search

unslothai / unsloth

Finetune Llama 3.2, Mistral, Phi, Qwen 2.5 & Gemma LLMs 2-5x faster with 80% less memory
https://unsloth.ai
Apache License 2.0
18.41k stars 1.29k forks source link

Potential bugfix in FlexAttention #1217

Closed AdityaKane2001 closed 3 weeks ago

AdityaKane2001 commented 3 weeks ago

Fixed potential incorrect flag assignment HAS_FLEX_ATTENTION. I am not aware if this is intentional (to prevent flaky code from running) or a genuine bug.

Erland366 commented 3 weeks ago

Yeah I think it's still disabled for now image

AdityaKane2001 commented 3 weeks ago

Got it.