Extract Probabilities from the BERT model (NLP), GitHub Repo Available

I need to extract document probabilities from the BERT model. There are many posts talking about implementation. The Github Repo provides docker, so install the repo should not be a problem. For quick starts, this might be helpful: [login to view URL] I have 2 types of input files for two types of speeches (L and Q): one contains sentence_ID and content (T1_Table). The other has 2 columns, each line records the beginning and end of a paragraph: The beginning sentence_ID (starting from 0), end sentence_ID. Please note that the sentence IDs start from 0 (T2_Table). We need to extract the probability of each paragraph for both two types of speech (L and Q) and save the embeddings for each sentence for both L and Q. All files can be found here [login to view URL] Final deliverables: 1. script 2. sentence embeddings with sentence_id for L and Q 3. paragraph probabilities with paragraph_id