salesforce / LAVIS

LAVIS - A One-stop Library for Language-Vision Intelligence
BSD 3-Clause "New" or "Revised" License
9.6k stars 942 forks source link

Why do I always encounter CUDA out of memory problem when I load the load_model_process function? Can the RTX 3090 be used for the BLIP-2 model?" #670

Open zhangmenghuan-mh opened 5 months ago

zhangmenghuan-mh commented 5 months ago

Why do I always encounter CUDA out of memory problem when I load the load_model_process function? Can the RTX 3090 be used for the BLIP-2 model?"

Thomas2419 commented 5 months ago

Yes, I'm using mine for that right now. More information may be needed to help, I know transformers version is an extremely common issue plaguing people. Specifically I'm using it for image captioning though so perhaps some nuance on your task may be in play here.