LlavaMistral1 / llava /train /train_xformers.py
lorocksUMD's picture
Upload 78 files
6351d80 verified
raw
history blame
379 Bytes
# Make it more memory efficient by monkey patching the LLaMA model with xformers attention.
# Need to call this before importing transformers.
from llava.train.llama_xformers_attn_monkey_patch import (
replace_llama_attn_with_xformers_attn,
)
replace_llama_attn_with_xformers_attn()
from llava.train.train import train
if __name__ == "__main__":
train()