You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I know that legacy uses openaiclip's github and main branch uses clip_trt's github. I used main branch to make coco's vector.bin and the search perfomance is very bad. I would like to ask where I didn't notice it?
The text was updated successfully, but these errors were encountered:
Hello!, yes I recall there being some question of projected vs non (this is unrelated to another mm_projector from VLM) and it resulting in embeddings of dimension 1024 vs 768. clip_trt is using HuggingFace for CLIP/SigLIP vs the legacy branch used openai directly, so will have to double-check exactly which weights are being used to restore the performance building new indexes.
I know that legacy uses openaiclip's github and main branch uses clip_trt's github. I used main branch to make coco's vector.bin and the search perfomance is very bad. I would like to ask where I didn't notice it?
The text was updated successfully, but these errors were encountered: