Skip to content

Pull requests: ROCm/vllm

New pull request New
33 Open 989 Closed
33 Open 989 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

Release/0.23.0
#1058 opened Jul 14, 2026 by kiran-thumma Collaborator • Draft
5 tasks
[ROCm][Kernel] W4A16 skinny GEMM: hand-asm kernel for gfx1151
#1045 opened Jul 1, 2026 by mgehre-amd • Draft
[bench] vit_attention/bench.py: add --backend, default to running all
#1044 opened Jul 1, 2026 by mgehre-amd • Draft
[CI] Re-enable performance test job
#1033 opened Jun 29, 2026 by marcusr-amd • Draft
5 tasks
1
[ROCm] Split skinny_gemms_int8.cu into per-N translation units
#1028 opened Jun 26, 2026 by marcusr-amd • Draft
3 of 5 tasks
1
tune Qwen3-VL-4B prefill unified-attention on gfx1150
#1024 opened Jun 26, 2026 by qingxuamd Loading…
Fix GLM 5.2 mxfp4 MTP loading issue
#1022 opened Jun 25, 2026 by amd-xiaoyu12 Loading…
[ROCm][MoE] W4A16 MoE routing-distribution benchmark suite for the gfx11 prefill GEMM
#1020 opened Jun 25, 2026 by roberteg16 • Draft
455 war room findings
#1001 opened Jun 12, 2026 by jpvillam-amd Loading…
@jpvillam-amd
MoE: Grouped Triton GEMM for TTFT improvements
#970 opened May 26, 2026 by mgehre-amd • Draft
feat: Add NPU+GPU async pipelining for vision-language models
#936 opened May 14, 2026 by liangliangchang • Draft
4 of 5 tasks
10
Hybrid
#918 opened May 4, 2026 by liangliangchang • Draft
5 tasks
1
Auto-build flash-attn wheels on push, upload to S3
#910 opened Apr 30, 2026 by mgehre-amd • Draft
1 task
[ROCm][DSv4] Share AITER decode dequant + fp8-cast buffers across layers (rebased, stacked on #902)
#903 opened Apr 27, 2026 by ChuanLi1101 • Draft
2 of 4 tasks
4
[ROCm][DSv4] Make AITER sparse decode cudagraph-clean (rebased, stacked on #901)
#902 opened Apr 27, 2026 by ChuanLi1101 • Draft
2 of 5 tasks
5
[ROCm][DSv4] AITER-accelerated MLA decode for DeepSeek V4 on MI355X (rebased on tj/dsv4prrebase)
#901 opened Apr 27, 2026 by ChuanLi1101 • Draft
1 of 4 tasks
5
[Do Not Merge] For review purpose: Rocm/aiter mla dsv4 decode cudagraph
#900 opened Apr 26, 2026 by tjtanaavllm • Draft
5 tasks
[ROCm] support topk_softplus for all number of experts
#899 opened Apr 25, 2026 by tjtanaa Loading…
5 tasks
Tune hybrid_triton_w4a16 prefill kernel for gfx1151
#879 opened Apr 15, 2026 by mgehre-amd • Draft
3 tasks done
Enable FLASH_ATTN backend with upstream flash-attn CK on ROCm for decode
#866 opened Apr 10, 2026 by mgehre-amd • Draft
1 task
[feat] Add ROCm ATOM model impl backend
#850 opened Dec 25, 2025 by zejunchen-zejun • Draft
[Triton] Add Fused GEMM A8W8 + Split + Concat to Deepseek FP8 Model
#831 opened Dec 3, 2025 by farlukas • Draft
1 of 3 tasks
[rocm]use aiter triton kernel as triton mha fallback path
#809 opened Nov 14, 2025 by zhuyuhua-v • Draft
1
add aiter fusion pattern for sequence parallel
#781 opened Oct 31, 2025 by zhuyuhua-v • Draft
5 tasks
Streaming logic for fused_exports and MOE
#768 opened Oct 27, 2025 by omuhamma • Draft
5 tasks
@omuhamma
ProTip! Adding no:label will show everything without a label.