Skip to content

Pull requests: pytorch/ao

New pull request New
253 Open 2,935 Closed
253 Open 2,935 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

Expand Triton autotune configs for MoE FP8 kernels to improve AMD GPU performance
#3952 opened Feb 25, 2026 by brucechanglongxu Loading…
3 tasks done
6
Fix operator precedence bug in is_Navi4() GPU detection
#3951 opened Feb 25, 2026 by brucechanglongxu Loading…
3 tasks done
2
Add better debug print for failed prepare 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
#3950 opened Feb 25, 2026 by JakeStevens Loading…
2
[mxfp8 training] remove mxfp8 from MXLinear and MXLinearConfig 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 mx
#3949 opened Feb 25, 2026 by danielvegamyhre Loading…
1
[mxfp8 training] unified MXFP8TrainingConfig and MXFP8TrainingTensor 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
#3948 opened Feb 25, 2026 by danielvegamyhre Loading…
1
Add FA4 fp8 backend to 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. topic: new feature Use this tag if this PR adds a new feature
#3947 opened Feb 25, 2026 by howardzhang-cv • Draft
1
Add RunningAbsMaxSmoothQuantObserver for memory-efficient calibration 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
#3946 opened Feb 25, 2026 by jcaip Loading…
2
Use relaxed memory ordering for Triton atomics on AMDGPU. ciflow/rocm 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 topic: improvement Use this tag if this PR is an improvement (doesn't fit into any of the other categories) topic: performance Use this tag if this PR improves the performance of a feature
#3945 opened Feb 25, 2026 by wenchenvincent Loading…
4
[ROCm] Enable float8 utils and base tests on ROCm module: rocm
#3942 opened Feb 24, 2026 by brucechanglongxu Loading…
2
[ROCm] Enable affine quantized, quant API, and integration tests on ROCm module: rocm
#3941 opened Feb 24, 2026 by brucechanglongxu Loading…
2
[ROCm] Remove redundant ROCm test skips in NF4 and use accelerator-ag… module: rocm
#3940 opened Feb 24, 2026 by brucechanglongxu Loading…
2
[ROCm] Enable low-bit optimizer tests on ROCm module: rocm
#3939 opened Feb 24, 2026 by brucechanglongxu Loading…
3
lowbit linear weight packing and shared embedding on x86 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
#3938 opened Feb 24, 2026 by JacobSzwejbka Loading…
2
Add MXFP4 support for GPTQ quantization and separate eval script CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#3935 opened Feb 23, 2026 by jcaip Loading…
1
[WIP] [mxfp8 moe training] add triton kernel for per group padding CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#3933 opened Feb 22, 2026 by danielvegamyhre Loading…
1
Added benchmark for LLaMA 3 model for attention tests benchmark 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
#3930 opened Feb 21, 2026 by howardzhang-cv • Draft
5
Added benchmark for single attention layer across different sequence lengths benchmark 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
#3929 opened Feb 21, 2026 by howardzhang-cv • Draft
1
Add Static Activation Quantization Subclass for Observation 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
#3925 opened Feb 20, 2026 by namgyu-youn Loading…
4 tasks
2
[mxfp8 moe training] migrate ep utils to mx_formats 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 topic: improvement Use this tag if this PR is an improvement (doesn't fit into any of the other categories)
#3922 opened Feb 20, 2026 by danielvegamyhre Loading…
1
Add MXFP4 support for GPTQ quantization and separate eval script CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#3921 opened Feb 19, 2026 by jcaip Loading…
1
Add support for flashinfer quantize kernel option for nvfp4 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
#3912 opened Feb 17, 2026 by jerryzh168 Loading…
7
Refactor use_triton_kernel to use nvfp4_quantize_kernel_choice 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
#3911 opened Feb 17, 2026 by jerryzh168 Loading…
20
Add asymmetric support for Int8Tensor + SmoothQuant 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
#3900 opened Feb 17, 2026 by jcaip Loading…
10
refactor GPTQ observer for dynamic quant CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
#3899 opened Feb 17, 2026 by jcaip Loading…
1
[mxfp8 moe training] move mxfp8 grouped mm code into mx_formats 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 topic: improvement Use this tag if this PR is an improvement (doesn't fit into any of the other categories)
#3898 opened Feb 17, 2026 by danielvegamyhre Loading…
4
ProTip! What’s not been updated in a month: updated:<2026-01-25.