How do we segment documents?

#1
by Kaguya-19 - opened

I am an undergraduate student at Tsinghua University and am currently working on embedded models. Your work GritLM is exciting, the publicly available Medi2 dataset is useful, and the ablation experiments in it are very helpful.

I am currently training a text embedding model using Medi2 and I find that documents in our dataset do not seem to be segmented by sentences. How do we segment documents? Can you provide processing code?

Sign up or log in to comment