mahmoodlab / CONCH

A vision-language foundation model for computational pathology - Nature Medicine
Other
281 stars 22 forks source link

Token Embedding Support in CONCH #21

Closed bryanwong17 closed 8 hours ago

bryanwong17 commented 1 day ago

Hi, I was wondering if CONCH can return token embedding similar to how CLIP returns token embeddings using clip_model.token_embedding

The output I expect is of shape[N, max_tokens, 512], but instead I get [N, 512] when using conch_model.encode_text

Thank you in advance!