search

IVGSZ / Flash-VStream

This is the official implementation of "Flash-VStream: Memory-Based Real-Time Understanding for Long Video Streams"
https://invinciblewyq.github.io/vstream-page/
Apache License 2.0
129 stars 7 forks source link

Numpy version confliction. Could you please provide a Release Note with configuration requirements? #25

Open robot007 opened 1 week ago

robot007 commented 1 week ago

I got a Numpy version confliction issue. When I run "bash scripts/train_and_eval.sh" the system said I need to download Numpy like the following image. image After I downgrade numpy to 1.24 or 1.23.5, I got following another error. image

There are simply too many versions to try out one by one.

Could you please clarify all python module versions? Could you use "pip freeze" and add the result to the github repo. Of course, a more professional method is to provide a requirements.txt file, where you may define version like: "numpy==x.y.z"

Also, I don't know how much disk space is required, so I failed several times when creating an VM in the cloud. Could you please provide a Release Note with hardware requirements? Thanks a lot.

zhang9302002 commented 6 hours ago

Hello, we are using following package versions in this project. I hope this will help.

torch==2.0.1
torchvision==0.15.2
transformers==4.31.0
numpy==1.26.0
cuda version: 12.1

(Since requirements.txt if more than 300 lines, which is redundant and helpless)

The model checkpoint takes 15GB on your disk, and it takes at least 15GB RAM (half precision) for inference.