search

snap-research / MyVLM

Official Implementation for "MyVLM: Personalizing VLMs for User-Specific Queries" (ECCV 2024)
https://snap-research.github.io/MyVLM/
Other
148 stars 9 forks source link

GPU question. #6

Open woonsango opened 2 months ago

woonsango commented 2 months ago

Thank you for submitting a good paper. I have three questions regarding gpu

Is there a way to make the code work using multiple GPUs?

What GPU was it trained on?

Did you use a model that applied quantization when learning concept embedding?

yuval-alaluf commented 3 days ago

Everything runs with a single 80-GB A100 GPU. I don't think it the code can work on multiple GPUs, but if you have 80GBs, this should work fine with a single GPU. The code provided in the repo is the exact code we used for all experiments. We did not use quantization, but you can try, although it may effect the results.