feat: Use SDPA as default attention for broader compatibility#149
Open
xander1421 wants to merge 1 commit intomultimodal-art-projection:mainfrom
Open
feat: Use SDPA as default attention for broader compatibility#149xander1421 wants to merge 1 commit intomultimodal-art-projection:mainfrom
xander1421 wants to merge 1 commit intomultimodal-art-projection:mainfrom
Conversation
…ibility Replace flash_attention_2 with sdpa (Scaled Dot Product Attention) as the default attention implementation. This provides better compatibility across different environments: - Works on systems without flash-attn installed - Compatible with Python 3.14+ - No additional dependencies required (uses PyTorch native SDPA) Users who have flash-attn installed can still modify this to use flash_attention_2 for potentially better performance. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
flash_attention_2withsdpa(Scaled Dot Product Attention) as the default attention implementation ininference/infer.pyMotivation
The current default requires
flash-attnto be installed, which:SDPA (Scaled Dot Product Attention) is:
Changes
attn_implementation="flash_attention_2"toattn_implementation="sdpa"in both stage1 and stage2 model loadingTesting
Tested inference successfully with:
Users who prefer flash_attention_2 can still modify this locally for potentially better performance on supported systems.
🤖 Generated with Claude Code