Skip to content

Pull requests: ROCm/aiter

New pull request New
430 Open 3,135 Closed
430 Open 3,135 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

Feat/gfx942 flydsl mxfp4 moe
#3926 opened Jun 25, 2026 by msaffari-amd • Draft
1 task
1
feat: support flydsl all2all
#3924 opened Jun 25, 2026 by JiaoliangYu • Draft
1 task
1
change default pa reduce kernel from cxx to flydsl ci:atom
#3923 opened Jun 25, 2026 by Bernard-Liu Contributor Loading…
3
[OPUS][ATOM] gfx1250 (wave32/WMMA) NoPE-fp8/RoPE-bf16 sparse prefill (separate module)
#3922 opened Jun 25, 2026 by carlushuang Collaborator Loading…
feat: update A8W8 MLA kernels to global-load ckv variant on gfx950
#3921 opened Jun 25, 2026 by fangche123 Contributor Loading…
1 task
2
[Bugfix] base_tuner: gfx-aware tuned CSV handling and column-safe row…
#3920 opened Jun 25, 2026 by yzhou103 Contributor Loading…
1 task
7
feat(gfx1151): allow gfx1151 in cpp_itfs JIT arch validation
#3919 opened Jun 25, 2026 by carlushuang Collaborator Loading…
1
Fix aot deadlock
#3918 opened Jun 25, 2026 by zhiding512 Contributor Loading…
1 task
6
feat(gfx1151): add INT8 W8A8 GEMM default config
#3917 opened Jun 25, 2026 by carlushuang Collaborator Loading…
1
CI: run vLLM DSv4 and MiniMax M3 on MI350X ci:vllm
#3916 opened Jun 25, 2026 by gyohuangxin Member Loading…
2
perf(unified_attention): use 8 warps for gfx1151 3D decode
#3915 opened Jun 25, 2026 by carlushuang Collaborator Loading…
1
[Triton] Optimize MoE
#3914 opened Jun 25, 2026 by vgokhale Contributor Loading…
1
flydsl: gfx942 FP8 MQA logits indexer kernel (+ Triton FN/FNUZ fix)
#3913 opened Jun 25, 2026 by akii96 Contributor • Draft
1 task
3
configs(fmoe): add expert=256/topk=8 MXFP4 tuned entries for DeepSeek-V3/R1 (gfx950)
#3911 opened Jun 24, 2026 by Rohan138 Contributor • Draft
1
Fix gfx950 fp8 persistent MLA folded dispatch ci:sglang
#3910 opened Jun 24, 2026 by hubertlu-tw Contributor Loading…
1
CI: run downstream tests on DO runners ci:all
#3908 opened Jun 24, 2026 by gyohuangxin Member Loading…
2
[FlyDSL] [gfx1250] gather moe support
#3907 opened Jun 24, 2026 by XingerZhu Contributor Loading…
1 task
6
[Triton] Add sigmoid score_mode to flat topk routing path
#3902 opened Jun 24, 2026 by leonling-ll Contributor • Draft
1
[FLYDSL] add MLA reduce decode kernel for gfx942
#3901 opened Jun 24, 2026 by anhminhnguyenhoang Contributor • Draft
@anhminhnguyenhoang
1
Satya/unify scale weight shuffling ci:all
#3900 opened Jun 24, 2026 by Boss2002n Contributor Loading…
@Boss2002n
2
[Communication] add quickreduce int3
#3899 opened Jun 24, 2026 by haoyangli0109 Contributor Loading…
1
[gfx1250][FLYDSL]moe gemm tune
#3897 opened Jun 24, 2026 by Zzz9990 Contributor • Draft
1 task
1
Fix HIP fp8 paged-attention kPerHead scale OOB page fault.
#3896 opened Jun 24, 2026 by JohnNikolay84 Contributor Loading…
1 task
@JohnNikolay84
1
[feat][gluon][mla] support MTP for mla
#3890 opened Jun 24, 2026 by yanxuer-999 Loading…
1 task
1
[MoE] Add swiglu_oai (OAI SwiGLU) for per-token fp8 CK XDL 2-stage MoE
#3886 opened Jun 24, 2026 by LJ-underdog Contributor Loading…
1 task
3
ProTip! Add no:assignee to see everything that’s not assigned.