fix: Use SDPA instead of flash_attention_2 for ROCm/hipblas environments#9155
Open
localai-bot wants to merge 2 commits intomudler:masterfrom
Open
fix: Use SDPA instead of flash_attention_2 for ROCm/hipblas environments#9155localai-bot wants to merge 2 commits intomudler:masterfrom
localai-bot wants to merge 2 commits intomudler:masterfrom
Conversation
- Detect ROCm environments using torch.version.hip or HIP_VISIBLE_DEVICES - When running on ROCm, use 'sdpa' attention implementation instead of 'flash_attention_2' - This fixes the issue where FlashAttention2 is not available in hipblas Docker images - Fixes issue mudler#9138 Signed-off-by: localai-bot <localai-bot@noreply.github.com>
- Move is_rocm detection from inline in cuda block to class initialization - Use self.is_rocm for cleaner code and single detection point - Simplify the attention implementation selection logic - Fixes review comment about device detection being in wrong location Signed-off-by: Team Coding Agent 1 <team-coding-agent-1@localai.dev> Signed-off-by: localai-bot <localai-bot@noreply.github.com>
Contributor
Author
DCO Compliance AchievedThis PR replaces #9147 which was closed due to DCO issues (missing Signed-off-by trailers). Changes made:
Ready for review. /cc @mudler |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This PR fixes issue #9138 where FlashAttention2 is not available in hipblas Docker images.
Problem
When using the hipblas Docker image with LocalAI, models like qwen3-tts and qwen3-asr fail to load with FlashAttention2 errors.
Solution
All commits have proper Signed-off-by trailers for DCO compliance.
Fixes #9138