Grouping tasks by similarity boosts fine-tuning of large language models

Original: The Parts Are Greater Than the Sum: Automated Task Sequencing for Efficient Training of Multi-Policy LLMs

Source: arxiv.org ↗