search

haotian-liu / LLaVA

[NeurIPS'23 Oral] Visual Instruction Tuning (LLaVA) built towards GPT-4V level capabilities and beyond.
https://llava.hliu.cc
Apache License 2.0
20k stars 2.2k forks source link

Clean the code #1077

Open luohao123 opened 8 months ago

luohao123 commented 8 months ago

As for time now ,transformers had support most monkey patch in the code, consider suppor tlatest transformers version make it simpler?

zsxm1998 commented 8 months ago

where are these monkey patches?

haotian-liu commented 8 months ago

Hi @luohao123

We've implemented the flash attention and removed that monkey patch: https://github.com/haotian-liu/LLaVA/blob/main/llava/train/train_mem.py

Can you explain what are other monkey patches that have been implemented by Transformers? Thanks.