FinePhrase is a large-scale synthetic dataset designed for high-quality language modeling, reasoning, and instruction-following tasks. It transforms raw educational web data into structured, instruction-rich formats suitable for training advanced language models.
This dataset has been extensively used in the pre-training pipeline of TNSA models, including:
