search

zhixiongz / CLIP4CMR

A Comprehensive Empirical Study of Vision-Language Pre-trained Model for Supervised Cross-Modal Retrieval
42 stars 3 forks source link

Can a single RTX 3090 GPU run? #5

Open limycml opened 2 years ago

limycml commented 2 years ago

Hello, May I ask what GPU does your model need to run? Is a single 3090 OK?

FutureTwT commented 2 years ago

I think it should work. (I use a single RTX2080Ti)

zhixiongz commented 2 years ago

Hello, We fixed the parameters of CLIP and only use its encoded features for cross-modal retrieval, so I think your computing resources are sufficient. We also tried training CLIP from scratch, which would require roughly 4 V100s.

Best, Zhixiong