CLASS-IT: Conversational and Lecture-Aligned Small-Scale Instruction Tuning Dataset
Dataset Summary
The CLASS-IT dataset provides instruction-tuning material for BabyLM-scale models, designed to investigate how small-scale language models benefit from interaction-driven and curriculum-based instruction tuning.
It contains two complementary components — Simple Wikipedia (instructional) and Switchboard (conversational) — enabling comparison between structured… See the full description on the dataset page: https://huggingface.co/datasets/colinglab/CLASS_IT.
