Skip to content

Pull requests: ROCm/aiter

New pull request New
553 Open 3,977 Closed
553 Open 3,977 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

[Gluon] Move fp4 GEMM gluon to _gluon_kernels
#4997 opened Aug 25, 2026 by vgokhale Contributor • Draft
1
[Bugfix] JIT: use torch's bundled pybind11 to avoid ABI (internals-version) skew
#4996 opened Aug 25, 2026 by Rohan138 Contributor • Draft
1
1
[ASM] fix(fmoe): add missing XQ scale barrier in a16 FP8-blockscale kernels
#4995 opened Aug 25, 2026 by alexioslyrakis-amd Contributor Loading…
3 tasks done
1
[AMD][DSV4] Fix(fmoe): fuse stage-1 fp8 quant on the heuristic FlyDSL fallback
#4994 opened Aug 25, 2026 by karverma-amd Contributor Loading…
1
[Triton/Gluon] Fix PR#4562 and PR#4470 for GPT-OSS-120b on gfx1250
#4993 opened Aug 25, 2026 by sogalin Contributor Loading…
1 task
1
[FlyDSL] gfx950 hd=72 varlen FMHA for Qwen3-VL prefill
#4992 opened Aug 25, 2026 by msaffari-amd • Draft
9
Fix inverse rope group quant gfx1250
#4991 opened Aug 25, 2026 by yzhou103 Contributor • Draft
1 task
1
[HIP] [ROCm] Widen the AR+RMSNorm reduce-scatter producer
#4990 opened Aug 25, 2026 by EricKing626 • Draft
1
[FlyDSL] comm fused moe
#4985 opened Aug 25, 2026 by yifehuan Contributor • Draft
1 task
1
feat(mega_moe/gfx1250): quantize before dispatch, on an fp8 or fp4 wire
#4984 opened Aug 25, 2026 by jhchouuu Contributor Loading…
1
[Config] config(glm5): restore MXFP4 MoE token=16/32/64/128/8192 to a4w4
#4983 opened Aug 25, 2026 by Jacob0226 Contributor Loading…
2
[FlyDSL] mega all gather merge stage1
#4981 opened Aug 25, 2026 by Bernard-Liu Contributor • Draft
1 task
@Bernard-Liu
10
[FlyDSL] [CI] [MegaMoE] A4W4 mega_moe operator multigpu
#4980 opened Aug 25, 2026 by Yaowu-Xiong Loading…
1 task
3
[Feature] Block-sparse MHA v4 forward with load-balanced work-table dispatch
#4979 opened Aug 25, 2026 by nsakkine Contributor • Draft
2 tasks done
[Triton/Gluon] [HIP] Dev lumen
#4978 opened Aug 25, 2026 by WuLei-AMD Loading…
16 of 17 tasks
1
[HIP] [Feat] Single-kernel Lamport fused all-reduce + RMSNorm (opt-in)
#4977 opened Aug 25, 2026 by EricKing626 • Draft
1
[Build] [Feature] Add MK1 persistent decoder provider
#4975 opened Aug 24, 2026 by ssharma4-amd • Draft
2
[HIP] Fix (custom_all_reduce): correct output RankData slot prediction during graph capture
#4974 opened Aug 24, 2026 by afriedri Loading…
2
[ASM] [HIP] [CK] feat(mha): gfx950 hd256 FP8 LINEAR paged-varlen asm prefill
#4971 opened Aug 24, 2026 by MiloLurati Loading…
1 task done
@MiloLurati
1
[FlyDSL] QRInt4: INT4 two-shot all-reduce for gfx942/gfx950
#4970 opened Aug 24, 2026 by samremes Contributor • Draft
3 tasks
11
[Triton/Gluon] [PA] Support Q/K D192 with asymmetric K/V head dim
#4968 opened Aug 24, 2026 by sammysun0711 Contributor Loading…
1 task done
10
[Triton/Gluon] [ASM] [HIP] MHA v4: support GQA, add gfx950 bf16, add gfx942 i8/fp8 ci:all ci:atom
#4967 opened Aug 24, 2026 by jcaraban Contributor Loading…
1 task done
3
[FlyDSL] Prevent duplicate JIT tracing on concurrent kernel startup
#4966 opened Aug 24, 2026 by xytpai Contributor Loading…
1
[HIP] topk_gating: fix dropped and out-of-range top-k slots
#4965 opened Aug 24, 2026 by yzhou103 Contributor Loading…
1 task
16
FIX MLA the nhead fold error for cp round robin
#4964 opened Aug 24, 2026 by minmengdie Contributor Loading…
1 task
7
ProTip! What’s not been updated in a month: updated:<2026-07-25.