPv4N14torch_tensorrt3ptq19Int8CacheCalibratorcvPN8nvinfer115IInt8CalibratorEEv", "torch_tensorrt::ptq::Int8CacheCalibrator::operator nvinfer1::IInt8Calibrator*"], [4, 1, 1, "_CPPv4I00EN14torch_tensorrt3ptq14Int8CalibratorE", "torch_tensorrt::ptq::Int8Calibrator"], [4, 7, 1, "_CPPv4I00EN14torch_tensorrt3ptq14Int8CalibratorE", "torch_tensorrt::ptq::Int8Calibrator::Algorithm"], [4, 7, 1, "_CPPv4I00EN14torch_tensorrt3ptq14Int8CalibratorE", "torch_tensorrt::ptq::Int8Calibrator::DataLoaderUniquePtr"], [4, 2, 1, "_CPPv4N14torch_tensorrt3ptq14Int8Calibrator14Int8CalibratorE19DataLoaderUniquePtrRKNSt6stringEb", "torch_tensorrt::ptq::Int8Calibrator::Int8Calibrator"], [4, 3, 1, "_CPPv4N14torch_tensorrt3ptq14Int8Calibrator14Int8CalibratorE19DataLoaderUniquePtrRKNSt6stringEb", "torch_tensorrt::ptq::Int8Calibrator::Int8Calibrator::cache_file_path"], [4, 3, 1, "_CPPv4N14torch_tensorrt3ptq14Int8Calibrator14Int8CalibratorE19DataLoaderUniquePtrRKNSt6stringEb", "torch_tensorrt::ptq::Int8Calibrator::Int8Calibrator::dataloader"], [4, 3, 1, "_CPPv4N14torch_tensorrt3ptq14Int8Calibrator14Int8CalibratorE19DataLoaderUniquePtrRKNSt6stringEb", "torch_tensorrt::ptq::Int8Calibrator::Int8Calibrator::use_cache"], [4, 2, 1, "_CPPv4N14torch_tensorrt3ptq14Int8CalibratorcvPN8nvinfer115IInt8CalibratorEEv", "torch_tensorrt::ptq::Int8Calibrator::operator nvinfer1::IInt8Calibrator*"], [29, 2, 1, "_CPPv4I0EN14torch_tensorrt3ptq26make_int8_cache_calibratorE19Int8CacheCalibratorI9AlgorithmERKNSt6stringE", "torch_tensorrt::ptq::make_int8_cache_calibrator"], [29, 7, 1, "_CPPv4I0EN14torch_tensorrt3ptq26make_int8_cache_calibratorE19Int8CacheCalibratorI9AlgorithmERKNSt6stringE", "torch_tensorrt::ptq::make_int8_cache_calibrator::Algorithm"], [29, 3, 1, "_CPPv4I0EN14torch_tensorrt3ptq26make_int8_cache_calibratorE19Int8CacheCalibratorI9AlgorithmERKNSt6stringE", "torch_tensorrt::ptq::make_int8_cache_calibrator::cache_file_path"], [30, 2, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator"], [30, 7, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::Algorithm"], [30, 7, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::DataLoader"], [30, 3, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::cache_file_path"], [30, 3, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::dataloader"], [30, 3, 1, "_CPPv4I00EN14torch_tensorrt3ptq20make_int8_calibratorE14Int8CalibratorI9Algorithm10DataLoaderE10DataLoaderRKNSt6stringEb", "torch_tensorrt::ptq::make_int8_calibrator::use_cache"], [36, 2, 1, "_CPPv4N14torch_tensorrt10set_deviceEKi", "torch_tensorrt::set_device"], [36, 3, 1, "_CPPv4N14torch_tensorrt10set_deviceEKi", "torch_tensorrt::set_device::gpu_id"], [49, 1, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpecE", "torch_tensorrt::torchscript::CompileSpec"], [49, 2, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecEN5torch3jit6IValueE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec"], [49, 2, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorI5InputEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec"], [49, 2, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorIN3c108ArrayRefI7int64_tEEEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec"], [49, 2, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorINSt6vectorI7int64_tEEEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec"], [49, 3, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorIN3c108ArrayRefI7int64_tEEEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec::fixed_sizes"], [49, 3, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorINSt6vectorI7int64_tEEEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec::fixed_sizes"], [49, 3, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecEN5torch3jit6IValueE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec::input_signature"], [49, 3, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec11CompileSpecENSt6vectorI5InputEE", "torch_tensorrt::torchscript::CompileSpec::CompileSpec::inputs"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec19allow_shape_tensorsE", "torch_tensorrt::torchscript::CompileSpec::allow_shape_tensors"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec10capabilityE", "torch_tensorrt::torchscript::CompileSpec::capability"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec5debugE", "torch_tensorrt::torchscript::CompileSpec::debug"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec6deviceE", "torch_tensorrt::torchscript::CompileSpec::device"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec12disable_tf32E", "torch_tensorrt::torchscript::CompileSpec::disable_tf32"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec20dla_global_dram_sizeE", "torch_tensorrt::torchscript::CompileSpec::dla_global_dram_size"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec19dla_local_dram_sizeE", "torch_tensorrt::torchscript::CompileSpec::dla_local_dram_size"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec13dla_sram_sizeE", "torch_tensorrt::torchscript::CompileSpec::dla_sram_size"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec18enabled_precisionsE", "torch_tensorrt::torchscript::CompileSpec::enabled_precisions"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec12graph_inputsE", "torch_tensorrt::torchscript::CompileSpec::graph_inputs"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec14min_block_sizeE", "torch_tensorrt::torchscript::CompileSpec::min_block_size"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec20num_avg_timing_itersE", "torch_tensorrt::torchscript::CompileSpec::num_avg_timing_iters"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec14ptq_calibratorE", "torch_tensorrt::torchscript::CompileSpec::ptq_calibrator"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec5refitE", "torch_tensorrt::torchscript::CompileSpec::refit"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec24require_full_compilationE", "torch_tensorrt::torchscript::CompileSpec::require_full_compilation"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec14sparse_weightsE", "torch_tensorrt::torchscript::CompileSpec::sparse_weights"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec22torch_executed_modulesE", "torch_tensorrt::torchscript::CompileSpec::torch_executed_modules"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec18torch_executed_opsE", "torch_tensorrt::torchscript::CompileSpec::torch_executed_ops"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec24truncate_long_and_doubleE", "torch_tensorrt::torchscript::CompileSpec::truncate_long_and_double"], [49, 6, 1, "_CPPv4N14torch_tensorrt11torchscript11CompileSpec14workspace_sizeE", "torch_tensorrt::torchscript::CompileSpec::workspace_size"], [31, 2, 1, "_CPPv4N14torch_tensorrt11torchscript29check_method_operator_supportERKN5torch3jit6ModuleENSt6stringE", "torch_tensorrt::torchscript::check_method_operator_support"], [31, 3, 1, "_CPPv4N14torch_tensorrt11torchscript29check_method_operator_supportERKN5torch3jit6ModuleENSt6stringE", "torch_tensorrt::torchscript::check_method_operator_support::method_name"], [31, 3, 1, "_CPPv4N14torch_tensorrt11torchscript29check_method_operator_supportERKN5torch3jit6ModuleENSt6stringE", "torch_tensorrt::torchscript::check_method_operator_support::module"], [32, 2, 1, "_CPPv4N14torch_tensorrt11torchscript7compileERKN5torch3jit6ModuleE11CompileSpec", "torch_tensorrt::torchscript::compile"], [32, 3, 1, "_CPPv4N14torch_tensorrt11torchscript7compileERKN5torch3jit6ModuleE11CompileSpec", "torch_tensorrt::torchscript::compile::info"], [32, 3, 1, "_CPPv4N14torch_tensorrt11torchscript7compileERKN5torch3jit6ModuleE11CompileSpec", "torch_tensorrt::torchscript::compile::module"], [34, 2, 1, "_CPPv4N14torch_tensorrt11torchscript28convert_method_to_trt_engineERKN5torch3jit6ModuleENSt6stringE11CompileSpec", "torch_tensorrt::torchscript::convert_method_to_trt_engine"], [34, 3, 1, "_CPPv4N14torch_tensorrt11torchscript28convert_method_to_trt_engineERKN5torch3jit6ModuleENSt6stringE11CompileSpec", "torch_tensorrt::torchscript::convert_method_to_trt_engine::info"], [34, 3, 1, "_CPPv4N14torch_tensorrt11torchscript28convert_method_to_trt_engineERKN5torch3jit6ModuleENSt6stringE11CompileSpec", "torch_tensorrt::torchscript::convert_method_to_trt_engine::method_name"], [34, 3, 1, "_CPPv4N14torch_tensorrt11torchscript28convert_method_to_trt_engineERKN5torch3jit6ModuleENSt6stringE11CompileSpec", "torch_tensorrt::torchscript::convert_method_to_trt_engine::module"], [33, 2, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module"], [33, 3, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module::device"], [33, 3, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module::engine"], [33, 3, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module::input_binding_names"], [33, 3, 1, "_CPPv4N14torch_tensorrt11torchscript26embed_engine_in_new_moduleERKNSt6stringE6DeviceRKNSt6vectorINSt6stringEEERKNSt6vectorINSt6stringEEE", "torch_tensorrt::torchscript::embed_engine_in_new_module::output_binding_names"], [75, 8, 0, "-", "torch_tensorrt"]], "torch_tensorrt": [[75, 9, 1, "", "Device"], [75, 9, 1, "", "DeviceType"], [75, 9, 1, "", "EngineCapability"], [75, 9, 1, "", "Input"], [75, 9, 1, "", "MutableTorchTensorRTModule"], [75, 12, 1, "", "compile"], [75, 12, 1, "", "convert_method_to_trt_engine"], [75, 9, 1, "", "dtype"], [111, 8, 0, "-", "dynamo"], [71, 8, 0, "-", "fx"], [75, 12, 1, "", "load"], [72, 8, 0, "-", "logging"], [75, 9, 1, "", "memory_format"], [74, 8, 0, "-", "runtime"], [75, 12, 1, "", "save"], [76, 8, 0, "-", "ts"]], "torch_tensorrt.Device": [[75, 10, 1, "", "__init__"], [75, 11, 1, "", "device_type"], [75, 11, 1, "", "dla_core"], [75, 11, 1, "", "gpu_id"]], "torch_tensorrt.DeviceType": [[75, 11, 1, "", "DLA"], [75, 11, 1, "", "GPU"], [75, 11, 1, "", "UNKNOWN"], [75, 10, 1, "", "to"], [75, 10, 1, "", "try_from"], [75, 10, 1, "", "try_to"]], "torch_tensorrt.EngineCapability": [[75, 11, 1, "", "DLA_STANDALONE"], [75, 11, 1, "", "SAFETY"], [75, 11, 1, "", "STANDARD"], [75, 10, 1, "", "to"], [75, 10, 1, "", "try_from"], [75, 10, 1, "", "try_to"]], "torch_tensorrt.Input": [[75, 10, 1, "", "__init__"], [75, 11, 1, "", "dtype"], [75, 10, 1, "", "example_tensor"], [75, 11, 1, "", "format"], [75, 10, 1, "", "from_tensor"], [75, 10, 1, "", "from_tensors"]], "torch_tensorrt.MutableTorchTensorRTModule": [[75, 10, 1, "", "__init__"], [75, 10, 1, "", "compile"], [75, 10, 1, "", "refit_gm"]], "torch_tensorrt.dtype": [[75, 11, 1, "", "b"], [75, 11, 1, "", "bf16"], [75, 11, 1, "", "f16"], [75, 11, 1, "", "f32"], [75, 11, 1, "", "f64"], [75, 11, 1, "", "f8"], [75, 11, 1, "", "i32"], [75, 11, 1, "", "i64"], [75, 11, 1, "", "i8"], [75, 10, 1, "", "to"], [75, 10, 1, "", "try_from"], [75, 10, 1, "", "try_to"], [75, 11, 1, "", "u8"], [75, 11, 1, "", "unknown"]], "torch_tensorrt.dynamo": [[70, 9, 1, "", "CompilationSettings"], [70, 12, 1, "", "compile"], [70, 12, 1, "", "export"], [70, 12, 1, "", "refit_module_weights"], [70, 12, 1, "", "trace"]], "torch_tensorrt.fx": [[71, 9, 1, "", "InputTensorSpec"], [71, 9, 1, "", "TRTInterpreter"], [71, 9, 1, "", "TRTInterpreterResult"], [71, 9, 1, "", "TRTModule"], [71, 12, 1, "", "compile"]], "torch_tensorrt.logging": [[72, 9, 1, "", "debug"], [72, 9, 1, "", "errors"], [72, 9, 1, "", "graphs"], [72, 9, 1, "", "info"], [72, 9, 1, "", "internal_errors"], [72, 9, 1, "", "warnings"]], "torch_tensorrt.memory_format": [[75, 11, 1, "", "cdhw32"], [75, 11, 1, "", "chw16"], [75, 11, 1, "", "chw2"], [75, 11, 1, "", "chw32"], [75, 11, 1, "", "chw4"], [75, 11, 1, "", "dhwc"], [75, 11, 1, "", "dhwc8"], [75, 11, 1, "", "dla_hwc4"], [75, 11, 1, "", "dla_linear"], [75, 11, 1, "", "hwc"], [75, 11, 1, "", "hwc16"], [75, 11, 1, "", "hwc8"], [75, 11, 1, "", "linear"], [75, 10, 1, "", "to"], [75, 10, 1, "", "try_from"], [75, 10, 1, "", "try_to"]], "torch_tensorrt.runtime": [[74, 9, 1, "", "PythonTorchTensorRTModule"], [74, 9, 1, "", "TorchTensorRTModule"], [74, 12, 1, "", "set_multi_device_safe_mode"]], "torch_tensorrt.runtime.PythonTorchTensorRTModule": [[74, 10, 1, "", "__init__"], [74, 10, 1, "", "cudagraphs_validate_shapes"], [74, 10, 1, "", "disable_profiling"], [74, 10, 1, "", "enable_profiling"], [74, 10, 1, "", "forward"], [74, 10, 1, "", "get_layer_info"]], "torch_tensorrt.runtime.TorchTensorRTModule": [[74, 10, 1, "", "__init__"], [74, 10, 1, "", "forward"], [74, 10, 1, "", "get_extra_state"], [74, 10, 1, "", "set_extra_state"]], "torch_tensorrt.ts": [[76, 12, 1, "", "TensorRTCompileSpec"], [76, 12, 1, "", "check_method_op_support"], [76, 12, 1, "", "compile"], [76, 12, 1, "", "convert_method_to_trt_engine"], [76, 12, 1, "", "embed_engine_in_new_module"], [73, 8, 0, "-", "ptq"]], "torch_tensorrt.ts.ptq": [[73, 9, 1, "", "CacheCalibrator"], [73, 9, 1, "", "CalibrationAlgo"], [73, 9, 1, "", "DataLoaderCalibrator"]], "torch_tensorrt.ts.ptq.CalibrationAlgo": [[73, 11, 1, "", "ENTROPY_CALIBRATION"], [73, 11, 1, "", "ENTROPY_CALIBRATION_2"], [73, 11, 1, "", "LEGACY_CALIBRATION"], [73, 11, 1, "", "MINMAX_CALIBRATION"]]}, "objtypes": {"0": "c:macro", "1": "cpp:class", "2": "cpp:function", "3": "cpp:functionParam", "4": "cpp:enum", "5": "cpp:enumerator", "6": "cpp:member", "7": "cpp:templateParam", "8": "py:module", "9": "py:class", "10": "py:method", "11": "py:attribute", "12": "py:function"}, "objnames": {"0": ["c", "macro", "C macro"], "1": ["cpp", "class", "C++ class"], "2": ["cpp", "function", "C++ function"], "3": ["cpp", "functionParam", "C++ function parameter"], "4": ["cpp", "enum", "C++ enum"], "5": ["cpp", "enumerator", "C++ enumerator"], "6": ["cpp", "member", "C++ member"], "7": ["cpp", "templateParam", "C++ template parameter"], "8": ["py", "module", "Python module"], "9": ["py", "class", "Python class"], "10": ["py", "method", "Python method"], "11": ["py", "attribute", "Python attribute"], "12": ["py", "function", "Python function"]}, "titleterms": {"class": [0, 1, 2, 3, 4, 20, 21, 38, 40, 41, 50, 70, 71, 73, 74, 75], "datatyp": 0, "document": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 16, 17, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 46, 47, 48, 49, 61, 68, 84, 85], "devic": [1, 46, 110], "devicetyp": 1, "nest": [1, 46], "relationship": [1, 3, 4, 46, 48], "tensorformat": 2, "templat": [3, 4, 29, 30], "int8cachecalibr": 3, "inherit": [3, 4, 48], "base": [3, 4, 48, 79], "type": [3, 4, 46, 48, 54], "int8calibr": 4, "defin": [5, 6, 7, 8, 9, 10, 11, 12, 19, 50, 105], "str": 5, "torch_tensorrt_patch_vers": 6, "torch_tensorrt_major_vers": 7, "torch_tensorrt_minor_vers": 8, "torchtrt_api": 9, "xstr": 10, "torchtrt_hidden": 11, "torch_tensorrt_vers": 12, "directori": [13, 14, 15, 51], "cpp": [13, 18, 19, 20, 21, 56], "subdirectori": [13, 14], "includ": [14, 18, 19, 20, 21], "torch_tensorrt": [15, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 45, 70, 71, 72, 73, 74, 75, 76, 99, 101, 102, 112], "file": [15, 18, 19, 20, 21, 42, 43, 44, 45, 50, 51], "enum": [16, 17, 38, 39, 50, 73, 75], "level": [16, 79, 81, 82], "enginecap": 17, "log": [18, 22, 23, 24, 25, 26, 27, 28, 39, 42, 72], "h": [18, 19, 20, 21, 42, 43, 44, 45, 56], "content": [18, 19, 20, 21, 38, 39, 40, 41, 79, 80, 81, 82, 83, 84], "definit": [18, 19, 20, 21, 82, 97, 98, 99, 100, 101, 102, 103, 104, 105], "By": [18, 19], "namespac": [18, 19, 20, 21, 38, 39, 40, 41, 50], "macro": [19, 43], "ptq": [20, 29, 30, 40, 44, 73, 90, 105], "function": [22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 50, 61, 70, 71, 74, 75, 76, 105], "get_logging_prefix": 22, "get_reportable_log_level": 23, "get_is_colored_output_on": 24, "set_reportable_log_level": 25, "set_is_colored_output_on": 27, "set_logging_prefix": 28, "make_int8_cache_calibr": 29, "make_int8_calibr": 30, "torchscript": [31, 32, 33, 34, 41, 60, 66, 68, 87, 88, 91, 111, 112], "check_method_operator_support": 31, "compil": [32, 57, 59, 63, 64, 66, 67, 68, 88, 94, 95, 97, 98, 99, 100, 101, 102, 103, 104, 106, 107, 109, 111, 112], "embed_engine_in_new_modul": 33, "convert_method_to_trt_engin": 34, "get_build_info": 35, "set_devic": 36, "dump_build_info": 37, "program": [42, 43, 44, 45, 63, 97, 110], "list": [42, 43, 44, 45, 82], "struct": [46, 47, 48, 49, 50], "graphinput": 47, "input": [48, 99, 101], "compilespec": 49, "torch": [50, 61, 63, 64, 65, 66, 67, 68, 88, 89, 91, 92, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112], "tensorrt": [50, 58, 61, 63, 64, 65, 66, 68, 88, 89, 91, 92, 96, 97, 99, 101, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112], "c": [50, 61, 66, 67, 68, 88, 90, 107], "api": [50, 51, 61, 66, 68], "hierarchi": 50, "full": [50, 51], "torchtrtc": [52, 88], "convers": [53, 57, 59, 60], "phase": [53, 55, 56, 57, 58, 59], "node": 53, "evalu": [53, 54, 69], "convert": [53, 54, 60, 65, 69, 88], "write": [54, 60, 62, 92], "dynamo": [54, 62, 68, 70, 95, 103, 104, 106, 111, 112], "implement": 54, "registr": 54, "capabl": 54, "valid": 54, "contract": [54, 60], "exampl": [54, 62, 81, 83, 107], "convolut": 54, "oper": [54, 64, 69, 88, 92], "decomposit": 54, "addmm": [54, 55], "lower": [55, 57, 59, 62], "pass": [55, 62], "us": [55, 61, 88, 89, 91, 92, 98, 99, 101, 102, 103, 104, 105, 107, 109], "eliminatecommonsubexpress": 55, "elimin": 55, "dead": 55, "code": [55, 68, 81], "except": 55, "Or": 55, "pattern": 55, "redund": 55, "guard": 55, "freez": 55, "modul": [55, 87, 88, 96, 112], "fuse": 55, "branch": 55, "linear": 55, "flatten": 55, "graph": [55, 58, 112], "tupl": 55, "fallback": [55, 56], "peephol": 55, "optim": [55, 67, 108], "remov": 55, "contigu": 55, "dropout": 55, "To": 55, "unpack": 55, "logsoftmax": 55, "unrol": 55, "loop": [55, 105], "replac": [55, 81], "tile": 55, "repeat": 55, "partit": [56, 57, 59], "partitoninfo": 56, "segmentedblock": 56, "shape_analysi": 56, "automat": 56, "depend": [56, 66], "awar": [56, 107], "runtim": [57, 58, 59, 74, 110], "background": [58, 60], "engin": [58, 65, 92, 93, 94], "executor": 58, "op": [58, 65, 92], "construct": 58, "result": 58, "serial": [58, 64, 67], "deseri": 58, "abi": [58, 66], "version": [58, 66], "format": [58, 112], "system": [59, 66], "overview": 59, "what": 60, "guarante": 60, "respons": 60, "context": [60, 79], "arg": [60, 80], "weight": [60, 97, 105], "other": 60, "advic": 60, "link": [61, 81], "develop": 61, "avail": 61, "layer": 61, "expect": 61, "dimens": 61, "python": [61, 66, 67, 68, 87, 89, 90], "sometim": 61, "easier": 61, "read": 61, "pytorch": [61, 65, 68, 91, 92, 103, 104, 107], "native_op": 61, "ir": [61, 111, 112], "aten": 62, "basic": 62, "requir": 62, "regist": [62, 88], "export": [63, 67, 102, 109], "customiz": [63, 64], "set": [63, 64, 96, 98, 102, 108], "under": [63, 88, 109], "hood": [63, 88, 109], "trace": 63, "backend": [64, 99, 103, 104], "kei": 64, "featur": 64, "custom": [64, 88, 92, 94, 98, 109], "usag": [64, 97, 98], "after": 64, "model": [64, 65, 92, 97, 98, 99, 100, 101, 102, 103, 104, 105, 107, 108, 111], "perform": 64, "coverag": 64, "feasibl": 64, "dynam": [64, 99, 107, 109], "shape": [64, 99, 107, 109], "support": [64, 69], "recompil": [64, 99], "condit": 64, "fx": [65, 68, 71, 107, 112], "frontend": [65, 66, 68, 91, 107, 112], "user": [65, 68], "guid": [65, 68], "acc": 65, "tracer": 65, "fx2trt": 65, "how": [65, 79, 90], "add": 65, "miss": 65, "instal": [66, 86], "precompil": 66, "binari": 66, "specif": 66, "cuda": [66, 98, 101], "nightli": 66, "build": [66, 79, 108], "onli": 66, "from": [66, 91], "sourc": 66, "linux": 66, "packag": [66, 110], "addit": 66, "option": [66, 67, 79, 80, 82, 99, 101, 112], "distribut": 66, "No": 66, "librari": [66, 110], "standalon": 66, "releas": 66, "debug": 66, "pre": [66, 105], "cxx11": 66, "choos": 66, "right": 66, "window": 66, "step": [66, 67, 108], "advanc": [66, 97, 98], "setup": 66, "troubleshoot": 66, "altern": 66, "cmake": 66, "nativ": 66, "aarch64": 66, "jetson": 66, "prerequisit": 66, "environ": 66, "cli": [66, 68], "quick": 67, "start": [67, 68], "1": [67, 83, 108], "2": [67, 83, 84, 108], "deploi": [67, 105, 107, 110], "deploy": 67, "In": [68, 97], "framework": 68, "infer": [68, 99, 100, 101, 102, 105, 108], "nvidia": 68, "gpu": 68, "get": 68, "tutori": [68, 106], "contributor": 68, "indic": 68, "legaci": [68, 112], "further": 68, "inform": 68, "current": 69, "through": 69, "ts": [73, 76, 112], "submodul": 75, "comput": 77, "time": [77, 112], "changelog": 78, "configur": 79, "project": 79, "wide": 79, "html": 79, "theme": [79, 85], "toc": 79, "page": 79, "tabl": [79, 80, 81, 82, 83, 84], "mod": 80, "test_py_modul": 80, "gener": [80, 103, 104], "index": 80, "paramet": 80, "data": 80, "paragraph": [81, 84], "markup": 81, "inlin": 81, "math": 81, "meta": 81, "block": 81, "liter": 81, "line": 81, "quot": 81, "doctest": 81, "emphas": 81, "number": [81, 82], "sidebar": 81, "ch": 81, "ien": 81, "The": [81, 88, 103, 104], "creativ": 81, "A": 81, "refer": 81, "footnot": 81, "citat": [81, 90], "glossari": 81, "target": 81, "direct": 81, "center": 81, "text": [81, 103, 104], "imag": [81, 82], "figur": 81, "admonit": 81, "And": 81, "wai": 81, "topic": 81, "rubric": 81, "titl": 81, "compound": 81, "download": [81, 86], "enumer": 82, "field": 82, "bullet": 82, "second": 82, "But": 82, "deeper": 82, "down": 82, "rabbit": 82, "hole": 82, "hlist": 82, "grid": 82, "giant": 82, "can": 82, "have": 82, "caption": [82, 85], "like": [82, 103, 104], "thi": [82, 85], "one": 82, "long": [83, 85], "sticki": 83, "nav": 83, "menu": [83, 85], "3": [83, 108], "4": 83, "5": 83, "6": 83, "7": 83, "8": 83, "9": 83, "10": 83, "11": 83, "12": 83, "13": 83, "14": 83, "15": 83, "16": 83, "17": 83, "18": 83, "19": 83, "20": 83, "submenu": 83, "subsubmenu": 83, "structur": 84, "element": 84, "section": 84, "subsect": 84, "subsubsect": 84, "demo": 85, "an": 85, "incred": 85, "via": 86, "git": 86, "creat": [87, 90], "work": [87, 88], "save": [87, 96, 111], "disk": 87, "quickstart": 88, "unsupport": 88, "post": 90, "train": [90, 105, 107], "quantiz": [90, 105, 107], "your": [90, 108], "own": 90, "applic": 90, "directli": 91, "kernel": 92, "within": 92, "test": 92, "our": 92, "wrap": 92, "insert": 92, "cach": [93, 94, 97], "bert": [93, 107], "jit": [94, 109], "aot": [94, 109], "mutabl": 96, "initi": 96, "make": [96, 97], "modif": 96, "stabl": [96, 100], "diffus": [96, 100], "huggingfac": 96, "refit": 97, "new": 97, "standard": 97, "workflow": 97, "import": [97, 98, 99, 100, 101, 102, 103, 104, 105], "pretrain": 97, "map": 97, "place": 97, "default": [98, 102], "cleanup": [98, 101], "driver": [98, 101], "error": [98, 101], "note": [98, 101], "resnet": 99, "argument": [99, 101], "avoid": 99, "specifi": 99, "befor": 99, "trt": 99, "transform": [101, 107], "cudagraph": [102, 110], "integr": 102, "gpt2": 103, "output": [103, 104], "decod": [103, 104], "sentenc": [103, 104], "should": [103, 104], "look": [103, 104], "i": [103, 104], "enjoi": [103, 104], "walk": [103, 104], "my": [103, 104], "cute": [103, 104], "dog": [103, 104], "m": [103, 104], "sure": [103, 104], "ll": [103, 104], "ever": [103, 104], "abl": [103, 104], "llama2": 104, "load": [105, 111], "dataset": 105, "loss": 105, "calibr": 105, "tune": 105, "fp8": 105, "notebook": 107, "citrinet": 107, "efficientnet": 107, "mask": 107, "languag": 107, "mlm": 107, "hug": 107, "face": 107, "acceler": 107, "serv": [107, 108], "resnet50": 107, "lenet": 107, "deep": 107, "learn": 107, "object": 107, "detect": 107, "ssd": 107, "int8": 107, "triton": 108, "up": 108, "server": 108, "client": 108, "queri": 108, "constraint": 109, "libtorchtrt": 110, "so": 110, "plugin": 110, "multi": 110, "safe": 110, "mode": 110, "exportedprogram": 111, "b": 111, "explain": 112, "just": 112, "accept": 112, "return": 112, "ahead": 112, "dla": 113}, "envversion": {"sphinx.domains.c": 2, "sphinx.domains.changeset": 1, "sphinx.domains.citation": 1, "sphinx.domains.cpp": 6, "sphinx.domains.index": 1, "sphinx.domains.javascript": 2, "sphinx.domains.math": 2, "sphinx.domains.python": 3, "sphinx.domains.rst": 2, "sphinx.domains.std": 2, "nbsphinx": 4, "sphinx.ext.intersphinx": 1, "sphinx.ext.todo": 2, "sphinx.ext.viewcode": 1, "sphinx": 56}}) \ No newline at end of file diff --git a/docs/sg_execution_times.html b/docs/sg_execution_times.html index 851ee0f54e..fcd5e37e95 100644 --- a/docs/sg_execution_times.html +++ b/docs/sg_execution_times.html @@ -10,7 +10,7 @@ - Computation times — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Computation times — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/src/pytorch-sphinx-theme/docs/changelog.html b/docs/src/pytorch-sphinx-theme/docs/changelog.html index 8a07be39e5..e646fcebe4 100644 --- a/docs/src/pytorch-sphinx-theme/docs/changelog.html +++ b/docs/src/pytorch-sphinx-theme/docs/changelog.html @@ -10,7 +10,7 @@ - Changelog — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Changelog — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/src/pytorch-sphinx-theme/docs/configuring.html b/docs/src/pytorch-sphinx-theme/docs/configuring.html index 8976d9d912..efa25349a2 100644 --- a/docs/src/pytorch-sphinx-theme/docs/configuring.html +++ b/docs/src/pytorch-sphinx-theme/docs/configuring.html @@ -10,7 +10,7 @@ - Configuration — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Configuration — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/src/pytorch-sphinx-theme/docs/demo/api.html b/docs/src/pytorch-sphinx-theme/docs/demo/api.html index 6f239da3fb..f264bd5294 100644 --- a/docs/src/pytorch-sphinx-theme/docs/demo/api.html +++ b/docs/src/pytorch-sphinx-theme/docs/demo/api.html @@ -10,7 +10,7 @@ - 5. :mod:`test_py_module` — Torch-TensorRT v2.5.0.dev0+8759736 documentation + 5. :mod:`test_py_module` — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/src/pytorch-sphinx-theme/docs/demo/demo.html b/docs/src/pytorch-sphinx-theme/docs/demo/demo.html index 527baa072a..7272488bec 100644 --- a/docs/src/pytorch-sphinx-theme/docs/demo/demo.html +++ b/docs/src/pytorch-sphinx-theme/docs/demo/demo.html @@ -12,7 +12,7 @@ - 3. Paragraph Level Markup — Torch-TensorRT v2.5.0.dev0+8759736 documentation + 3. Paragraph Level Markup — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • @@ -654,7 +654,7 @@

    3.4.4.

    3.4.5. Code Blocks

    # parsed-literal test
    -curl -O http://someurl/release-v2.5.0.dev0+8759736.tar-gz
    +curl -O http://someurl/release-v2.5.0.dev0+a4a9419.tar-gz
    Code Blocks can have captions.
    {
    diff --git a/docs/src/pytorch-sphinx-theme/docs/demo/lists_tables.html b/docs/src/pytorch-sphinx-theme/docs/demo/lists_tables.html
    index fa99dad0ff..5e1246004e 100644
    --- a/docs/src/pytorch-sphinx-theme/docs/demo/lists_tables.html
    +++ b/docs/src/pytorch-sphinx-theme/docs/demo/lists_tables.html
    @@ -10,7 +10,7 @@
     
       
       
    -  4. Lists & Tables — Torch-TensorRT v2.5.0.dev0+8759736 documentation
    +  4. Lists & Tables — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation
       
     
       
    @@ -273,7 +273,7 @@
                   
                   
                     
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/src/pytorch-sphinx-theme/docs/demo/long.html b/docs/src/pytorch-sphinx-theme/docs/demo/long.html index e472c7384d..1aae3b7da0 100644 --- a/docs/src/pytorch-sphinx-theme/docs/demo/long.html +++ b/docs/src/pytorch-sphinx-theme/docs/demo/long.html @@ -10,7 +10,7 @@ - 1. Long Sticky Nav — Torch-TensorRT v2.5.0.dev0+8759736 documentation + 1. Long Sticky Nav — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/src/pytorch-sphinx-theme/docs/demo/structure.html b/docs/src/pytorch-sphinx-theme/docs/demo/structure.html index dabdc9cfda..b6bc26b3bd 100644 --- a/docs/src/pytorch-sphinx-theme/docs/demo/structure.html +++ b/docs/src/pytorch-sphinx-theme/docs/demo/structure.html @@ -10,7 +10,7 @@ - 1. Structural Elements — Torch-TensorRT v2.5.0.dev0+8759736 documentation + 1. Structural Elements — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/src/pytorch-sphinx-theme/docs/index.html b/docs/src/pytorch-sphinx-theme/docs/index.html index 61ed7ea89f..a2b55e5edf 100644 --- a/docs/src/pytorch-sphinx-theme/docs/index.html +++ b/docs/src/pytorch-sphinx-theme/docs/index.html @@ -10,7 +10,7 @@ - <no title> — Torch-TensorRT v2.5.0.dev0+8759736 documentation + <no title> — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/src/pytorch-sphinx-theme/docs/installing.html b/docs/src/pytorch-sphinx-theme/docs/installing.html index c99e90e31e..b247fb7447 100644 --- a/docs/src/pytorch-sphinx-theme/docs/installing.html +++ b/docs/src/pytorch-sphinx-theme/docs/installing.html @@ -10,7 +10,7 @@ - Installation — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Installation — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/ts/creating_torchscript_module_in_python.html b/docs/ts/creating_torchscript_module_in_python.html index 111f15a2ce..e69108f233 100644 --- a/docs/ts/creating_torchscript_module_in_python.html +++ b/docs/ts/creating_torchscript_module_in_python.html @@ -10,7 +10,7 @@ - Creating a TorchScript Module — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Creating a TorchScript Module — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/ts/getting_started_with_cpp_api.html b/docs/ts/getting_started_with_cpp_api.html index 92f810d822..27e5cdb666 100644 --- a/docs/ts/getting_started_with_cpp_api.html +++ b/docs/ts/getting_started_with_cpp_api.html @@ -10,7 +10,7 @@ - Using Torch-TensorRT in C++ — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Using Torch-TensorRT in C++ — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/ts/getting_started_with_python_api.html b/docs/ts/getting_started_with_python_api.html index e267e4247b..f1226b66fb 100644 --- a/docs/ts/getting_started_with_python_api.html +++ b/docs/ts/getting_started_with_python_api.html @@ -10,7 +10,7 @@ - Using Torch-TensorRT in Python — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Using Torch-TensorRT in Python — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/ts/ptq.html b/docs/ts/ptq.html index f03b5af3cf..637d131cbb 100644 --- a/docs/ts/ptq.html +++ b/docs/ts/ptq.html @@ -10,7 +10,7 @@ - Post Training Quantization (PTQ) — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Post Training Quantization (PTQ) — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/ts/torchscript_frontend_from_pytorch.html b/docs/ts/torchscript_frontend_from_pytorch.html index be94a301d8..12e3b01adf 100644 --- a/docs/ts/torchscript_frontend_from_pytorch.html +++ b/docs/ts/torchscript_frontend_from_pytorch.html @@ -10,7 +10,7 @@ - Using Torch-TensorRT TorchScript Frontend Directly From PyTorch — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Using Torch-TensorRT TorchScript Frontend Directly From PyTorch — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/tutorials/_rendered_examples/dynamo/custom_kernel_plugins.html b/docs/tutorials/_rendered_examples/dynamo/custom_kernel_plugins.html index 532214d7a1..4713924e27 100644 --- a/docs/tutorials/_rendered_examples/dynamo/custom_kernel_plugins.html +++ b/docs/tutorials/_rendered_examples/dynamo/custom_kernel_plugins.html @@ -10,7 +10,7 @@ - Using Custom Kernels within TensorRT Engines with Torch-TensorRT — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Using Custom Kernels within TensorRT Engines with Torch-TensorRT — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -275,7 +275,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -316,6 +316,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -343,7 +344,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/tutorials/_rendered_examples/dynamo/engine_caching_bert_example.html b/docs/tutorials/_rendered_examples/dynamo/engine_caching_bert_example.html index b100072501..c2978b115e 100644 --- a/docs/tutorials/_rendered_examples/dynamo/engine_caching_bert_example.html +++ b/docs/tutorials/_rendered_examples/dynamo/engine_caching_bert_example.html @@ -10,7 +10,7 @@ - Engine Caching (BERT) — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Engine Caching (BERT) — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -273,7 +273,7 @@
    - v2.5.0.dev0+8759736 + v2.5.0.dev0+a4a9419
    @@ -314,6 +314,7 @@
  • Deploying Torch-TensorRT Programs
  • DLA
  • Torch Compile Advanced Usage
  • +
  • Deploy Quantized Models using Torch-TensorRT
  • Engine Caching
  • Refitting Torch-TensorRT Programs with New Weights
  • @@ -341,7 +342,6 @@
  • Example notebooks
  • Compiling ResNet using the Torch-TensorRT torch.compile Backend
  • Compiling a Transformer using torch.compile and TensorRT
  • -
  • Torch Compile Advanced Usage
  • Torch Compile Stable Diffusion
  • Torch Export with Cudagraphs
  • Using Custom Kernels within TensorRT Engines with Torch-TensorRT
  • diff --git a/docs/tutorials/_rendered_examples/dynamo/engine_caching_example.html b/docs/tutorials/_rendered_examples/dynamo/engine_caching_example.html index 0d5f095791..8cd0a4f45d 100644 --- a/docs/tutorials/_rendered_examples/dynamo/engine_caching_example.html +++ b/docs/tutorials/_rendered_examples/dynamo/engine_caching_example.html @@ -10,7 +10,7 @@ - Engine Caching — Torch-TensorRT v2.5.0.dev0+8759736 documentation + Engine Caching — Torch-TensorRT v2.5.0.dev0+a4a9419 documentation @@ -40,7 +40,7 @@ - +