Running Featured 1.31k FineWeb: decanting the web for the finest text data at scale π· 1.31k Generate a curated webβtext dataset for LLM training