pytorch/pytorch issues and pull requests

#131958 - [DEBUG] Strange behavior observed with PyTorch 2.4.0 + Windows + CPU inference

Issue - State: closed - Opened by Burhan-Q over 1 year ago - 42 comments
Labels: high priority, needs reproduction, module: windows, module: cpu, triaged, module: regression

#131864 - 2.4.0: third_party/QNNPACK is expected by the build but there is no such directry or git submodule

Pull Request - State: closed - Opened by yurivict over 1 year ago - 6 comments
Labels: module: build, triaged, module: regression, module: devx

#131662 - OSError: [WinError 126] The specified module could not be found.

Issue - State: closed - Opened by Noor-Nizar over 1 year ago - 76 comments
Labels: high priority, module: binaries, module: windows, triaged, module: regression

#130619 - PyTorch 2.4 windows performance regression compared with 0410 nightly

Issue - State: closed - Opened by WeizhuoZhang-intel over 1 year ago - 21 comments
Labels: high priority, module: binaries, module: windows, triaged, module: regression

#130414 - [c10d] collective failure with single-device style (#119421) and nccl <= 2.12

Issue - State: closed - Opened by oraluben over 1 year ago - 5 comments
Labels: high priority, triage review, oncall: distributed, module: regression

#130034 - Windows nightly Wheel and Libtorch builds are broken Jul 3, 2024

Issue - State: closed - Opened by atalman over 1 year ago - 6 comments
Labels: high priority, module: binaries, module: ci, triaged, module: regression

#129777 - Failed to compute shorthash for libnvrtc.so

Issue - State: closed - Opened by judicaelclair over 1 year ago - 2 comments
Labels: module: build, triaged, module: regression

#129060 - Incompatability between torch>=2.3 and torchdatasets==0.2.0

Issue - State: closed - Opened by Zerhigh over 1 year ago - 2 comments
Labels: module: binaries, module: windows, triaged, module: regression

#128394 - [export] Failed to trace HF Llama2 model

Issue - State: closed - Opened by kwen2501 over 1 year ago - 12 comments
Labels: high priority, triage review, module: regression, oncall: pt2, oncall: export

#128130 - PyTorch fails to build on ppc64le

Issue - State: closed - Opened by mikejuliet13 over 1 year ago - 1 comment
Labels: module: build, triaged, module: regression, module: POWER, actionable, module: multi-headed-attention

#127924 - recent torchinductor changes seems to break torchao CI

Issue - State: closed - Opened by jerryzh168 over 1 year ago - 1 comment
Labels: high priority, triage review, module: regression, oncall: pt2

#127821 - Inductor lowering ignores auto_functionalized custom op

Issue - State: closed - Opened by zou3519 over 1 year ago - 8 comments
Labels: high priority, triaged, module: regression, oncall: pt2, module: inductor, module: pt2-dispatcher

#127383 - 4 GPT2 can't run into `_scaled_dot_product_flash_attention_for_cpu` using AOTI due to export related change in https://github.com/pytorch/pytorch/pull/123732

Issue - State: closed - Opened by chunyuan-w over 1 year ago - 5 comments
Labels: triaged, module: regression, oncall: pt2, module: aotinductor, module: sdpa

#126919 - Tensor computation error on MPS backend

Issue - State: closed - Opened by Caz-T over 1 year ago - 8 comments
Labels: triaged, module: regression, module: mps

#125879 - CUDA nightly docker actually includes CPU build of torch

Issue - State: closed - Opened by bhack over 1 year ago - 41 comments
Labels: high priority, triaged, module: regression, module: docker

#125871 - Error linking to libtorch (specifically mkl) on Windows with OneAPI and Visual Studio

Issue - State: open - Opened by jatkinson1000 over 1 year ago - 10 comments
Labels: high priority, module: binaries, module: windows, module: cpp, triaged, module: regression

#125832 - failed assertion `[MPSNDArrayDescriptor sliceDimension:withSubrange:] error: subRange.start (6) is not less than length of dimension[0] (6)'

Issue - State: closed - Opened by shink over 1 year ago - 5 comments
Labels: needs reproduction, module: ci, triaged, module: regression, module: mps

#125803 - Some MPS functionality regressed in MacOS-14.4

Issue - State: closed - Opened by malfet over 1 year ago - 4 comments
Labels: high priority, module: ci, triaged, module: regression, module: mps

#125548 - linux aarch64: CI: test_mkldnn started failing with oneDNN3.4+ACL24.04 version upgrade

Issue - State: closed - Opened by snadampal over 1 year ago - 17 comments
Labels: high priority, triaged, module: mkldnn, module: regression, module: arm

#125544 - spectral_norm not working with autocast

Issue - State: closed - Opened by nagadomi over 1 year ago - 5 comments
Labels: high priority, triaged, module: regression, module: amp (automated mixed precision), module: norms and normalization, module: nn.utils.parametrize

#125094 - Broken Docker Image on dockerhub

Issue - State: closed - Opened by lharri73 over 1 year ago - 6 comments
Labels: high priority, triaged, module: regression, module: docker

#124941 - Torch nightly wheels no longer include `torchgen` YAML files

Issue - State: closed - Opened by antoniojkim over 1 year ago - 4 comments
Labels: high priority, triage review, module: binaries, module: regression

#124922 - Time taken to data loading increased in newer builds (ARM)

Issue - State: closed - Opened by dilililiwhy over 1 year ago - 7 comments
Labels: module: binaries, module: performance, triaged, module: regression, module: arm

#124907 - 2.3.0 not backward compatible with torchdata

Issue - State: closed - Opened by npuichigo over 1 year ago - 4 comments
Labels: module: binaries, triaged, module: regression

#124897 - Certain .pyi files are not encoded as UTF-8 in Windows

Issue - State: closed - Opened by marovira over 1 year ago - 11 comments
Labels: module: binaries, module: typing, triaged, module: regression

#124821 - torch.compile error: Unsupported reduction type from torch.float32 to torch.int64

Issue - State: closed - Opened by syheliel over 1 year ago - 9 comments
Labels: high priority, triaged, module: regression, oncall: pt2, oncall: cpu inductor

#124732 - ARM libtorch links to libomp but libomp is no longer bundled

Issue - State: closed - Opened by hugary1995 over 1 year ago - 4 comments
Labels: high priority, module: binaries, module: cpp, triaged, module: macos, module: openmp, module: regression

#124497 - torch.compile does not work since 2.2.1 on MacOS for some models

Issue - State: closed - Opened by maxbartel over 1 year ago - 10 comments
Labels: high priority, triaged, module: macos, module: regression, has workaround, oncall: pt2, oncall: cpu inductor

#123343 - MacOS nightly builds are broken

Issue - State: closed - Opened by malfet over 1 year ago - 3 comments
Labels: module: binaries, triaged, module: macos, module: regression

#123225 - Nightly MacOS wheels are no longer build with OpenMP support

Issue - State: closed - Opened by malfet over 1 year ago - 9 comments
Labels: module: binaries, triaged, module: macos, module: regression

#123178 - [MPS] Gradient error with LeakyRELU/Mish/GELU + .sum() on 1D and scalar tensors

Issue - State: closed - Opened by jtang98 over 1 year ago - 4 comments
Labels: triaged, module: regression, module: correctness (silent), module: mps

#122394 - [MPS] Severe Performance Degradation with Deterministic Algorithm on macOS 14.4

Issue - State: open - Opened by 1155159410 over 1 year ago - 10 comments
Labels: module: performance, triaged, module: regression, module: mps

#122206 - AWS config change stripped public access from number of CI buckets, but no alerts were raised

Issue - State: closed - Opened by malfet over 1 year ago - 7 comments
Labels: high priority, module: ci, triaged, module: regression

#122016 - [MPS] Regression from macOS 14.3 to 14.4 in PyTorch 2.2.0/2.2.1

Issue - State: closed - Opened by AuroraWright over 1 year ago - 6 comments
Labels: high priority, triaged, module: regression, module: correctness (silent), module: mps

#120982 - conv_transpose1d is 1000x times slow in torch 2.2.1+cpu vs torch 1.13.1+cpu

Issue - State: closed - Opened by codetogamble over 1 year ago - 22 comments
Labels: high priority, module: performance, module: cpu, module: convolution, triaged, module: regression

#120547 - F.conv1d fails with `RuntimeError: cannot create std::vector larger than max_size()` after transpose of input tensor

Issue - State: closed - Opened by AlexanderDokuchaev over 1 year ago - 7 comments
Labels: high priority, module: cpu, module: convolution, triaged, module: mkldnn, module: regression

#120406 - Segmentation Fault in PyTorch 2.2 with `F.conv3d` function

Issue - State: closed - Opened by KumoLiu over 1 year ago - 17 comments
Labels: module: nn, triaged, module: mkldnn, module: regression

#120020 - `Unknown layout` while initializing tensors

Issue - State: closed - Opened by sssrijan-amazon almost 2 years ago - 13 comments
Labels: triage review, module: binaries, module: cpp, module: regression

#119989 - PyTorch 2.2 conflict with locally installed cuDNN (version:

Issue - State: closed - Opened by cblmemo almost 2 years ago - 15 comments
Labels: module: binaries, module: cuda, triaged, module: regression

#118862 - Using package Torch from CMake fails

Issue - State: closed - Opened by chrismile almost 2 years ago - 13 comments
Labels: high priority, triage review, module: build, triaged, module: mkl, module: regression

#118737 - PermissionError: [WinError 32] The process cannot access the file because it is being used by another process

Issue - State: closed - Opened by nabenabe0928 almost 2 years ago - 15 comments
Labels: high priority, triage review, module: binaries, oncall: distributed, module: windows, module: regression

#118349 - Why torch.mm of torch2 is slower than torch1?

Issue - State: closed - Opened by TXacs almost 2 years ago - 4 comments
Labels: module: performance, triaged, module: regression, needs research

#118135 - TestDecompCUDA.test_quick_all_cuda_float32 fails with cuda illegal memory access

Issue - State: closed - Opened by xwang233 almost 2 years ago - 12 comments
Labels: high priority, triage review, module: cuda, module: regression

#117465 - Apple MPS: Failed to create indexing library

Issue - State: closed - Opened by iSach almost 2 years ago - 6 comments
Labels: high priority, triage review, triaged, module: regression, module: mps

#117081 - PyTorch 2.1.0 Performance regression from PyTorch 2.0.1

Issue - State: closed - Opened by sirutBuasai almost 2 years ago - 8 comments
Labels: triage review, module: performance, module: cuda, module: regression

#116623 - Pytorch 2.1.X SIGILL on CPUs that do not support SSE4.1

Issue - State: closed - Opened by fpoisson35 almost 2 years ago - 19 comments
Labels: high priority, module: binaries, module: cpu, triaged, module: mkldnn, module: regression

#116276 - `import torch` results in `cuInit` call

Issue - State: closed - Opened by malfet almost 2 years ago - 8 comments
Labels: high priority, module: cuda, module: ci, triaged, module: regression

#115575 - Performance regression using latest nightlies and HF transformers

Issue - State: closed - Opened by tdoublep almost 2 years ago - 3 comments
Labels: high priority, module: performance, triaged, module: regression, oncall: pt2

#115482 - `python -c "import torch;print(torch.nn.GELU()(torch.rand(2)))"` crashes on aarch64

Issue - State: closed - Opened by malfet almost 2 years ago - 8 comments
Labels: high priority, triaged, module: mkldnn, module: regression, has workaround, module: arm

#114529 - Bug in element-wise multiplication of `torch.sparse_csr_tensor`s on GPU - 0's in result considered significant - PyTorch 2.1.1

Issue - State: open - Opened by Mystic-Slice almost 2 years ago - 5 comments
Labels: module: sparse, triaged, module: regression

#113652 - [ONNX] ONNX export fails when combining tracing and scripting in the presence of symbolic functions

Issue - State: closed - Opened by ekaden about 2 years ago - 3 comments
Labels: module: onnx, triaged, module: regression

#113037 - `torch.div` on empty tensors causes segmentation fault

Issue - State: open - Opened by malfet about 2 years ago - 5 comments
Labels: high priority, module: crash, module: cpu, triaged, module: regression

#112941 - Segmentation Fault on torch.div with Empty Tensor Dimensions

Issue - State: open - Opened by uruasxpi about 2 years ago - 3 comments
Labels: module: crash, module: error checking, triaged, module: regression, module: edge cases

#112305 - `_foreach_add` segfaults when passed tensors with different lengths

Issue - State: closed - Opened by Chillee about 2 years ago - 6 comments
Labels: high priority, module: crash, triaged, module: regression

#111859 - Ignore beartype if its version is 0.16.0

Pull Request - State: closed - Opened by BowenBao about 2 years ago - 3 comments
Labels: module: regression, open source, Merged, ciflow/trunk, release notes: onnx

#111695 - Runnings SentenceTransformer encoding step causes Docker containers on Mac (Silicon) to crash with code 139

Issue - State: open - Opened by sabaimran about 2 years ago - 2 comments
Labels: high priority, module: crash, triaged, module: mkldnn, module: regression, module: arm

#111251 - Version 2.1 breaks torch.equal when tensors contain nans

Issue - State: open - Opened by Erotemic about 2 years ago - 2 comments
Labels: triage review, module: bc-breaking, module: NaNs and Infs, module: regression

#110980 - DISABLED test_jvp_linalg_det_singular_cpu_float32 (main.TestOperatorsCPU)

Issue - State: closed - Opened by malfet about 2 years ago - 8 comments
Labels: triaged, module: macos, module: regression, skipped, module: functorch

#110597 - [PyTorch 2.1 regression] TorchScript behavior changed from 2.0.1 (and older) to 2.1

Issue - State: open - Opened by fxmarty about 2 years ago - 16 comments
Labels: high priority, triage review, oncall: jit, module: regression

#109819 - ValueError: only one element tensors can be converted to Python scalars

Issue - State: open - Opened by Atcold about 2 years ago - 8 comments
Labels: needs reproduction, triaged, module: regression

#109312 - pytorch-2.1.0 aarch64 wheels crash on A1/Raspberry Pie

Issue - State: open - Opened by malfet about 2 years ago - 8 comments
Labels: high priority, triage review, oncall: binaries, module: regression, module: arm

#108474 - CNN w variable sized input performance regression 1.10.2 cu113 -> 2.0.1 cu117

Issue - State: open - Opened by jxtps about 2 years ago - 4 comments
Labels: module: performance, module: nn, triaged, module: regression

#108442 - Torch compile generates incorrect graph on Llama model

Issue - State: closed - Opened by kkontny about 2 years ago - 2 comments
Labels: triaged, module: regression, oncall: pt2, module: dynamic shapes

#107955 - PyTorch profile issues summary

Issue - State: closed - Opened by npuichigo about 2 years ago - 4 comments
Labels: triage review, module: regression, oncall: profiler

#107396 - [regression] Not getting `CUDA error: device-side assert triggered` on main for CUDA_KERNEL_ASSERT2

Issue - State: closed - Opened by mikaylagawarecki over 2 years ago - 9 comments
Labels: module: cuda, triaged, module: regression

#106051 - `torch.conj()` slice assignment not working

Issue - State: open - Opened by OverLordGoldDragon over 2 years ago - 10 comments
Labels: high priority, good first issue, triaged, module: complex, module: regression, module: partial aliasing, actionable, module: correctness (silent), module: viewing and reshaping

#103117 - Observed regress in DataLoader spawn from PyTorch1.13 to PyTorch2.0

Issue - State: open - Opened by boryiing over 2 years ago - 5 comments
Labels: high priority, module: performance, triaged, module: regression, module: data

#103056 - [Pytorch 2.0] torch::nn::Dropout output is incorrect on Windows

Issue - State: closed - Opened by mszhanyi over 2 years ago - 3 comments
Labels: module: binaries, module: windows, module: cpp, triaged, module: regression

#100974 - Pytorch 2.0.1 pypi wheel does not install dependent cuda libraries

Issue - State: open - Opened by Martin4R over 2 years ago - 3 comments
Labels: triage review, oncall: binaries, module: regression, needs design

#100690 - python3.10 windows wheels >=1.12.0 silently requires numpy>=1.23.0

Issue - State: open - Opened by jvesely over 2 years ago - 9 comments
Labels: high priority, triage review, oncall: binaries, module: windows, module: regression

#100253 - profiler.export_stacks doesn't return stack trace unless experimental_config is provided

Issue - State: open - Opened by NicolayP over 2 years ago - 20 comments
Labels: high priority, module: regression, oncall: profiler

#99981 - Group Norm crashes on Apple M1/MPS devices for versions 2.0+

Issue - State: closed - Opened by KE7 over 2 years ago - 9 comments
Labels: needs reproduction, triaged, module: regression, module: mps

#96153 - mps bug: failed assertion `[MPSNDArrayDescriptor sliceDimension:withSubrange:] error: subRange.start (6) is not less than length of dimension[0] (6)'

Issue - State: closed - Opened by hchau630 over 2 years ago - 18 comments
Labels: high priority, triaged, module: regression, module: viewing and reshaping, module: mps

GitHub / pytorch/pytorch issues and pull requests