energy-eval-filtered_responses_multichoice_CEIA-RL_energyv2-dpo-offline-GRPO_v3
收藏资源简介:
该数据集包含447个训练样本,专为问答或推理任务设计,特别集成了检索增强生成(RAG)和思维链(THINK)技术。每个样本包含6个文本字段:原始问题(question)、带选项的问题(question_with_choices)、答案键(answerKey)、一个由CEIA-RL、energyv2-dpo-offline-GRPO、RAG、多语言E5大型微调模型、BGE重排序器等多种技术组合生成的复杂输出字段、RAG提示词(prompt_rag)以及该复杂输出字段的上下文信息。所有字段均为长字符串类型。数据集名称和字段命名表明其与能源领域(energyv2)和特定NLP发布版本(cemig-nlp-releases)相关,可能用于评估或训练结合了强化学习、指令微调和检索增强的复杂问答系统。
This dataset contains 447 training samples, specifically designed for question answering or reasoning tasks, with integrated Retrieval-Augmented Generation (RAG) and Chain of Thought (THINK) techniques. Each sample includes 6 text fields: the original question (question), the question with options (question_with_choices), the answer key (answerKey), a complex output field generated by a combination of various technologies such as CEIA-RL, energyv2-dpo-offline-GRPO, RAG, multilingual E5 large fine-tuned model, BGE reranker, the RAG prompt (prompt_rag), and contextual information for that complex output field. All fields are of the long string type. The dataset name and field naming suggest its association with the energy domain (energyv2) and a specific NLP release version (cemig-nlp-releases), and it may be used for evaluating or training complex question-answering systems that combine reinforcement learning, instruction fine-tuning, and retrieval augmentation.
- 数据集名称:energy-eval-filtered_responses_multichoice_CEIA-RL_energyv2-dpo-offline-GRPO_v3
- 数据集链接:https://huggingface.co/datasets/CEIA-RL/energy-eval-filtered_responses_multichoice_CEIA-RL_energyv2-dpo-offline-GRPO_v3
- 数据集大小:9,075,359 字节(约 8.66 MB)
- 下载大小:3,492,986 字节(约 3.33 MB)
- 数据集划分:仅包含训练集 (
train) - 训练集样本数:447 条
- 数据集特征:
question:问题文本(字符串)question_with_choices:包含选项的问题文本(字符串)answerKey:正确答案(字符串)CEIA-RL_energyv2-dpo-offline-GRPO_RAG-TRUE_ENC-cemig-nlp-releases_multilingual-e5-large-ft-v2_RR-cemig-nlp-releases_bge-reranker-v2-m3-ft-v2_THINK-TRUE:模型响应(字符串)prompt_rag:检索增强生成(RAG)提示词(字符串)CEIA-RL_energyv2-dpo-offline-GRPO_RAG-TRUE_ENC-cemig-nlp-releases_multilingual-e5-large-ft-v2_RR-cemig-nlp-releases_bge-reranker-v2-m3-ft-v2_THINK-TRUE_context:RAG 上下文(字符串)
- 数据集配置:
- 配置名称:
default - 数据文件路径:
data/train-*
- 配置名称:




