solsticestudioai/coding-intel-pack

这是一个名为Coding Agent MCTS Reasoning Trace Pack (Sample)的合成数据集，专为自主编码代理设计，包含蒙特卡洛树搜索（MCTS）推理轨迹。每一行代表一个完整的推理生命周期，标注了推理表型（如TEST_DRIVEN、HACKER等）并携带UCB分数和奖励。数据集包含10,000条推理轨迹，按推理表型、任务类型、语言和生产影响层级分层。该数据集适用于基于MCTS的编码代理架构训练、过程奖励模型训练等相关研究。数据集以CC BY 4.0许可证发布，提供Parquet和JSONL格式。

The Coding Agent MCTS Reasoning Trace Pack (Sample) is a synthetic dataset designed for autonomous coding agents, featuring Monte Carlo Tree Search (MCTS) reasoning traces. Each row represents a complete reasoning lifecycle, labeled with reasoning phenotypes (e.g., TEST_DRIVEN, HACKER) and carrying UCB scores and rewards. The dataset includes 10,000 reasoning traces, stratified by reasoning phenotypes, task types, languages, and production impact tiers. It is useful for MCTS-based coding agent architecture training, process reward model training, and other related research. The dataset is released under the CC BY 4.0 license and is available in Parquet and JSONL formats.