Giters

dvlab-research / LISA

Project Page for "LISA: Reasoning Segmentation via Large Language Model"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

how to merge lora weight when there are more than 1 model.bin

hxx-who opened this issue · comments

commented

Hi, great work!
I want to Merge LoRA Weight of LISA-7B by using the following code

python merge_lora_weights_and_save_hf_model.py \
  --version="./LLaVA/LLaVA-Lightning-7B-v1-1"\
  --weight="PATH_TO_pytorch_model.bin" \
  --save_path="./LISA7B"

but the --weight only receives the path to one model.bin，but the model of lisa7B has two as pytorch_model-00001-of-00002.bin and pytorch_model-00002-of-00002.bin，how can I merge them?

commented

I aim to train on the base of LISA-7B, and it seems like there's no need to merge.
I tried to set the '''--version''' as the model path of LISA-7B and follow the debug instruction in #85, it finally works though #85 also doesn't know why changing that line would be useful.