PES2O Materials Science Pretraining Corpus 780M

This HuggingFace dataset provides a large-scale pretraining corpus specifically curated for materials science applications. The dataset contains approximately 780 million tokens derived from RefinedWeb sources collected before January 1, 2019. It is designed to support training of language models for materials discovery, molecular dynamics simulation, and computational materials science tasks. The dataset has been downloaded 221 times on HuggingFace.