#language model fine-tuning

Internal Coherence Maximization (ICM) introduces a novel label-free, unsupervised training framework for large language models, achieving performance on par with human-supervised methods and enabling advanced capabilities without human feedback.