The script lora_video_infer.py expects a file named non_lora_trainables.bin to be present in the model_path. However, I could not find any instructions or references on where to obtain or generate this file.
I am running lora_video_infer.py with --pretrain_path=baichuan-inc/Baichuan-7B (HF hub).
Thanks for sharing and very cool work! I have a couple of suggestions to improve the codebase:
Update Hardcoded Model Location
In the file get_blip_fea.py, could you change the hardcoded location of the model? The current hardcoded path doesn't specify which model is being used. For example:
In the file process_video.py, the code expects the directory images/video_cuts which does not exist by default. Could you please add the following code to ensure the directory is created if it doesn't exist?