NanoStories Releases!
retained text
It is a Tinystories style dataset, that contains 50M stories, about 25x that of Tinystories, it was generated using a rule based algorithm, instead of a language model. It took just under 10 Minutes to generate and a few minutes to convert to parquet and upload to huggingface on a phone CPU. Any questions, feedback, or criticism is welcome!
document fetched 2026-10-24 10:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1xdc991/nanostories_releases