transformers
858d4a26 - Optimize flash attention max seqlen computation in vision attention (#47170)

Commit
124 days ago
Optimize flash attention max seqlen computation in vision attention (#47170) * Optimize flash attention max seqlen computation * Use shared max seqlen precomputation helpers * Refine max seqlen precomputation flow * Address flash max seqlen review feedback * Refine max sequence length fallback handling * Share max sequence length utility
Author
Parents
Loading