ResearchBenchmarksData Generation on GSM8K, Live Code Bench, and Dolly15k 53K labelsFollow8.5Generation Time (hours)SelfJudge4.0434.14564.2594.355Sep 26, 2025Evaluation ResultsMethodMethodLinksGeneration Time (hours)SelfJudgeTarget/Draft Model=Lla...Target/Draft Model=Llama-3.1-70B/8B, Hardware=4x H100 GPUs2025.098.5AutoJudgeTarget/Draft Model=Lla...Target/Draft Model=Llama-3.1-70B/8B, Hardware=4x H100 GPUs2025.09120