haotian-liu / LLaVA

[NeurIPS'23 Oral] Visual Instruction Tuning (LLaVA) built towards GPT-4V level capabilities and beyond.

Home Page:https://llava.hliu.cc

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

[Question] How to use the pretrain checkpoint

hellangleZ opened this issue · comments

Question

I only found that there are some file like:

image

How can I merge them to base model , or something I should do, any help is very appreciated

commented

I have the same question...

You can use the load_pretrained_model function. Check here.

You need to:

  1. pass a model-base argument, which is the base language model you used for your llava model.
  2. pass a model-path argument, which is where your pretrained weights' folder is.

This way, you can load the pretrained weight into a base llava model.

Is it possible to use the pretrained checkpoints with SGLang?