Search results

reports of torch.compile'd rmsnorm numerics different from eager

module: inductor

oncall: pt2

vllm-compile

Status: Open.

#158699 In pytorch/pytorch;

· zou3519 opened on Jul 19, 2025

[fsdp2] Possible to fuse all-reduce (in fp32) + dtype conversion (to bf16)

Status: Open.

#158698 In pytorch/pytorch;

· vadimkantorov opened on Jul 19, 2025

DISABLED test_sum_all_cpu_float64 (__main__.TestReductionsCPU)

oncall: pt2

skippedDenotes a (flaky) test currently skipped in CI.Denotes a (flaky) test currently skipped in CI.

Status: Open.

#158664 In pytorch/pytorch;

· clee2000 opened on Jul 18, 2025

Autocast + bfloata16 + onnx.export: fp32 scalar tensor -> complex128

Status: Open.

#158658 In pytorch/pytorch;

· ShijunK opened on Jul 18, 2025

extend torch.util.checkpoint to support offload to cpu/nvme

Status: Open.

#158657 In pytorch/pytorch;

· stas00 opened on Jul 18, 2025

Remove tuple of ints as supported dtype for torch.max()

actionable

module: docsRelated to our documentation, both in docs/ and docblocksRelated to our documentation, both in docs/ and docblocks

module: python frontendFor issues relating to PyTorch's Python frontendFor issues relating to PyTorch's Python frontend

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.

#158645 In pytorch/pytorch;

· aalbayrak394 opened on Jul 18, 2025

torch.compile() produces inconsistent outputs with models containing dropout and in-place ReLU operations

oncall: pt2

Status: Open.

#158643 In pytorch/pytorch;

· zhouxiaoyaozzz opened on Jul 18, 2025

PyTorch Dynamo Error Report: Tensor. item() Operation Causes Graph Compilation Interruption

Status: Open.

#158642 In pytorch/pytorch;

· zhouxiaoyaozzz opened on Jul 18, 2025

support checkPoolLiveAllocations for cudaMallocAsync

module: cudaRelated to torch.cuda, and CUDA support in generalRelated to torch.cuda, and CUDA support in general

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.

#158641 In pytorch/pytorch;

· spawner1145 opened on Jul 18, 2025

Allowing <code>optimizer.step(..., *, set_grad_to_none=True)</code> (explicit kwarg opt-in, seems should not break existing code)

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

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.

#158638 In pytorch/pytorch;

· vadimkantorov opened on Jul 18, 2025

nansum crash on complex32 cuda non-JIT

module: cudaRelated to torch.cuda, and CUDA support in generalRelated to torch.cuda, and CUDA support in general

needs reproductionSomeone else needs to try reproducing the issue given the instructions. No action needed from userSomeone else needs to try reproducing the issue given the instructions. No action needed from user

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.

#158635 In pytorch/pytorch;

· pzzp opened on Jul 18, 2025

RandomSampler docs wrongly states type for data_source is Dataset, when it is Sized

module: datatorch.utils.datatorch.utils.data

module: docsRelated to our documentation, both in docs/ and docblocksRelated to our documentation, both in docs/ and docblocks

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.

#158631 In pytorch/pytorch;

· Parskatt opened on Jul 18, 2025