why can not see visual encoder weights?
#1
by
lucasjin
- opened
why can not see visual encoder weights?
This comment has been hidden
Our vision encoder is not finetuned, so it is loaded directly from openai clip vit large 336px.
liuhaotian
changed discussion status to
closed
Our vision encoder is not finetuned, so it is loaded directly from openai clip vit large 336px.
Can we fine tune them ? Because it seems imp is inspired on LLava and prevents us to train the vision tower https://huggingface.co/MILVLG/imp-v1-3b/discussions/9.