edbeeching/fineproofs-gpt-oss-120b

FineProofs GPT-OSS 120B是一个与FineProofs兼容的训练快照数据集，使用openai/gpt-oss-120b模型生成。数据集包含1382行数据，时间戳为2026-05-27T10:04:24Z。其模式与lm-provers/FineProofs-SFT完全匹配，包括问题、推理内容、证明、类别、竞赛、gemini-3-pro-grade、qwen3-4b-thinking-reward@128、来源和消息等字段。GPT-OSS Harmony输出被重新格式化为Qwen风格的助手消息结构。仅包含完整停止生成且非空推理和证明的数据。评分和奖励列存在但留空，因为这些生成的证明尚未评分。

FineProofs GPT-OSS 120B is a FineProofs-compatible training snapshot generated with the openai/gpt-oss-120b model. The dataset contains 1382 rows with a snapshot timestamp of 2026-05-27T10:04:24Z. Its schema exactly matches lm-provers/FineProofs-SFT, including fields such as problem, reasoning_content, proof, category, competition, gemini-3-pro-grade, qwen3-4b-thinking-reward@128, source, and messages. GPT-OSS Harmony outputs are reformatted into Qwen-style assistant messages. Only complete generations with finish_reason=stop and non-empty reasoning and proof are included. The grade and reward columns are present for schema compatibility but left null as these generated proofs have not been graded.