💡 [REQUEST] - 能调用backbone抽feature吗？ about qwen-vl HOT 2 CLOSED

Hi, @jyC23333,

1. No, without fine-tuning, you can not use the model directly on Image-Text retrieval.
2. Our ViT model is the same as openclip except we drop the class token. You can directly load the model. The details of the model are on hugginface.

from qwen-vl.

@simonJJJ Hi, Thanks a lot.

1. As mentioned, I can't regard the whole model as image encoder + text encoder, is it?
2. does it make any sense to extract image feature using ViT and to extract text feature using LLM backbone?
   Because after alignment between the two kinds of features, I think the model has learned the relationship of the two domain.

from qwen-vl.