Update coremltools from 9b to 9.0 (#15723)

As titled

Author: metascroy

backends/apple/coreml/compiler/torch_ops.py

@@ -20,8 +20,6 @@
     NUM_TO_TORCH_DTYPE,
     split,
     to,
-    transpose,
-    unbind,
 )
 from coremltools.converters.mil.frontend.torch.torch_op_registry import (
     register_torch_op,
@@ -30,18 +28,6 @@
 from executorch.exir.dim_order_utils import get_memory_format
 
 
-# https://github.com/apple/coremltools/pull/2556
-@register_torch_op(override=False)
-def transpose_copy(context, node):
-    transpose(context, node)
-
-
-# https://github.com/apple/coremltools/pull/2557
-@register_torch_op(override=False)
-def unbind_copy(context, node):
-    unbind(context, node)
-
-
 # https://github.com/apple/coremltools/pull/2563
 @register_torch_op(override=False)
 def split_copy(context, node):
@@ -117,7 +103,9 @@ def _clone_dim_order(context, node):
 # https://github.com/apple/coremltools/pull/2558
 @register_torch_op(
     torch_alias=["torchao::dequantize_affine", "torchao.dequantize_affine"],
-    override=False,
+    # coremltools did not merge the fix into 9.0 (https://github.com/apple/coremltools/pull/2589),
+    # so we override here
+    override=True,
 )
 def dequantize_affine(context, node):
     inputs = _get_inputs(context, node, expected=[7, 8])

backends/apple/coreml/test/test_coreml_recipes.py

@@ -326,7 +326,7 @@ def forward(self, x):
         )
         self.check_fully_delegated(session)
 
-        self._compare_eager_quantized_model_outputs(session, example_inputs, atol=1e-3)
+        self._compare_eager_quantized_model_outputs(session, example_inputs, atol=1e-2)
         self._compare_eager_unquantized_model_outputs(session, model, example_inputs)
 
     def test_int8_weight_only_pt2e(self):

pyproject.toml

@@ -74,7 +74,7 @@ dependencies=[
   # See also third-party/TARGETS for buck's typing-extensions version.
   "typing-extensions>=4.10.0",
   # Keep this version in sync with: ./backends/apple/coreml/scripts/install_requirements.sh
-  "coremltools==9.0b1; platform_system == 'Darwin' or platform_system == 'Linux'",
+  "coremltools==9.0; platform_system == 'Darwin' or platform_system == 'Linux'",
   # scikit-learn is used to support palettization in the coreml backend
   "scikit-learn==1.7.1",
   "hydra-core>=1.3.0",