Slow inference
#3
by floschne - opened
Hi and thanks for you M-CLIP publications ! :)
I'm just trying this model and compared to the multilingual models from sentence-transformers or the large laion xlm-roberta models, the inference performance is very slow. I.e., about 61x slower than 'sentence-transformers-clip-ViT-B-32-multilingual-v1' and about 4x slower than the 'laion/CLIP-ViT-H-14-frozen-xlm-roberta-large-laion5B-s13B-b90'.
Do you have an idea, how to boost performance? I'm running the models on a RTX A 6000 (50GB)