├── README.md
├── README_zh.md
├── data
├── dataset.json
├── law_library.jsonl
└── samples
│ ├── generated_responses.jsonl
│ ├── retrieval_Qwen2-1.5B.jsonl
│ └── rewrite_question.jsonl
└── src
├── config
├── config.py
└── template
│ └── prompt.txt
├── eval
├── evaluator.py
├── llm_as_judge
│ ├── make_prompt.py
│ ├── run_judge.py
│ └── use_template.py
├── metrics.py
└── retrieval_metrics.py
├── generate
├── data_processor.py
├── generator.py
└── prompt_builder.py
├── pipeline.py
├── process
├── processor.py
└── rewriter.py
├── retrieval
├── dense_retriever.py
├── lexical_matching.py
└── run_retrieval.py
└── utils
└── utils.py
/README.md:
--------------------------------------------------------------------------------
1 |
2 |
3 |
LexRAG: Benchmarking Retrieval-Augmented Generation in Multi-Turn Legal Consultation Conversation
6 |
7 | :book:中文 |
8 | English
9 |
10 | Welcome to LexiT, the dedicated toolkit for RAG in the legal domain.
11 |
12 | ## :link: Introduction
13 | To advance RAG system research in the legal domain, we’ve proposed LexiT, a modular and scalable RAG toolkit for legal researchers. Although there are some general-domain RAG toolkits available, they do not support multi-turn conversations and evaluations tailored to the legal domain. LexiT consists of three components: **Data**, **Pipeline**, and **Evaluation**. It integrates all elements of the RAG process into a unified framework and supports standalone applications. This modular design enhances flexibility and allows for high customizability in evaluating different legal scenarios.
14 |
15 |
16 |
25 |
26 |
2 |
3 |
LexRAG: 多轮法律咨询对话中的检索-增强生成基准测试
6 |
7 | :book:中文 |
8 | English
9 |
10 | 欢迎来到LexiT，一个法律领域的检索增强生成（RAG）专用工具包。
11 |
12 | ## :link: Introduction
13 | 为了推进法律领域的 RAG 系统研究，我们为法律研究人员提出了模块化、可扩展的 RAG 工具包 LexiT。虽然目前已有一些通用领域的 RAG 工具包，但它们并不支持多轮对话和针对法律领域的评估。LexiT 由三个部分组成： ```Data``` ```Pipeline``` ```Evaluation```。LexiT将 RAG 流程的所有要素整合到一个统一的框架中，并支持独立应用。这种模块化设计提高了灵活性，在评估不同的法律情况时具有高度的可定制性。
14 |
15 |
16 |
24 |
25 |