search

CircleRadon / TokenPacker

The code for "TokenPacker: Efficient Visual Projector for Multimodal LLM".
148 stars 6 forks source link

load model issue (about vision tower) #10

Closed Eric-is-good closed 2 weeks ago

Eric-is-good commented 2 weeks ago

In sunshine-lwt/TokenPacker-7b-144token, use openai/clip-vit-large-patch14-336, when I load model, appear : image many weight can not use

Eric-is-good commented 2 weeks ago

image

Eric-is-good commented 2 weeks ago

image

when i use the cli.py, what the diff with old llava cli

LiWentomng commented 2 weeks ago

@Eric-is-good

Hello, the above information about loading model is normal. Regarding cli.py, we have not tested it yet and there are some missing. We will update it as soon as possible. You can try the pretrained models by benchmark evaluation, such as GQA.

Eric-is-good commented 2 weeks ago

@LiWentomng Thank you !!!

LiWentomng commented 2 weeks ago

@Eric-is-good Hello, we have updated cli.py, please check it.