{ "base": "general_knowledge_explanation_sft_12k_backup", "continuation": "10k strong-filtered MMLU-Pro-CoT", "evaluation": { "max_new_tokens": 16000, "do_sample": false, "system_prompt": "You are a careful multiple-choice reasoning model.", "heldout_100": { "boxed_accuracy": 0.77, "fallback_accuracy": 0.79, "boxed_rate": 0.98, "clean_boxed_rate": 0.98, "format_gap": 0.02 } } }