Repository navigation

All pull requests

Labels (24)Milestones (15)New pull request

Open

(96)

Closed

(701)

Declare the test dependencies in requirements_dev.txt
#2039ConnorBaker opened Oct 22, 2026
Remove tests of removed amp, fp16_utils and parallel modules
#2038ConnorBaker opened Oct 22, 2026
[contrib/group_norm] Compute NHWC GroupNorm variance from shifted sums to avoid fp32 cancellation
#2037jayden1711 opened Oct 15, 2026
[pre-commit.ci] pre-commit autoupdate
#2035pre-commit-ci[bot] opened Sep 28, 2026Bot
Honor set_grad_none=False when clearing gradients
#2034sylvesterkaczmarek opened Sep 23, 2026
1
Restore graph lowering after compiler failures
#2033sylvesterkaczmarek opened Sep 23, 2026
1
Prune unstructured weights by magnitude
#2032sylvesterkaczmarek opened Sep 23, 2026
1
Preserve float64 accumulation in RMSNorm fallback
#2031sylvesterkaczmarek opened Sep 23, 2026
1
Replace deprecated torch.cuda.amp API with torch.amp equivalents
#2025xyf5432 opened Aug 20, 20261 task done
fix(contrib): call torch.cuda.is_available() in FP16_Optimizer guard
#2021andrewwhitecdw opened Jul 26, 2026
4
extend 64bit indexing for other multi tensor ops
#2014POI-WX opened Jun 9, 2026
Avoid retaining saved tensors in fused norm custom ops under no_grad
#2012LeSingh1 opened May 31, 2026Contributor
1
Constrain APEX_ASP_CACHE_DIR to a safe base directory
#2011LeSingh1 opened May 31, 2026Contributor
fix missing standard headers in group_norm_v2 host template
#2010ZhiyuanChen opened May 28, 2026Contributor
Use TORCH_LIBRARY*, Make Python ABI Stable
#2009crcrpar opened May 27, 2026Collaborator
2
Guard CUDA probe when CUDA_HOME is unset
#2008fallintoplace opened May 26, 2026
[contrib][test_clip_grad] Increase rtol for fp16 test to 5e-3
#2002crcrpar opened May 12, 2026Collaborator
Group norm v2: add missing include
#1997RemiLacroix-IDRIS opened Apr 27, 2026
Add 64-bit indexing fallback for large multi_tensor_l2norm kernels
#1989SongXiaoXi opened Mar 25, 2026
4
Fix traceback formatting in permutation_lib.py
#1987yujincheng08 opened Mar 17, 2026Member
3
Fixed Win11 install issues of compile flags and isfnite() crashing.
#1977original-doc opened Jan 18, 2026
3
Fix FMHA bwd_nl kernel issue for batch_size == 3
#1975crcrpar opened Jan 14, 2026Collaborator
Fix E731: Convert lambda assignments to def functions
#1965crcrpar with CopilotAI opened Nov 29, 2025
1
Add LibTorch Stable ABI infrastructure (#1946)
#1956jackulau opened Nov 24, 2025
20
fix build python,cpp_ext and cuda_ext error for Visual Studio 2022 on…
#1931huhai463127310 opened Aug 29, 2025

Navigation Menu

Repository navigation

All pull requests

Declare the test dependencies in requirements_dev.txt

Remove tests of removed amp, fp16_utils and parallel modules

[contrib/group_norm] Compute NHWC GroupNorm variance from shifted sums to avoid fp32 cancellation

[pre-commit.ci] pre-commit autoupdate

Honor set_grad_none=False when clearing gradients

Restore graph lowering after compiler failures

Prune unstructured weights by magnitude

Preserve float64 accumulation in RMSNorm fallback

Replace deprecated torch.cuda.amp API with torch.amp equivalents

fix(contrib): call torch.cuda.is_available() in FP16_Optimizer guard

extend 64bit indexing for other multi tensor ops

Avoid retaining saved tensors in fused norm custom ops under no_grad

Constrain APEX_ASP_CACHE_DIR to a safe base directory

fix missing standard headers in group_norm_v2 host template

Use `TORCH_LIBRARY*`, Make Python ABI Stable

Guard CUDA probe when CUDA_HOME is unset

[contrib][test_clip_grad] Increase rtol for fp16 test to 5e-3

Group norm v2: add missing include

Add 64-bit indexing fallback for large multi_tensor_l2norm kernels

Fix traceback formatting in permutation_lib.py

Fixed Win11 install issues of compile flags and isfnite() crashing.

Fix FMHA bwd_nl kernel issue for batch_size == 3

Fix E731: Convert lambda assignments to def functions

Add LibTorch Stable ABI infrastructure (#1946)

fix build python,cpp_ext and cuda_ext error for Visual Studio 2022 on…

Pull requests