Research
Research paper proposes data-free self-training for large language models
The method has models generate their own question-answer pairs and reasoning traces from high-level topic descriptions, with no human-curated data. The authors report gains across text and visual question answering benchmarks.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
On-Policy Self-Training: Data-Free Self-Improvement in Large Language Models