annon124816/mmlu

该数据集包含用于论文工件的Ergon-native分片rollout-card导出，专注于MMLU（大规模多任务语言理解）任务中的提示/提取比较行。证据类型为一步追踪，仅包含一步提示/响应证据，不涉及完整的追踪过程。数据集旨在支持相关研究中的比较分析，基于上游MMLU和harness发布，并具有特定的元数据分类，如重分发类为metadata-plus-fetch，公共证据类为one_step_completion_trace。This dataset contains Ergon-native sharded rollout-card exports for the paper artifact, focusing on prompt/extraction comparison rows for the MMLU (Massive Multitask Language Understanding) task. The evidence type is one-step trace, which includes one-step prompt/response evidence and is not a full rollout trace. It is designed to support comparative analysis in related research, based on upstream MMLU and harness releases, with specific metadata classifications such as redistribution class as metadata-plus-fetch and public evidence class as one_step_completion_trace.