LPDG-results
收藏资源简介:
本数据集是LPDG任务运行后生成的结果集合文件,以JSON Lines (.jsonl) 格式存储。数据内容整合了多种基线方法与推理策略在特定测试场景下的输出结果,具体包括:SelfRAG/CRAG基线模型的答案、基于GRPO奖励的消融推理结果、结合知识图谱(KG)的query_to_cot(思维链)生成结果,以及这些方法在噪声鲁棒性测试下的表现结果。该数据集主要用于对比分析不同方法(如检索增强生成、奖励建模、知识增强推理等)在应对噪声或进行复杂推理时的性能差异。文件目录结构保留了原始实验代码(src/xiugai)的组织方式。
This dataset is a collection of result files generated upon completion of the LPDG task, stored in JSON Lines (.jsonl) format. It integrates the outputs of multiple baseline methods and reasoning strategies across specific test scenarios, specifically including: the answers from SelfRAG/CRAG baseline models, ablation study results based on GRPO rewards, query_to_cot (Chain-of-Thought, CoT) generation results combined with Knowledge Graph (KG), and the performance results of these methods during noise robustness tests. This dataset is primarily used for comparative analysis of performance differences between various methods (such as Retrieval-Augmented Generation, Reward Modeling, Knowledge-Augmented Reasoning, etc.) when coping with noise or conducting complex reasoning. The file directory structure follows the organization of the original experimental code (src/xiugai).
数据集概述
- 数据集名称:LPDG 跑完的结果文件
- 数据集来源:Hugging Face(https://huggingface.co/datasets/lfqian/LPDG-results)
内容描述
该数据集包含以下实验结果文件(JSONL格式):
- SelfRAG/CRAG 基线答案
- GRPO reward 消融推理 结果
- KG query_to_cot 结果
- 噪声鲁棒性 结果
目录结构
保留自 src/xiugai 目录结构。




