search

CircleRadon / TokenPacker

The code for "TokenPacker: Efficient Visual Projector for Multimodal LLM".
141 stars 5 forks source link

report bugs #3

Closed Gaffey closed 1 month ago

Gaffey commented 1 month ago

https://github.com/CircleRadon/TokenPacker/blob/305ce146ec8b6d8b5ec4959f6cac699e7c8b9ed4/llava/train/train.py#L320

There is a comment on Line 321 which makes the indented block wrong. Maybe there should be a comment on Line 320?

Gaffey commented 1 month ago

And there is no arguments in train.py named down_rate. Is that a bug?

Gaffey commented 1 month ago

https://github.com/CircleRadon/TokenPacker/blob/305ce146ec8b6d8b5ec4959f6cac699e7c8b9ed4/llava/model/multimodal_projector/builder.py#L54

Undefined name grid. Maybe it should be raw_grid?

daixiangzi commented 1 month ago

https://github.com/CircleRadon/TokenPacker/blob/305ce146ec8b6d8b5ec4959f6cac699e7c8b9ed4/llava/model/multimodal_projector/builder.py#L54

Undefined name grid. Maybe it should be raw_grid?

yes,i also find the bug,i think grid is raw_grid

CircleRadon commented 1 month ago

@Gaffey @daixiangzi Sorry for mistakes, we have fixed it. Please try our latest code!