pytorch

1635 issues found

DISABLED test_foreach_max_empty_tensor_error__foreach_max_xpu_bfloat16 (main.TestForeachXPU)

5/25/2026

torch.save silently truncates files at INT_MAX bytes when the output path contains non-ASCII characters

5/25/2026

DISABLED test_native_matmul_config_block_numel_limit (main.TestTritonHeuristics)

5/25/2026

[inductor] Scheduler phases scale poorly on lowered aten.lstm/aten.gru graphs

5/25/2026

DISABLED test_forward_nn_LinearCrossEntropyLoss_cuda_bfloat16 (main.TestModuleCUDA)

5/25/2026

[Dynamo] torchvision.io functions raise "RuntimeError when making fake tensor call" instead of graph breaking

5/25/2026

RFC Test class classification

5/25/2026

DTensor: `F.nll_loss(weight=..., reduction='mean')` over a sharded batch returns wrong result

5/25/2026

[Bug] `torch.library._register_fake` crashes when any unrelated `LazyLoader` module in `sys.modules` fails to import

5/24/2026

[RFC] torch.compile + CUDA graph capture: Inductor JIT has no capture-state awareness, causes CPU→GPU copy error on first call

5/24/2026

[torch.compile] AOTAutograd fails to coerce plain Tensor tangent for expected AsyncCollectiveTensor

5/23/2026

deprecation of derived quantized dtypes (quint8, qint8, qint32)

5/23/2026

`torch.compile` crashes when `torch.cat` uses `axis=` keyword — `sink_cat_after_pointwise` pass

5/23/2026

torch.export.export accepts a parameter whose storage is too small, then save/runtime fails with low-level setStorage error

5/23/2026

ConstantLR accepts factor=0 and crashes with ZeroDivisionError after total_iters steps

5/23/2026

libshm/manager.cpp:49:8: error: ‘erase_if’ is not a member of ‘std’

5/23/2026

ModuleList not recognized as Iterable by type checkers

5/23/2026

register_buffer-created attributes and Parameter.data incorrectly typed as Tensor | Module

5/23/2026

`torch.export.export` raises `RuntimeError: Unhandled FakeTensor Device Propagation` for CUDA/MPS example inputs instead of a clear user-facing error

5/22/2026

`torch.autograd.functional.jacobian` raises `NotImplementedError` at execution time for `strategy='forward-mode'` with `vectorize=False` instead of raising `ValueError` at argument validation

5/22/2026

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

pytorch

DISABLED test_foreach_max_empty_tensor_error__foreach_max_xpu_bfloat16 (main.TestForeachXPU)

torch.save silently truncates files at INT_MAX bytes when the output path contains non-ASCII characters

DISABLED test_native_matmul_config_block_numel_limit (main.TestTritonHeuristics)

[inductor] Scheduler phases scale poorly on lowered aten.lstm/aten.gru graphs

DISABLED test_forward_nn_LinearCrossEntropyLoss_cuda_bfloat16 (main.TestModuleCUDA)

[Dynamo] torchvision.io functions raise "RuntimeError when making fake tensor call" instead of graph breaking

RFC Test class classification

DTensor: `F.nll_loss(weight=..., reduction='mean')` over a sharded batch returns wrong result

[Bug] `torch.library._register_fake` crashes when any unrelated `LazyLoader` module in `sys.modules` fails to import

[RFC] torch.compile + CUDA graph capture: Inductor JIT has no capture-state awareness, causes CPU→GPU copy error on first call

[torch.compile] AOTAutograd fails to coerce plain Tensor tangent for expected AsyncCollectiveTensor

deprecation of derived quantized dtypes (quint8, qint8, qint32)

`torch.compile` crashes when `torch.cat` uses `axis=` keyword — `sink_cat_after_pointwise` pass

torch.export.export accepts a parameter whose storage is too small, then save/runtime fails with low-level setStorage error

ConstantLR accepts factor=0 and crashes with ZeroDivisionError after total_iters steps

libshm/manager.cpp:49:8: error: ‘erase_if’ is not a member of ‘std’

ModuleList not recognized as Iterable by type checkers

register_buffer-created attributes and Parameter.data incorrectly typed as Tensor | Module

`torch.export.export` raises `RuntimeError: Unhandled FakeTensor Device Propagation` for CUDA/MPS example inputs instead of a clear user-facing error

`torch.autograd.functional.jacobian` raises `NotImplementedError` at execution time for `strategy='forward-mode'` with `vectorize=False` instead of raising `ValueError` at argument validation

pytorch

DISABLED test_foreach_max_empty_tensor_error__foreach_max_xpu_bfloat16 (__main__.TestForeachXPU)

torch.save silently truncates files at INT_MAX bytes when the output path contains non-ASCII characters

DISABLED test_native_matmul_config_block_numel_limit (__main__.TestTritonHeuristics)

[inductor] Scheduler phases scale poorly on lowered aten.lstm/aten.gru graphs

DISABLED test_forward_nn_LinearCrossEntropyLoss_cuda_bfloat16 (__main__.TestModuleCUDA)

[Dynamo] torchvision.io functions raise "RuntimeError when making fake tensor call" instead of graph breaking

RFC Test class classification

DTensor: `F.nll_loss(weight=..., reduction='mean')` over a sharded batch returns wrong result

[Bug] `torch.library._register_fake` crashes when any unrelated `LazyLoader` module in `sys.modules` fails to import

[RFC] torch.compile + CUDA graph capture: Inductor JIT has no capture-state awareness, causes CPU→GPU copy error on first call

[torch.compile] AOTAutograd fails to coerce plain Tensor tangent for expected AsyncCollectiveTensor

deprecation of derived quantized dtypes (quint8, qint8, qint32)

`torch.compile` crashes when `torch.cat` uses `axis=` keyword — `sink_cat_after_pointwise` pass

torch.export.export accepts a parameter whose storage is too small, then save/runtime fails with low-level setStorage error

ConstantLR accepts factor=0 and crashes with ZeroDivisionError after total_iters steps

libshm/manager.cpp:49:8: error: ‘erase_if’ is not a member of ‘std’

ModuleList not recognized as Iterable by type checkers

register_buffer-created attributes and Parameter.data incorrectly typed as Tensor | Module

`torch.export.export` raises `RuntimeError: Unhandled FakeTensor Device Propagation` for CUDA/MPS example inputs instead of a clear user-facing error

`torch.autograd.functional.jacobian` raises `NotImplementedError` at execution time for `strategy='forward-mode'` with `vectorize=False` instead of raising `ValueError` at argument validation

DISABLED test_foreach_max_empty_tensor_error__foreach_max_xpu_bfloat16 (main.TestForeachXPU)

DISABLED test_native_matmul_config_block_numel_limit (main.TestTritonHeuristics)

DISABLED test_forward_nn_LinearCrossEntropyLoss_cuda_bfloat16 (main.TestModuleCUDA)