Important
Summary of Bugs
System | #Fixed | #Confirmed | #Pending | #Total |
---|---|---|---|---|
PyTorch | 57 | 11 | 16 | 84 |
PyTorch-ONNX Converter | 12 | 0 | 0 | 12 |
ONNX | 1 | 0 | 0 | 1 |
ONNXRuntime | 3 | 4 | 4 | 11 |
TVM | 34 | 0 | 6 | 40 |
TensorRT | 6 | 2 | 2 | 10 |
TensorFlow | 0 | 14 | 0 | 14 |
Hidet | 13 | 0 | 0 | 13 |
Sum | 126 | 31 | 28 | 185 |
Note
- Status: ✅ means fixed; 🔵 means confirmed; 🚨 means this bug has been marked with a
high-priority
label (PyTorch) - Symptom: 💥 Crash or exception; 🧮 Result inconsistency (silent semantic bug); 🧴 Sanitizers
Note
Table of Content
- ✅🧮🚨
torch.compile
produce wrong result in backward ad withconv2d + interpolate
when interpolatemode=nearest/bilinear/bicubic
· Issue #100794 · pytorch/pytorch - ✅💥🚨 SIGIOT when running model with conv2d and avgpool2d after
optimize_for_inference
· Issue #86535 · pytorch/pytorch · GitHub - ✅🧮
optimize_for_inference
leads to wrong results for model with conv2d, max and clip · Issue #86556 · pytorch/pytorch · GitHub - ✅💥🚨 RuntimeError: could not construct a memory descriptor using a format tag · Issue #86664 · pytorch/pytorch · GitHub
- ✅💥 [NNPack] Runtime error with padded
Conv1d
and>=16
batch size · Issue #90142 · pytorch/pytorch · GitHub - ✅💥 stable
torch.sort
crash with expanded tensor · Issue #91420 · pytorch/pytorch · GitHub - ✅💥 [Crash]
torch.searchsorted
with out-of-bound sorter · Issue #91606 · pytorch/pytorch · GitHub - ✅🧮
index_select
with scalar input and 0-dimed vector leads to undeterministic output · Issue #94340 · pytorch/pytorch · GitHub - ✅🧮
index_select
with scalar input and 0-dimed vector leads to undeterministic output · Issue #94340 · pytorch/pytorch · GitHub - ✅💥
torch.compile
failed ontorch.add
with a constant python number · Issue #92324 · pytorch/pytorch · GitHub - ✅💥
torch.compile
generates wrong profiling program forrandn_like
· Issue #92368 · pytorch/pytorch · GitHub - ✅💥
torch.compile
generates wrong profiling program for function havingtranspose
andlerp
· Issue #93229 · pytorch/pytorch · GitHub - ✅💥
torch.compile
triggers assertion error when explicitly provideout=None
· Issue #92814 · pytorch/pytorch · GitHub - ✅💥 INTERNAL ASSERT FAILED in
torch.compile
when the input tensor oftorch.clamp
hasrequires_grad=True
· Issue #93225 · pytorch/pytorch · GitHub - ✅💥
torch.compile
failed to run in-place operationunsqueeze_(0)
· Issue #93259 · pytorch/pytorch · GitHub - ✅🧮
stack
+ inplace operator produce wrong results intorch.compile
· Issue #93283 · pytorch/pytorch · GitHub - ✅🧮 [pt2] compiled model with cat and expand gives wrong results · Issue #93357 · pytorch/pytorch · GitHub
- ✅🧮🚨 [pt2] compiled function with cat and mul gives wrong results · Issue #93365 · pytorch/pytorch · GitHub
- ✅🧮 [pt2] cannot compile model with linear layer when the input has rank 1 · Issue #93372 · pytorch/pytorch · GitHub
- ✅💥
softmax
+transpose
+div_
triggers assertion fail in compile mode · Issue #93371 · pytorch/pytorch · GitHub - ✅🧮🚨 [pt2]
torch.where
gives wrong results withtorch.compile
· Issue #93374 · pytorch/pytorch · GitHub - ✅💥
torch.rsub
withalpha=xxx
triggers assertion fail in compile mode · Issue #93376 · pytorch/pytorch · GitHub - ✅🧮 [pt2] compile gives wrong result for function having
expand
anddiv_
· Issue #93377 · pytorch/pytorch · GitHub - ✅💥 [pt2]
torch._inductor.exc.CppCompileError: C++ compile error
when compilingneg
andmax
· Issue #93380 · pytorch/pytorch · GitHub - ✅💥 [pt2] exception when compiling
max_pool2d_with_indices
· Issue #93384 · pytorch/pytorch · GitHub - ✅💥 [pt2] cannot compile function having
gt
,expand
andadd_
· Issue #93386 · pytorch/pytorch · GitHub - ✅💥🚨
torch.compile
trigger assertion error when executinghistogramdd
· Issue #93274 · pytorch/pytorch · GitHub - ✅🧮 [pt2]
torch.compile
produces wrong results formasked_fill
· Issue #93823 · pytorch/pytorch · GitHub - ✅🧮 [pt2]
torch.compile
produces wrong results for function withreciprocal_
· Issue #93824 · pytorch/pytorch · GitHub - ✅🧮 [pt2]
torch.compile
produces wrong results for function withneg
onuint8
tensor · Issue #93829 · pytorch/pytorch · GitHub - ✅💥
log_softmax
+pad
triggers assertion fail in compile mode · Issue #93819 · pytorch/pytorch · GitHub - ✅💥 [pt2] Cannot compile model with
neg
andlinear
· Issue #93836 · pytorch/pytorch · GitHub - ✅🧮
pad
+gt
produce wrong results in compile mode · Issue #93351 · pytorch/pytorch · GitHub - ✅💥 [pt2] (
interpolate
withmode=nearest
) +kthvalue
triggers assertion fail in compile mode · Issue #93849 · pytorch/pytorch · GitHub - ✅💥 [pt2]
torch._inductor.exc.CppCompileError: C++ compile error
when compilingargmax
andmin
· Issue #94055 · pytorch/pytorch · GitHub - ✅💥
Tensor.select
+add_
triggers C++ Compile Error · Issue #94960 · pytorch/pytorch · GitHub - ✅💥
torch.compile
fails when usingtorch.sub
with python constant · Issue #95181 · pytorch/pytorch · GitHub - ✅💥
Tensor.copy_
+moveaxis
Trigger Exception in Compile Mode · Issue #95262 · pytorch/pytorch · GitHub - ✅🧮
torch.ge
produces wrong results in compile mode when given int tensors · Issue #95695 · pytorch/pytorch · GitHub - ✅💥 [pt2]
bitwise_and
+clamp_max
Triggers Compilation Error · Issue #97968 · pytorch/pytorch · GitHub - ✅🧮 [pt2]
add
+unfold
+abs_
produces wrong results · Issue #98143 · pytorch/pytorch · GitHub - ✅🧮 [pt2]
pow
+cos
produces wrong result · Issue #98149 · pytorch/pytorch · GitHub - ✅💥
torch._C._nn.fractional_max_pool3d
Trigger Segmentation Fault · Issue #89648 · pytorch/pytorch · GitHub - ✅💥🚨
torch.nn.functional.embedding_bag
Trigger "IOT instruction" Failure · Issue #89677 · pytorch/pytorch · GitHub - ✅🧴
torch.Tensor.index_select
Trigger heap-buffer-overflow with AddressSanitizer · Issue #88940 · pytorch/pytorch · GitHub - ✅🧴
nn.utils.rnn.pack_sequence
Trigger heap-buffer-overflow with AddressSanitizer · Issue #88334 · pytorch/pytorch · GitHub - ✅🚨🧴
MultiMarginLoss
Trigger out-of-bound Read under Compute Sanitizer · Issue #88724 · pytorch/pytorch · GitHub - ✅🧴
nn.functional.max_unpool3d
Trigger heap-buffer-overflow with AddressSanitizer · Issue #88032 · pytorch/pytorch · GitHub - ✅🧴
torch.nn.functional.interpolate
Trigger heap-buffer-overflow with AddressSanitizer · Issue #88939 · pytorch/pytorch · GitHub - ✅🧴
torch.fft.hfft
Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88985 · pytorch/pytorch · GitHub - ✅🧴
torch.nn.functional.interpolate
Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88951 · pytorch/pytorch · GitHub - ✅💥🚨
torch.compile
failed ontorch.bitwise_xor
with a constant python number · Issue #93224 · pytorch/pytorch · GitHub - ✅💥 [CPU Inductor] Compile error when passing float16 tensors to
vector_norm
+remainder
· Issue #97758 · pytorch/pytorch · GitHub - ✅💥 [pt2]
movedim
+add_
+cat
triggers exception · Issue #98122 · pytorch/pytorch · GitHub - ✅🧮
dstack
+reciprocal
produce wrong result in compile mode · Issue #93078 · pytorch/pytorch · GitHub - ✅💥
min
reduction on float16 tensor failed on certain shapes · Issue #93249 · pytorch/pytorch · GitHub - ✅💥
argmin
+view
Trigger Exception in compile mode · Issue #95370 · pytorch/pytorch · GitHub - 🔵💥 [JIT] Zero-channel conv2d cannot be applied with
optimize_for_inference
· Issue #91396 · pytorch/pytorch · GitHub - 🔵💥 [JIT] Applying
conv2d
over Constants Leads to Exception · Issue #92740 · pytorch/pytorch · GitHub - 🔵🧮🚨
torch.compile
produce wrong result ininterpolate
whenmode=bilinear
· Issue #93262 · pytorch/pytorch · GitHub - 🔵🧮
torch.fmod
produces inconsistent results in eager and compile mode · Issue #97333 · pytorch/pytorch · GitHub - 🔵💥
torch.Tensor.flatten
Trigger Segmentation Fault when trying to provide and output named dim · Issue #89718 · pytorch/pytorch · GitHub - 🔵🧴
nn.functional.embedding_bag
Trigger out-of-bound Read under Compute Sanitizer · Issue #88563 · pytorch/pytorch · GitHub - 🔵🧴
torch.nn.CTCLoss
Trigger heap-buffer-overflow under AddressSanitizer · Issue #88047 · pytorch/pytorch · GitHub - 🔵🧴
torch.nn.ReplicationPad2D
Report "invalid configuration argument" Error under Compute Sanitizer · Issue #89254 · pytorch/pytorch · GitHub - 🔵🧴
torch.nn.LayerNorm
Abort with "invalid device ordinal" Error · Issue #89218 · pytorch/pytorch · GitHub - 🔵🧴
torch.svd_lowrank
Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88942 · pytorch/pytorch · GitHub - 🔵🧴
torch.linalg.lstsq
Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88941 · pytorch/pytorch · GitHub - 💥 Adding a linear layer leads to failure of
optimize_for_mobile
· Issue #86667 · pytorch/pytorch · GitHub - 💥 [JIT] INTERNAL ASSERT FAILED when dispatching for
torch.Tensor.view
· Issue #90365 · pytorch/pytorch · GitHub - 💥 [JIT] INTERNAL ASSERT FAILED
torch.add
with boolean primitive constant · Issue #90367 · pytorch/pytorch · GitHub - 💥 [JIT] INTERNAL ASSERT FAILED
torch.mul
with boolean primitive constant · Issue #90366 · pytorch/pytorch · GitHub - 💥 [JIT] Wrong type inference leads to misleading error message · Issue #90369 · pytorch/pytorch · GitHub
- 💥 [JIT] INTERNAL ASSERT FAILED when
Conv2d
andclamp
used together · Issue #92563 · pytorch/pytorch · GitHub - 💥 [JIT] Inconsistency in tensor shape between eager mode and JIT · Issue #92548 · pytorch/pytorch · GitHub
- 💥 [JIT][TracingCheckError] inplace ops incompatible with
contiguous(.., channels_last)
· Issue #92558 · pytorch/pytorch · GitHub - 💥 [JIT] Consecutive use of
addmm
Leads to Exception · Issue #92742 · pytorch/pytorch · GitHub - 🧴
torch.topk
Trigger RuntimError under UndefinedBehaviorSanitizer · Issue #88944 · pytorch/pytorch · GitHub - 🧴
torch.vander
Trigger RuntimeError with UndefinedBehaviorSanitizer · Issue #88943 · pytorch/pytorch · GitHub - 🧴
torch.nn.CTCLoss
Trigger out-of-bound Read under Compute Sanitizer · Issue #89208 · pytorch/pytorch · GitHub - 🧴
torch.nn.functional.embedding_bag
Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88950 · pytorch/pytorch · GitHub - 🧴
torch.set_rng_state
Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88949 · pytorch/pytorch · GitHub - 🧴
torch.Tensor.msort
Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88947 · pytorch/pytorch · GitHub - 🧴
torch.linalg.eigvals
Trigger RuntimeError under UndefinedBehaviorSanitizer · Issue #88945 · pytorch/pytorch · GitHub
- ✅ [ONNX]
f64 * LeakyReLU(f64)
mistakingly returns f32 · Issue #85316 · pytorch/pytorch · GitHub - ✅ [ONNX] Converter did not consider the implicit casting specifically for
Max
· Issue #87609 · pytorch/pytorch · GitHub - ✅ fix: onnx PReLU unidirectional broadcasting by ganler · Pull Request #70571 · pytorch/pytorch · GitHub
- ✅ [Clip] [ONNX] Make Non-Float Op Exportation Compatible to Avoid Invalid ONNX Models by ganler · Pull Request #72401 · pytorch/pytorch · GitHub
- ✅ [Min] [ONNX] Make Non-Float Op Exportation Compatible to Avoid Invalid ONNX Models by ganler · Pull Request #72401 · pytorch/pytorch · GitHub
- ✅ [Max] [ONNX] Make Non-Float Op Exportation Compatible to Avoid Invalid ONNX Models by ganler · Pull Request #72401 · pytorch/pytorch · GitHub
- ✅ [ReLU] [ONNX] Make Non-Float Op Exportation Compatible to Avoid Invalid ONNX Models by ganler · Pull Request #72401 · pytorch/pytorch · GitHub
- ✅ [Pad] [ONNX] Make Non-Float Op Exportation Compatible to Avoid Invalid ONNX Models by ganler · Pull Request #72401 · pytorch/pytorch · GitHub
- ✅ [onnx export] Add broadcast to matmul shape inference by lazycal · Pull Request #70534 · pytorch/pytorch · GitHub
- ✅ [Bug][ONNX] Specification Inconsistency in Flatten · Issue #74142 · pytorch/pytorch · GitHub
- ✅ [ONNX] Fix shape inconsistency when exporting scalar log2 by lazycal · Pull Request #78701 · pytorch/pytorch · GitHub
- ✅ [ONNX Export] Interpolation likely should be exported with
half_pixel
instead ofpytorch_half_pixel
· Issue #79361 · pytorch/pytorch · GitHub
- ✅ Crashes when relu is followed by a clip · Issue #9753 · microsoft/onnxruntime · GitHub
- ✅ MatMul fusion failed at scalar input · Issue #10950 · microsoft/onnxruntime · GitHub
- ✅ GemmTransposeFusion error when C is transposed (
Gemm(A,B,Transpose(C)
), complained with confusing name_transformed_transformed_transformed...
· Issue #12071 · microsoft/onnxruntime · GitHub - 🔵 [Bug] Mixing negative and positive paddings causes segfault/uninitialized memory values produced in reflected pad · Issue #11828 · microsoft/onnxruntime · GitHub
- 🔵 Runtime Exception when relu is followed by a clip · Issue #10936 · microsoft/onnxruntime · GitHub
- 🔵 Inconsistent result to NumPy and PyTorch when consecutively casting a float tensor to int32 and then to bool · Issue #11994 · microsoft/onnxruntime · GitHub
- 🔵 Wrong output shape due to MergeShape failure · Issue #11870 · microsoft/onnxruntime · GitHub
- Wrong Floor output on very large input · Issue #12076 · microsoft/onnxruntime · GitHub
- Resize with mode linear always produces 0.5 on GPU regardless of the input · Issue #12091 · microsoft/onnxruntime · GitHub
- Resize with
nearest
mode have inconsistent results compared to PyTorch and TVM · Issue #12098 · microsoft/onnxruntime · GitHub - Parameters are optimized out even if it is a needed return value · Issue #13425 · microsoft/onnxruntime · GitHub
- ✅ [Bug] shape int32-int64 check error in
trilu
'ste.compute
· Issue #13029 · apache/tvm · GitHub - ✅ [Bug]
trilu
not tagged withinjective
and thus miss reduce schedule · Issue #13030 · apache/tvm · GitHub - ✅ [Bug] Wrong results of
cast<int32>( cast<bool>(-1i64) )
· Issue #13048 · apache/tvm · GitHub - ✅ [BugFix] resolve integer 32. ~ 64. mismatch by casting by ganler · Pull Request #9582 · apache/tvm · GitHub
- ✅ [onnx] fix onnx where broadcast by lazycal · Pull Request #10106 · apache/tvm · GitHub
- ✅ Fix broadcast InferCorrectLayout by lazycal · Pull Request #10156 · apache/tvm · GitHub
- ✅ [BUGFIX][ARITH] Fix FloorMod Simplifier by lazycal · Pull Request #10336 · apache/tvm · GitHub
- ✅ [BugFix]: select node type error in NarrowDataType pass by ganler · Pull Request #10519 · apache/tvm · GitHub
- [Bug] GetStoreRule failure at simple Conv2d + Squeeze model · Issue #10528 · apache/tvm · GitHub
- ✅ [Relay][ONNX][Fix] Flatten in OnnxConverter by ganler · Pull Request #10593 · apache/tvm · GitHub
- ✅ [NarrowDataType] [TIR] Fix Ramp int32~64 mismatch in VectorizeLoop and NarrowDataType passes by lazycal · Pull Request #10172 · apache/tvm · GitHub
- ✅ [VectorizeLoop] [TIR] Fix Ramp int32~64 mismatch in VectorizeLoop and NarrowDataType passes by lazycal · Pull Request #10172 · apache/tvm · GitHub
- ✅ [Bug][TE Schedule] Unsupported nested parallel created by Softmax TE schedule · Issue #12001 · apache/tvm · GitHub
- ✅ [fix] vec * mat in matmul in onnx converter by ganler · Pull Request #11174 · apache/tvm · GitHub
- ✅ fix vec*mat in PyTorch converter by ganler · Pull Request #11347 · apache/tvm · GitHub
- ✅ [TIR] Fix int32 vs int64 mismatch in For construct. by lazycal · Pull Request #10595 · apache/tvm · GitHub
- ✅ Add missing Slice layout fallback check of
stride=1
. by lazycal · Pull Request #10690 · apache/tvm · GitHub - ✅ Onnx squeeze enabled with auto axis handling. by ganler · Pull Request #10742 · apache/tvm · GitHub
- ✅ [Reduce] [ONNX] fix reduce crash on scalar inputs by ganler · Pull Request #10780 · apache/tvm · GitHub
- ✅ [ReduceSumSquare] [ONNX] fix reduce crash on scalar inputs by ganler · Pull Request #10780 · apache/tvm · GitHub
- ✅ [ReduceL1] [ONNX] fix reduce crash on scalar inputs by ganler · Pull Request #10780 · apache/tvm · GitHub
- ✅ [ReduceL2] [ONNX] fix reduce crash on scalar inputs by ganler · Pull Request #10780 · apache/tvm · GitHub
- ✅ [ReduceLogSum][ONNX] fix reduce crash on scalar inputs by ganler · Pull Request #10780 · apache/tvm · GitHub
- ✅ [FIX] resolve int64/32 for AttrStmtNode by ganler · Pull Request #10983 · apache/tvm · GitHub
- ✅ Fix onnx round import with float64 inputs. by lazycal · Pull Request #11685 · apache/tvm · GitHub
- ✅ Fix 1d-softmax schedule. by lazycal · Pull Request #11719 · apache/tvm · GitHub
- ✅ [Fix] int32/64 mismatch of buffer elem_offset at HandleBufferBindScope by ganler · Pull Request #11755 · apache/tvm · GitHub
- ✅ [Bug] Int64 BroadCast-ArgMax triggers assertion error at graph runtime · Issue #11794 · apache/tvm · GitHub
- ✅ [TE Schedule] Fix broken 2D softmax TE schedules when axis=0 by lazycal · Pull Request #11803 · apache/tvm · GitHub
- ✅ [Bug]
concat([x], axis=1)
return random results · Issue #11895 · apache/tvm · GitHub - ✅ Fix infercorrect layout in Layoutrewrite and improve naming. by lazycal · Pull Request #12007 · apache/tvm · GitHub
- ✅ Several type mismatch fixes and checks by kparzysz-quic · Pull Request #12041 · apache/tvm · GitHub
- ✅ [FIX][ONNX][Relay] onnx converter on matmul with scalar; bring back nn.matmul check by ganler · Pull Request #13448 · apache/tvm · GitHub
- ✅ [Bug] Layout Error when Putting
argmin
afterconv2d
· Issue #9813 · apache/tvm · GitHub - ✅ Fix LayoutRewriter by lazycal · Pull Request #10118 · apache/tvm · GitHub
- [Bug] concatenating strided slice and negative padding causes wrong buffer binding · Issue #11897 · apache/tvm
- [Bug] GPU
lower_thread_allreduce
is_zero(index) check false · Issue #11898 · apache/tvm - Resize does not reject unsupported layout during AlterOpLayout · Issue #12008 · apache/tvm
- [Bug] Compilation failure for
broadcast-argmax
in internal type inference · Issue #13031 · apache/tvm · GitHub - [Bug] Compiled
squeeze-broadcast_to-argmin
fails at graph runtime · Issue #13045 · apache/tvm · GitHub
- ✅ Segfault on const+prelu+reduce_mean+comparison_op · Issue #1738 · NVIDIA/TensorRT · GitHub
- ✅ Gemm conversion error, seem to be caused by squeeze · Issue #824 · onnx/onnx-tensorrt · GitHub
- ✅ [Bug] crash on poolings with larger-than-317 pool sizes · Issue #2094 · NVIDIA/TensorRT
- ✅ [Bug] --loadInputs not working: input name mismatch when Flatten is the input node · Issue #1990 · NVIDIA/TensorRT · GitHub
- ✅ Cuda OutOfMemory when creating tensor with 2^29 (~0.5 G) elements - TensorRT - NVIDIA Developer Forums
- ✅ Myelin error on onnx model: Assertion `i < crds_.size() < failed · Issue #1781 · NVIDIA/TensorRT · GitHub
- Segmentation fault when using TensorRT to compile a model - TensorRT - NVIDIA Developer Forums
- Internal Error: GPU error during getBestTactic: PWN(LeakyRelu_4) : misaligned address - TensorRT - NVIDIA Developer Forums
- 🔵 Duplicated reshapes triggers "[graphOptimizer.cpp::findOne::510] Error Code 2: Internal Error (Assertion it != v.end() failed. )" - TensorRT - NVIDIA Developer Forums
- 🔵 Incorrect slicing of boolean constant tensor with step size > 1 - TensorRT - NVIDIA Developer Forums
- 🔵💥 Inconsistant behavior of Conv2D between eager mode and tracing · Issue #57664 · tensorflow/tensorflow · GitHub
- 🔵💥 TFLite fails to run a model with a dense layer following an Add operator · Issue #57697 · tensorflow/tensorflow · GitHub
- 🔵💥 TFLite throws an error with certain tensor value · Issue #57708 · tensorflow/tensorflow · GitHub
- 🔵🧮 TFLite's max operator has wrong broadcasting behavior · Issue #57759 · tensorflow/tensorflow · GitHub
- 🔵💥 Issues · tensorflow/tensorflow · GitHub
- 🔵🧮 pow operation gives valid output even the input is invalid · Issue #57757 · tensorflow/tensorflow · GitHub
- 🔵🧮 TFLite produce wrong results when add follows a leakyrelu · Issue #57818 · tensorflow/tensorflow · GitHub
- 🔵💥 TFLite runner crashes with XOR and squeeze in the model · Issue #57882 · tensorflow/tensorflow · GitHub
- 🔵💥 Conv2D with XLA jit_compile=True fails to run · Issue #57748 · tensorflow/tensorflow · GitHub
- 🔵🧮 log operator outputs wrong results with XLA compilation · Issue #57744 · tensorflow/tensorflow · GitHub
- 🔵🧮 pow operator output nan for valid inputs · Issue #57747 · tensorflow/tensorflow · GitHub
- 🔵🧮 LRN operator outputs wrong results with
jit_compile=True
· Issue #57746 · tensorflow/tensorflow · GitHub - 🔵💥 Conv2D layer fails to run with XLA on CUDA · Issue #57838 · tensorflow/tensorflow · GitHub
- 🔵🧴
tf.raw_ops.SegmentMax
Behaves Differently Under CPU and GPU · Issue #58469 · tensorflow/tensorflow · GitHub
Based on NNSmith, @soodoshll found a number of bugs for Hidet, including:
- ✅ [Bug] Use int64 in argmax · Issue #103 · hidet-org/hidet
- ✅ [Bug] broadcast_shape parameter type error · Issue #85 · hidet-org/hidet
- ✅ [Bug] Data type casting from onnx · Issue #87 · hidet-org/hidet
- ✅ [Bug] cuda code compilation error · Issue #89 · hidet-org/hidet
- ✅ [Bug] MinOp generates max code · Issue #90 · hidet-org/hidet
- ✅ [Bug] FP64 reduce · Issue #91 · hidet-org/hidet
- ✅ [Bug] Inconsistent definition of the inputs parameter of operators · Issue #93 · hidet-org/hidet
- ✅ [Bug] Slice indexing in ONNX · Issue #94 · hidet-org/hidet
- ✅ [Bug] binary arithmetic with CUDA scalar · Issue #95 · hidet-org/hidet
- ✅ [Bug] Unexpected behavior when inputs and outputs overlap · Issue #96 · hidet-org/hidet
- ✅ [Followup][Bug] Unexpected behavior when inputs and outputs overlap · Issue #96 · hidet-org/hidet
- ✅ [Bug] arguments of clip drop after fusion · Issue #97 · hidet-org/hidet
- ✅ [Bug] fusion rewrite fails · Issue #99 · hidet-org/hidet
Note
Methodology
- Though most bugs are identified via individual reports, there are cases where multiple similar-looking bugs are merged into one report to avoid potential duplication. Nonetheless, they might be counted for multiple times according to the actual required different fixes.
- "won't fix" bugs are omitted.
- Detected bugs come from the ASPLOS'23 (NNSmith) and FSE'23 (NeuRI) projects.