Freeze the model, train the harness: gains transfer across LLMs and benchmarks

• Posted 2 hours ago by megadragon9
• 1 points