lihaoxin2020/agentic-search-refiner-shortform-curated-v1

该数据集名为Agentic Search Refiner Shortform Curated v1，是一个经过整理的短格式问答混合数据集，用于`/home/nvidia/workspace/dr-tulu/rl/open-instruct/run_train_refiner_shortform_api_agent.sh`。数据集已标准化为ASearcher兼容的schema，包含`question`（用户查询）、`answer`（精确参考答案）、`source`、`subset`、`source_id`和`metadata`等字段。数据集由多个子集组成，总计1500行数据，包括asearcher_lrm_multihop（300行）、asearcher_lrm_webwalker_chain（300行）、webaggregatorqa_train（300行）、webexplorerqa（100行）和webshaper（500行）。每个子集都有特定的筛选和去重标准。数据集的预期用途是作为短格式API-agent refiner RL运行的单一JSONL/Hugging Face数据集。

The dataset is named Agentic Search Refiner Shortform Curated v1 and is a curated short-form QA mixture for `/home/nvidia/workspace/dr-tulu/rl/open-instruct/run_train_refiner_shortform_api_agent.sh`. The dataset is normalized to the ASearcher-compatible schema, including fields such as `question` (user query), `answer` (exact reference answer), `source`, `subset`, `source_id`, and `metadata`. The dataset consists of multiple subsets, totaling 1500 rows, including asearcher_lrm_multihop (300 rows), asearcher_lrm_webwalker_chain (300 rows), webaggregatorqa_train (300 rows), webexplorerqa (100 rows), and webshaper (500 rows). Each subset has specific filtering and deduplication criteria. The intended use of the dataset is as a single JSONL/Hugging Face dataset in the short-form API-agent refiner RL run.