salesforce / query-focused-sum Goto Github PK

Hello there, thank you for a great paper and piece of work!

I tried to train multiencoder, but when I try to get raw data from https://github.com/Yale-LILY/QMSum, it seems to have a slightly different format

Failed to run preprocess.py, missing meeting_id and meeting_transcripts expects list of str but the oroginal data has list of dict

I can hack around and change to format to introduce dummy meeting_id and make it look as expected but I wanted to first check if I am missing something or if there is an cleaner way to do so.

Question is: before running preprocess.py should one just get the jsonl files from https://github.com/Yale-LILY/QMSum or is there additional and different data expected beyond a simple transform to the original data?

Thank you in advance!

I didn't find the preprocessing code for generating {split}.rouge.256.jsonl in this repo, did i miss it?

Hi, thanks for the great work!

Are the preprocessed AquaMuse documents (and summaries) also available for download, similar to the QMSum dataset? It would be really helpful for us to get identical datasets (and save the overhead of processing the data from commoncrawl).

Thanks!