Don't Repeat Yourself: Self-Supervised Fine-Tuning for Coverage
Improving model reliability in math and coding through enhanced output diversity during post-training.
In verifiable domains such as math and coding, finding one correct solution among many attempts can matter more than the pass rate of each attempt. Post-training can concentrate large language model outputs around a few modes, while increasing sampling temperature has limited eff...
ArXiv CS.CL
Source
Generated insight by OnAirToday AI
This research addresses the critical limitation of model collapse in iterative post-training by shifting focus from individual accuracy to collective output diversity. By effectively covering the solution space, developers can significantly enhance the reliability of AI agents in logic-heavy domains like coding and mathematics.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.