16/05/2023 I found that doing downsampling isn't a good idea. The accuracy is very low (25%) and the loss is more than one (>1) in every epoch. Maybe there are some loss information. So I'll try to ...

LLM-based pragmatic QA (CooperativeRAG) with comparison to an extractive DistilBERT baseline. A project for the BGU NLP with LLMs course. - mayanim/nlp-llms-pragmaticqa ...