This repo contains the evaluation datasets for our paper Adapting Large Language Models via Reading Comprehension.
We explore continued pre-training on domain-specific corpora for large language models. While this approach enriches LLMs with domain knowledge, it significantly hurts their prompting ability for question answering. Inspired by human learning via reading comprehension, we propose a simple method to… See the full description on the dataset page:
https://huggingface.co/datasets/AdaptLLM/law-tasks.
