Maximum Sequence Length
#1
by tomaarsen - opened
Hello @EdoardoMosca and co!
The model card mentions both a Context Length of 32,768 tokens, as well as a "Document length" of 512 tokens. Sentence Transformers uses the "max_seq_length": 512, from https://huggingface.co/LiquidAI/LFM2.5-Embedding-350M/blob/main/sentence_bert_config.json#L2 and adopts the 512. Is this indeed the expected behaviour? Just making sure 🤗
- Tom Aarsen
Hey @tomaarsen thanks for flagging this!
512 is indeed the right number. The 32k context length refers to the original LFM2.5 backbone. But I do agree it can be confusing, I'll remove it across cards