Training AI models on their own improved outputs without external scorers

Original: Sharing our latest work on offline self-distillation:

Source: arxiv.org ↗