Skip to content

Pull requests: AMD-AGI/Primus-Turbo

New pull request New
22 Open 483 Closed
22 Open 483 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/mxfp8 topk grad blowup
#529 opened Sep 22, 2026 by Xiaoming-AMD Collaborator • Draft
12 tasks
fix: flydsl grouepd gemm kernel racing condition bug ci:gpu
#528 opened Sep 22, 2026 by RuibinCheung Collaborator Loading…
5 of 12 tasks
1
fix: triton grouped gemm kernel work steal numerical issue on v26.6 ci:gpu
#527 opened Sep 22, 2026 by RuibinCheung Collaborator Loading…
5 of 12 tasks
1
Dev/kyle/flydsl flash attn meta
#526 opened Sep 22, 2026 by wenxie-amd Collaborator • Draft
12 tasks
Optimize FlyDSL MXFP4 grouped GEMM on gfx950
#523 opened Sep 21, 2026 by suachong Collaborator • Draft
4 of 12 tasks
Add option to enforce single compute stream in case it blocks communication ci:gpu
#520 opened Sep 18, 2026 by lorri-rao Loading…
feat(flydsl): add standalone QKV RoPE fwd+bwd for Llama-3.1-8B ci:gpu
#519 opened Sep 17, 2026 by kylasa Contributor Loading…
12 tasks
14
feat(attn): running-max softmax for FlyDSL BF16 dualwave forward ci:gpu
#517 opened Sep 16, 2026 by kylasa Contributor Loading…
12 tasks
6
[WIP]perf(flydsl): MXFP4 dense GEMM optimization ci:gpu
#514 opened Sep 15, 2026 by kyle-256 Collaborator Loading…
6 of 12 tasks
19
flydsl attn: cover the D=64/D=128 backward's uncovered issue windows ci:gpu
#512 opened Sep 15, 2026 by kyle-256 Collaborator Loading…
4 of 12 tasks
4
opt(gemm): reduce MXFP4 dense dispatch overhead ci:gpu
#504 opened Sep 13, 2026 by suachong Collaborator Loading…
14
feat(gemm): add a gfx1250 FlyDSL GEMM for bf16 / fp8 / mxfp8
#499 opened Sep 8, 2026 by JohnQinAMD Collaborator Loading…
12 tasks
27
fix(grouped-gemm): cap FP8 BLOCK_K at 64 on gfx1250 ci:gpu
#477 opened Aug 24, 2026 by JohnQinAMD Collaborator Loading…
12 tasks
10
Add torch-compatible flex_attention entry backed by Primus-Turbo ci:gpu
#472 opened Aug 21, 2026 by botaohu001 Contributor Loading…
12 tasks
fix: loosen flydsl install requires on gfx1250 ci:gpu
#467 opened Aug 20, 2026 by RuibinCheung Collaborator Loading…
5 of 12 tasks
2
feat(moe): support HIP and Triton backends for moe_permute, default to Triton ci:gpu
#455 opened Aug 12, 2026 by zhenhuang12 Collaborator Loading…
10 of 12 tasks
7
[WIP]feat(tuning): offline tune framework and configs
#439 opened Jul 30, 2026 by xiaobochen-amd Collaborator Loading…
12 tasks
4
opt(gemm): add hipBLASLt algorithm cache and thread-local workspace
#321 opened Apr 30, 2026 by jasainio Contributor Loading…
6 of 12 tasks
5
Refactor: moe dispatch combine autotune ci:gpu
#312 opened Apr 24, 2026 by zhenhuang12 Collaborator Loading…
7 of 12 tasks
65
feat: enable hybrid FP8 dtypes on Triton grouped GEMM backends
#288 opened Apr 15, 2026 by sarthak-amd • Draft
2
feat: Online tuning for hipblaslt gemm
#277 opened Apr 10, 2026 by Z-Y00 Contributor Loading…
5
refactor: reorganize moe ops and kernels
#243 opened Mar 5, 2026 by zhenhuang12 Collaborator Loading…
1
ProTip! Filter pull requests by the default branch with base:main.