Web-Instruct’s Instruction Tuning for MAmmoTH2 and MAmmoTH2-Plus Models: The Power of Web-Mined Data in Enhancing Large Language Models

The Web-Instruct method and the subsequent development of the MAmmoTH2 and MAmmoTH2-Plus models mark significant advances in instruction tuning for large language models. MAmmoTH2 demonstrated a surge in accuracy on complex reasoning tasks, and MAmmoTH2-Plus outperformed base models on standard reasoning benchmarks.