formal-umd/T-SHIRT

T-SHIRT: Token-Selective Hierarchical Data Selection for Instruction Tuning Yanjun Fu, Faisal Hamman, Sanghamitra Dutta [📖 Paper] Overview Abstract Instruction tuning is essential for Large Language Models (LLMs) to effectively follow user instructions. To improve training efficiency and reduce data redundancy, recent works use LLM-based scoring functions, e.g., Instruction-Following Difficulty (IFD), to select high–quality instruction-tuning data with scores… See the full description on the dataset page: https://huggingface.co/datasets/formal-umd/T-SHIRT.