Search results

Voxtral transcription example produce gibbersih on MPS device

has workaround

module: correctness (silent)issue that returns an incorrect result silentlyissue that returns an incorrect result silently

module: mpsRelated to Apple Metal Performance Shaders frameworkRelated to Apple Metal Performance Shaders framework

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#162092 In pytorch/pytorch;

· malfet opened on Sep 3, 2025

torch.arange not implemented for complex numbers

enhancementNot as big of a feature, but technically not a bug. Should be easy to fixNot as big of a feature, but technically not a bug. Should be easy to fix

has workaround

module: complexRelated to complex number support in PyTorchRelated to complex number support in PyTorch

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#160629 In pytorch/pytorch;

· springcrane opened on Aug 14, 2025

Compile error with gcc-14 when building vec_test_all_types

actionable

has workaround

module: buildBuild system issuesBuild system issues

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#159962 In pytorch/pytorch;

· jsmolic opened on Aug 6, 2025

autograd precision: float32 gradients of bfloat16 weights

has workaround

module: autogradRelated to torch.autograd, and the autograd engine in generalRelated to torch.autograd, and the autograd engine in general

needs designWe want to add this feature but we need to figure out how firstWe want to add this feature but we need to figure out how first

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#158709 In pytorch/pytorch;

· martin-marek opened on Jul 19, 2025

NCCL + cudagraphs + expandable segments result in IMA

has workaround

module: dependency bugProblem is not caused by us, but caused by an upstream library we useProblem is not caused by us, but caused by an upstream library we use

module: ncclProblems related to nccl supportProblems related to nccl support

oncall: distributedAdd this issue/PR to distributed oncall triage queueAdd this issue/PR to distributed oncall triage queue

Status: Open.

#158029 In pytorch/pytorch;

· ngimel opened on Jul 10, 2025

RuntimeError: operator torchvision::nms does not exist

has workaround

low priorityWe're unlikely to get around to doing this in the near futureWe're unlikely to get around to doing this in the near future

module: hub

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#157648 In pytorch/pytorch;

· ralphrmartin opened on Jul 4, 2025

Can't alter the learning rate though loading optimizer states

has workaround

module: LrScheduler

module: optimizerRelated to torch.optimRelated to torch.optim

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#156846 In pytorch/pytorch;

· Alex17Li opened on Jun 25, 2025

Change error to warnning when tensor size mismatch while loading params

has workaround

module: nnRelated to torch.nnRelated to torch.nn

needs researchWe need to decide whether or not this merits inclusion, based on research worldWe need to decide whether or not this merits inclusion, based on research world

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#155368 In pytorch/pytorch;

· PistonY opened on Jun 6, 2025

matmul uses excessive memory in batch cases with more than 3 dimensions

actionable

has workaround

matrix multiplication

module: memory usagePyTorch is using more memory than it should, or it is leaking memoryPyTorch is using more memory than it should, or it is leaking memory

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#154128 In pytorch/pytorch;

· JackBuck opened on May 22, 2025

CMake Error (CMAKE_MINIMUM_REQUIRED) when build 2.7.0

has workaround

module: buildBuild system issuesBuild system issues

module: third_party

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#154096 In pytorch/pytorch;

· fbezruchko opened on May 22, 2025

undefined symbol: __nvJitLinkCreate_12_8, version libnvJitLink.so.12

has workaround

module: binariesAnything related to official binaries that we release to usersAnything related to official binaries that we release to users

needs reproductionEnsure you have actionable steps to reproduce the issue. Someone else needs to confirm the repro.Ensure you have actionable steps to reproduce the issue. Someone else needs to confirm the repro.

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#152783 In pytorch/pytorch;

· FurkanGozukara opened on May 4, 2025

RMS norm causes NaNs when used with torch.compile + float8 with rowwise scales

has workaround

high priority

module: correctness (silent)issue that returns an incorrect result silentlyissue that returns an incorrect result silently

module: floatx (formerly float8)For torch.float8_e5m2 and torch.float8_e4m3 and other sub 8-bit float typesFor torch.float8_e5m2 and torch.float8_e4m3 and other sub 8-bit float types

module: inductor

module: norms and normalization

months

oncall: pt2

triagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Status: Open.

#150859 In pytorch/pytorch;

· danielvegamyhre opened on Apr 8, 2025