Hi everyone ! I am trying to do manual full embeddings generate from my python for

sentence-transformers/distiluse-base-multilingual-cased-v2

and when I tried to get the output it is different for a simple word

Hello

. I tried onnx it gave me 768 dimension output however when I tried the sentence-transformer version it gives me 512 dimension so I am kind of confused why is it like that ?

Difference is Sentence transformers version has an additional dense layer that maps 768 dim embedding outputs to 512 dim. https://huggingface.co/sentence-transformers/distiluse-base-multilingual-cased-v2#full-model-architecture