Skip to content

[AMD] Enable mamba-extra-buffer for Qwen3.5 on ROCm - #30359

Merged
HaiShaw merged 2 commits into
mainfrom
bingxche/test-r195-rocm-mamba-extra-buffer-27141
Jul 15, 2026
Merged

HaiShaw merged 2 commits into
mainfrom
bingxche/test-r195-rocm-mamba-extra-buffer-27141

Conversation

@bingxche

@bingxche bingxche commented Jul 7, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

  • Allow mamba-extra-buffer on ROCm by accepting is_hip() in the extra_buffer platform guard.
  • Update the assertion text so ROCm is listed with the supported FLA backends.
  • This will unblock Qwen/Qwen3.5-397B-A17B nightly test in AMD Nightly CI
  • credit to @ntgiang71096

Testing

image image

CI States

Latest PR Test (Base): ❌ Run #28848756832
Latest PR Test (Extra): ❌ Run #28848756410

@gemini-code-assist

Copy link
Copy Markdown
Contributor

Warning

Gemini encountered an error creating the review. You can try again by commenting /gemini review.

@yctseng0211

yctseng0211 commented Jul 15, 2026 •

Copy link
Copy Markdown
Collaborator
image

result verified

@bingxche bingxche changed the title [AMD] Support mamba extra buffer for Qwen3.5 [AMD] Enable mamba-extra-buffer for Qwen3.5 on ROCm Jul 15, 2026
@HaiShaw
HaiShaw merged commit d36e96c into main Jul 15, 2026
94 of 108 checks passed
@HaiShaw
HaiShaw deleted the bingxche/test-r195-rocm-mamba-extra-buffer-27141 branch July 15, 2026 17:18
ColinZ22 added a commit to ColinZ22/sglang that referenced this pull request Jul 15, 2026
Revert the is_cuda/is_musa/is_npu guard in _mamba_radix_cache_resolution.
main already enables extra_buffer on ROCm via is_hip() (sgl-project#30359); this gate
contradicted that and broke test_model_overrides.

Co-Authored-By: Claude <noreply@anthropic.com>
Chronostasys pushed a commit to MindLab-Research/sglang that referenced this pull request Aug 24, 2026
Co-authored-by: ntgiang71096 <nguyentruonggiang71096@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants