fix(qwen-tts): install flash-attn on CUDA 13 - #11251
Open
localai-org-maint-bot wants to merge 4 commits into
Open
fix(qwen-tts): install flash-attn on CUDA 13#11251localai-org-maint-bot wants to merge 4 commits into
localai-org-maint-bot wants to merge 4 commits into
Conversation
Add the missing cublas13 post-install requirements hook so the CUDA 13 backend uses FlashAttention instead of falling back to SDPA. Keep CUDA packaging parity covered by the backend test script. Assisted-by: Codex:gpt-5
Limit the CUDA 13 FlashAttention source build to one ninja worker by default so it stays within CI runner memory. Preserve explicit MAX_JOBS overrides and leave other build profiles unchanged. Assisted-by: Codex:gpt-5 [web]
Keep the CUDA 13 FlashAttention packaging check tied to its serialized source-build guard so the runner OOM fix cannot be removed unnoticed.\n\nAssisted-by: Codex:gpt-5 [Codex]
localai-org-maint-bot
force-pushed
the
bot/issue-9293-qwen-tts-cuda13
branch
from
August 1, 2026 03:03
1952ca2 to
7916c48
Compare
The CUDA 13 FlashAttention build still exhausts hosted-runner memory with a single ninja worker because nvcc can compile multiple threads internally. Limit nvcc to one thread for that profile and guard the setting in the backend test script. Assisted-by: Codex:gpt-5
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
This PR fixes the confirmed CUDA 13 qwen-tts packaging defect in #9293. The shared backend installer already loads a profile-specific
requirements-<profile>-after.txt, but qwen-tts only provided that hook for CUDA 12. Adding the CUDA 13 counterpart installsflash-attnafter PyTorch, avoiding the reported ImportError and SDPA fallback.The backend test script now checks both supported CUDA profiles retain the post-install dependency hook. The separate vLLM-omni failure reported in the issue is intentionally out of scope because no current traceback is available.
Notes for Reviewers
Verification run locally:
bash -n backend/python/qwen-tts/test.sh backend/python/qwen-tts/install.shpython3 -m py_compile backend/python/qwen-tts/backend.py backend/python/qwen-tts/test.pygit diff --checkThe full backend image build was not run locally because it is a long GPU packaging build; CI should exercise that path.
Signed commits