Skip to content

Pull requests: ROCm/aiter

New pull request New
509 Open 3,827 Closed
509 Open 3,827 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

fix(custom_all_reduce): use SYSTEM scope + ACQUIRE ordering for cross-device signal loads
#4786 opened Aug 16, 2026 by hekhong-png Loading…
1
[feat] mega stage2 of gfx1250
#4785 opened Aug 16, 2026 by yanboshao Contributor • Draft
1 task
34
MI350 MLA BF16/FP8 decode: support causal=False (msk0 kernels)
#4784 opened Aug 16, 2026 by sajandhy • Draft
1 task
1
[gfx950][FlyDSL] Add direct dense A4W4 MXFP4 GEMM
#4782 opened Aug 16, 2026 by LiuYinfeng01 Contributor • Draft
6 tasks done
1
[gfx950] Retune the small-M tiles and wave counts in the A16W16 fallback config
#4781 opened Aug 16, 2026 by akii96 Contributor Loading…
1
[Bugfix] Handle GroupNorm autocast safely
#4779 opened Aug 15, 2026 by akshatvishu Loading…
5 of 8 tasks
1
1
[gfx1100] Enable RDNA3 in arch allow-list + Triton GEMM A8W8 tuning config
#4778 opened Aug 15, 2026 by okone1995 Loading…
1
perf(mla): chunk the non-FP4 gather_kv_b_proj over KV
#4776 opened Aug 15, 2026 by zejunchen-zejun Contributor Loading…
2
[Attention] Expose paged MQA SplitKV override
#4775 opened Aug 15, 2026 by AMD-yanfeiwang Contributor • Draft
1
[Triton] [Gluon] [GFX12] mxfp8 gemm
#4773 opened Aug 15, 2026 by k50112113 Contributor • Draft
1
[gfx950][FlyDSL] Add dense BF16 x MXFP4 GEMM
#4772 opened Aug 15, 2026 by LiuYinfeng01 Contributor • Draft
5 tasks done
1
[Triton][FMHA] Fused paged-prefill kernel for page_size=1, head_dim 64/256
#4771 opened Aug 15, 2026 by amd-ethany • Draft
1 task done
1
[Bugfix] Include headers necessary for ROCm 10.0.0
#4768 opened Aug 14, 2026 by rjrock Loading…
1 task done
1
1
[Qwen3.8 MXFP4][Tuning] Add MXFP4 fused-MoE configs for gfx950 (TP8)
#4767 opened Aug 14, 2026 by nholmber Contributor Loading…
1 task done
1
[WIP][TRITON][GLUON][GFX950][DSV4] Sparse MLA training backward
#4766 opened Aug 14, 2026 by wangye805 • Draft
6
feat(moe): consume prepared stage1 activation scales
#4762 opened Aug 14, 2026 by JohnQinAMD Contributor Loading…
6
[PERF] Optimize Triton unified attention prefill and decode
#4761 opened Aug 14, 2026 by vorapolsiloai • Draft
5 of 7 tasks
1
[gfx1250][FlyDSL] Update a8w8 gemm tuned config
#4760 opened Aug 14, 2026 by aoli26 Contributor Loading…
1 task
2
Add full 16-tier Qwen3-VL FP4 MoE tuned config for gfx950
#4759 opened Aug 14, 2026 by johannes-graner Contributor Loading…
1 task
4
[Gluon][MLA] Deeper async-copy pipeline in the bh16 stage-1 decode loop
#4758 opened Aug 14, 2026 by amd-ethany Loading…
1 task done
1
[detorch] remove vestigial py_itfs_common.h includes (non-CK); top_k_per_row now torch-free ci:atom
#4756 opened Aug 14, 2026 by amd-ruitang3 Contributor Loading…
1
[Opus MoE] Unify A8W4 metadata and runtime dispatch
#4755 opened Aug 14, 2026 by yifehuan Contributor Loading…
1 task
1
[Bugfix] Disable cpp_itfs sampling on ROCm 10
#4754 opened Aug 14, 2026 by fsx950223 Contributor Loading…
3 tasks done
4
fix: LL proto ar dispatch
#4753 opened Aug 14, 2026 by TennyWang1223 Contributor Loading…
1 task
1
ci: centralize PyTorch test image config
#4750 opened Aug 14, 2026 by gyohuangxin Member Loading…
3
ProTip! no:milestone will show everything without a milestone.