THUDM

/

cogvlm2-llama3-caption

like 80

Follow

Knowledge Engineering Group (KEG) & Data Mining at Tsinghua University 1.55k

Hi, We want to enable this model on intel xpu platform, which has been supported in stock pytorch. Pls help review, thx!

The video card's memory of the xpu should be insufficient to run this model, how did you manage to run it?
Additionally, I noticed that you seem to have set transformers>=4.44.0 and <=4.44.0, which is very good. However, in the requirements.txt on GitHub, we have already requested to use transformers and above, so this part does not need to be modified.

Hi, The memory of our xpu card is 128GB, so it is OK to run with one single card. For transformers version, we want to add support for latest versions, so that it can run in our project.

I am confused, we did not update this transformers file, for 4.44-4.48, you are right.