search

zzxslp / SoM-LLaVA

[COLM-2024] List Items One by One: A New Data Source and Learning Paradigm for Multimodal LLMs
123 stars 3 forks source link

Attention map extraction #4

Open yi-ming-qian opened 6 months ago

yi-ming-qian commented 6 months ago

Hello, thanks for sharing the work, it is very inspiring. I wonder if you can share the attention extraction and visualization script used for creating Figure 2 in the paper?

zzxslp commented 5 months ago

Hi! We use the code in this repo. Currently swamped with other things but can share the probing code for LLaVA later.