search

dvlab-research / LLaMA-VID

LLaMA-VID: An Image is Worth 2 Tokens in Large Language Models (ECCV 2024)
Apache License 2.0
693 stars 43 forks source link

Demo Crash #25

Closed QiSu77 closed 9 months ago

QiSu77 commented 9 months ago

Hi, when i using the demo page your great team provided, I find that even a video which is more than 1min will leads to the page crash. Which is as same as what happens when I doing inference, what is strange is that there is no error reported on my terminal. So I think it maybe something general, did you ever meet it? Could you please try it on the demo page provided to have more know about it

wcy1122 commented 9 months ago

Hi. I have deleted some files on my local server and restarted the online demo. In my test, the online demo looks fine now for more than 1 minute video. Maybe you can try again on our online demo.

QiSu77 commented 9 months ago

@wcy1122 Oh, you are my hero! Can you tell me more details about the files you deleted? Where are they in the Structure which the official provides, and what are their names. Our team is very interested in your excellent work and wants to reproduce your work on our sever. Thank you very much!

wcy1122 commented 9 months ago

Hi, thanks for your interest in our work. I found that the code will dump uploaded videos to /tmp/gradio in my local server, but the memory of which is limited. I just deleted these files and restarted the demo. Not sure if you face the same problem.

QiSu77 commented 9 months ago

Thank you ! I want to know where can I download "llama-vid-vicuna-7b-short" model. I just find "llama-vid-7b-full-224-video-fps-1" and "llama-vid-13b-full-224-video-fps-1" for short video.

wcy1122 commented 9 months ago

Hi. "llama-vid-vicuna-7b-short" is the same model as "llama-vid-7b-full-224-video-fps-1", we rename the folder.

QiSu77 commented 9 months ago

Thank you very much!