Repository navigation
[AMD] Enable mamba-extra-buffer for Qwen3.5 on ROCm - #32552
Closed
nemanjaudovic wants to merge 27 commits into
Closed
nemanjaudovic wants to merge 27 commits into
nemanjaudovic wants to merge 27 commits into
Conversation
… NVFP4 online quantization (sgl-project#30397) (sgl-project#30424) Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca> Co-authored-by: Brayden Zhong <brayden@radixark.ai>
… NVILA weight loading (sgl-project#30400) (sgl-project#30425) Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca> Co-authored-by: Brayden Zhong <brayden@radixark.ai>
…TP: clamp padded-row seq_lens to >= 0 (sgl-project#30378) (sgl-project#30427) Co-authored-by: DarkSharpness <76582120+DarkSharpness@users.noreply.github.com> Co-authored-by: ziyi.xu <ziyi.xu@radixark.ai> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
…er by default for large prefill chunks (sgl-project#30140) (sgl-project#30436) Co-authored-by: YAMY <74099316+YAMY1234@users.noreply.github.com>
…ensor contiguous (sgl-project#27926) (sgl-project#30449) Co-authored-by: Kaixi <matteochen3@gmail.com> Co-authored-by: liqichao <liqichao@baidu.com> Co-authored-by: chenbong <bhchen@stu.xmu.edu.cn>
…#30439) (sgl-project#30465) Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
…te all output slots on tie overflow (sgl-project#30512) (sgl-project#30559) Co-authored-by: Jimmy Shong <69131491+Jiminator@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…te dispatch input for precision job (sgl-project#30495) (sgl-project#30566) Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
…indices under tie overflow / inf scores (IMA in FA3 sparse decode) (sgl-project#30645) (sgl-project#30698) Co-authored-by: DarkSharpness <76582120+DarkSharpness@users.noreply.github.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
…sgl-project#30699) (sgl-project#30704) Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
…E to main-first to avoid CUDA graph stream explosion (sgl-project#30460) (sgl-project#30714) Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Jiminator <jimmysh341@gmail.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… scale layout (sgl-project#29275) (sgl-project#30724) Co-authored-by: sonle5 <51179712+hdt98@users.noreply.github.com> Co-authored-by: sunxxuns <126995791+sunxxuns@users.noreply.github.com>
…h on gfx950 (sgl-project#29479) (sgl-project#30725) Co-authored-by: billishyahao <bill.he@amd.com> Co-authored-by: HAI <hixiao@gmail.com>
…t kernel in decode graph when not recording (sgl-project#30302) (sgl-project#30726) Co-authored-by: Rita Brugarolas <Rita.BrugarolasBrufau@amd.com>
…okens (sgl-project#30313) (sgl-project#30727) Co-authored-by: Bingxu Chen <bingxche@amd.com> Co-authored-by: YC Yen-Ching Tseng <yctseng@amd.com>
…DA-graph capture crash under torch_memory_saver (sgl-project#30557) (sgl-project#30730) Co-authored-by: Zhiyao Jiang <jessicajiang324@gmail.com> Co-authored-by: Xinyu Jiang <xinyuj2@andrew.cmu.edu>
…error (sgl-project#30374) (sgl-project#30729) Co-authored-by: Wang, FangYuan <39615225+At1a8@users.noreply.github.com>
…trtllm long-context "!!!!" collapse (NaN routing) (sgl-project#31001) (sgl-project#31065) Co-authored-by: Khoa Pham <khoa.pham@radixark.ai>
…oss PD and CUDA graph replay (sgl-project#30839) (sgl-project#31083) Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
…th prefill CP (sgl-project#30992) (sgl-project#31106) Co-authored-by: zRzRzRzRzRzRzR <Yuxuan.Zhang2@liverpool.ac.uk>
Contributor
|
Caution The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased. |
Signed-off-by: nemanjaudovic <nudovic@amd.com>
sogalin
requested review from
1am9trash,
Alisehen,
AniZpZ,
BBuf,
ByronHsu,
Duyi-Wang,
Edwardf0t1,
FlamingoPg,
Fridge003,
HaiShaw,
Kangyan-Zhou,
OrangeRedeng,
ShangmingCai,
YAMY1234,
b8zhong,
bingxche,
ch-wan,
hebiao064,
hubertlu-tw,
ishandhanani,
ispobock,
kkHuang-amd,
mmangkad,
rainj-me,
sogalin,
xiezhq-hermann,
yctseng0211,
yichiche and
yizhang2077
as code owners
September 15, 2026 02:52
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
10 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Ports upstream #30359 to amd_march: adds
is_hip()to the_validate_mamba_extra_bufferplatform check so hybrid-mamba models (Qwen3.5 / Qwen3-Next)can use the extra_buffer radix-cache strategy on ROCm instead of asserting at startup. The
required kernels (FLA chunk-scan, selective_state_update) are Triton and AMD-aware; the assert
was a validated-platform whitelist, not a hard limit.
CI States
Latest PR Test (Base): ❌ Run #34828715619
Latest PR Test (Extra): ❌ Run #34828715349
Latest PR Test (AMD ROCm 10): ➖ No AMD PR run found for this commit.