Skip to content

Pull requests: pytorch/ao

Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort

Pull requests list

Add CuTeDSL backend for FP8 2:4 sparse conversion and linear, on by default CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: not user facing Use this tag if you don't want this PR to show up in release notes sparsity
#4803 opened Aug 20, 2026 by alexsamardzic Collaborator Loading…
Add device-specific tensor registration pattern and XPU MX example ciflow/xpu label used to trigger xpu CI jobs CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: core changes affecting multiple modules, e.g. base config/tensor, observers, quant ops xpu Intel XPU related features
#4800 opened Aug 20, 2026 by xiaowangintel Collaborator Draft
[nvfp4_training] Optimize CuteDSL kernels for NVFP4 training CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4798 opened Aug 20, 2026 by rdspring1 Contributor Loading…
Fix broken example code in the static quantization tutorial CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4794 opened Aug 19, 2026 by amanyagami Loading…
Mask a_s/b_s scale loads in blockwise FP8 training GEMM kernels CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4793 opened Aug 19, 2026 by amanyagami Loading…
Fix axis-swapped bounds mask in _to_fp8_row_major_t transposed store CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4792 opened Aug 19, 2026 by amanyagami Loading…
Fix dtype mismatch in Float8Linear backward when weight scaling is DISABLED CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4791 opened Aug 19, 2026 by amanyagami Loading…
Preserve dropout p / batch_norm eps,momentum across QAT train<->eval switches CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4790 opened Aug 19, 2026 by amanyagami Loading…
Fix constant_fold deleting attrs still referenced by live get_attr nodes CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4789 opened Aug 19, 2026 by amanyagami Loading…
Make autocast dtype detection device-agnostic (not CUDA-only) CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4788 opened Aug 19, 2026 by amanyagami Loading…
Make GPTQ device-agnostic (remove CUDA-only assert and torch.cuda.synchronize) CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4787 opened Aug 19, 2026 by amanyagami Loading…
Raise ImportError (not RuntimeError) when the MPS lib can't load on non-macOS CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4786 opened Aug 19, 2026 by amanyagami Loading…
Remove dead, broken torchao/prototype/attention/fp8_fa3/setup.py CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4785 opened Aug 19, 2026 by amanyagami Loading…
Fix inverted bias flag in swap_conv2d_1x1_to_linear CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4784 opened Aug 19, 2026 by amanyagami Loading…
Fix ZeroPointDomain.NONE branch comparing class instead of variable CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4783 opened Aug 19, 2026 by amanyagami Loading…
Allow DCP to load Float8 MoE training weights CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. meta-exported
#4782 opened Aug 19, 2026 by ivy-zhou Loading…
Remove Pyre mode headers (#4781) CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. meta-exported
#4781 opened Aug 19, 2026 by meta-codesync Bot Loading…
[XPU][CI]Add timeout for single case in Intel GPU CI ciflow/xpu label used to trigger xpu CI jobs CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. xpu Intel XPU related features
#4780 opened Aug 19, 2026 by zxd1997066 Contributor Loading…
Add standalone mxfp8 32x32 Triton cast kernels CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4777 opened Aug 18, 2026 by vkuzo Contributor Loading…
Support passthrough ops for batch norm fusion (#4776) CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. meta-exported
#4776 opened Aug 17, 2026 by DrJessop Loading…
Add opt-in KleidiAI support to BUCK (#4775) CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. meta-exported module: inference quantize_ api inference flow
#4775 opened Aug 17, 2026 by JakeStevens Contributor Loading…
Delete unused GPTQ prototype CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: bc-breaking Use this tag if this PR breaks backward compatibility module: inference quantize_ api inference flow
#4768 opened Aug 17, 2026 by iamzainhuda Contributor Loading…
Delete deprecated NF4 quantization API CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: bc-breaking Use this tag if this PR breaks backward compatibility module: inference quantize_ api inference flow
#4767 opened Aug 17, 2026 by iamzainhuda Contributor Loading…
Fix cvt_packfloat compatibility with nvidia-cutlass-dsl 4.6.0+ CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: training quantize_ api training flow
#4766 opened Aug 17, 2026 by andrewor14 Contributor Loading…
Fix RoundingModeKind incompatible with nvidia-cutlass-dsl 4.6.0+ CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: training quantize_ api training flow
#4765 opened Aug 17, 2026 by andrewor14 Contributor Loading…
ProTip! no:milestone will show everything without a milestone.