metadata
license: mit
The dataset contains approximately 3.5 million texts and dialogues sourced from the "FineWeb-Edu," "Wikipedia," and "TinyChat" datasets—with each text or dialogue comprising up to 100 words—and includes TinyChat data in the <|user|> - <|assistant|> format.
The dataset was used to train VDrontGPT50m-Base3.
It can also be used for fine-tuning or training lightweight models.