Skip to content

Pull requests: pytorch/ao

New pull request New
287 Open 3,047 Closed
287 Open 3,047 Closed
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
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Delete deprecated Float8Layout and Float8AQTTensorImpl 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
#4127 opened Mar 20, 2026 by jerryzh168 Loading…
3
Delete deprecated CutlassSemiSparseLayout and related code 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
#4126 opened Mar 20, 2026 by jerryzh168 Loading…
4
Fix test_sparse_api failures for builds without hipSPARSELt (#4125) CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. fb-exported meta-exported module: not user facing Use this tag if you don't want this PR to show up in release notes
#4125 opened Mar 20, 2026 by bbeckca Loading…
3
[WIP]torchao support npu fp8 CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4124 opened Mar 20, 2026 by frankleitong • Draft
1
Delete deprecated autoquant v1 and all references 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
#4122 opened Mar 20, 2026 by jerryzh168 Loading…
3
[sam2] Replace complex-number RoPE with PyTorch F.apply_rotary_emb CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4118 opened Mar 19, 2026 by jainapurva • Draft
1
Fix float8 operand precision validation CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4115 opened Mar 19, 2026 by GotFusion Loading…
1
[moe training] Optimize triton_fp8_per_group_colwise_scales for AMDGPU ciflow/rocm ciflow/4xh100 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
#4113 opened Mar 19, 2026 by wenchenvincent Loading… FP8 Rowwise Training
3
[ROCm] Enable MXFP8 MoE training kernel and grouped GEMM tests on ROCm CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: rocm
#4110 opened Mar 18, 2026 by brucechanglongxu Loading…
4 tasks done
2
Enable Float8 dynamic activation quantization on Navi4 CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4108 opened Mar 18, 2026 by zhangnju Loading…
2
[X86] intmm: Use u8s8 when only support avx512-vnni 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
#4103 opened Mar 18, 2026 by cyxlily Loading…
14
enable fp8 qlinear patterns for AOT Inductor
#4099 opened Mar 17, 2026 by CaoE • Draft
2
Fix aten._has_compatible_shallow_copy_type', overload='default' dispatcher failed for LinearActivationQuantizedTensor in inference mode bug Something isn't working 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
#4098 opened Mar 17, 2026 by xiaowangintel Loading…
1
1
[xpu][feat] Add XPU support for blockwise FP8 CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. quantize_ quantize_ API topic: for developers Use this tag if this PR is mainly developer facing xpu Intel XPU related features
#4096 opened Mar 17, 2026 by Stonepia • Draft
3
[ROCm] Enable Int4 weight-only quantization (TILE_PACKED_TO_4D) on ROCm CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: rocm
#4092 opened Mar 16, 2026 by brucechanglongxu Loading…
2
Fix _is_128_128_scaled false positive for PerTensor on 128x128 linears CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4091 opened Mar 16, 2026 by brucechanglongxu Loading…
1
2
[WIP] npu support: int4 and fp8 CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4087 opened Mar 16, 2026 by frankleitong • Draft
2
Fuse expert grouping and padding into the MXFP8 EP dispatch path 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 moe mx
#4086 opened Mar 14, 2026 by MagellaX Loading… MXFP8 Training
@MagellaX
14
speed up mxfp8 rceil quantization kernel CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: inference quantize_ api inference flow
#4073 opened Mar 13, 2026 by vkuzo Loading…
3
[moe training] Optimize FP8 MoE backward pass: fused colwise kernel + AMD tuning CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. float8 module: training quantize_ api training flow moe
#4069 opened Mar 13, 2026 by lizamd Loading…
3 tasks
FP8 Rowwise Training
@lizamd
18
Use in-place ops in _quantize_affine_float8 to reduce peak memory (#4068) CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. fb-exported meta-exported module: inference quantize_ api inference flow
#4068 opened Mar 12, 2026 by r3t2 Loading…
3
[mxfp8 moe training] fused all2all dispatch + token group padding + expert major kernel 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
#4066 opened Mar 12, 2026 by danielvegamyhre Loading… MXFP8 Training
@danielvegamyhre
1
soften version guard check for low precision attention API CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#4057 opened Mar 11, 2026 by howardzhang-cv Loading…
1
[xpu][test] Skip WIP config for Intel GPU in test_safetensors_support.py and test_x86inductor_fusion.py 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
#4049 opened Mar 11, 2026 by zxd1997066 Loading…
8
[xpu][test] Port test/prototype/quant_logger/test_quant_logger.py to Intel GPU 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: inference quantize_ api inference flow
#4046 opened Mar 11, 2026 by zxd1997066 Loading…
4
ProTip! Follow long discussions with comments:>50.