Skip to content

TrustRAG:The RAG Framework within Reliable input,Trusted output

Notifications You must be signed in to change notification settings

gomate-community/TrustRAG

Repository files navigation

TrustRAG

可配置的模块化RAG框架。

Python workflow status codecov pydocstyle PEP8

🔥TrustRAG 简介

TrustRAG是一款配置化模块化的Retrieval-Augmented Generation (RAG) 框架,旨在提供可靠的输入与可信的输出 ,确保用户在检索问答场景中能够获得高质量且可信赖的结果。

TrustRAG框架的设计核心在于其高度的可配置性和模块化,使得用户可以根据具体需求灵活调整和优化各个组件,以满足各种应用场景的要求。

🔨TrustRAG 框架

framework.png

✨主要特色

“Reliable input,Trusted output”

可靠的输入,可信的输出

🎉 更新记录

  • TrustRAG 打包构建,支持pip和source两种方式安装
  • 添加MinerU文档解析 :一站式开源高质量数据提取工具,支持PDF/网页/多格式电子书提取[20240907]
  • RAPTOR:递归树检索器实现
  • 支持多种文件解析并且模块化目前支持解析的文件类型包括:text,docx,ppt,excel,html,pdf,md
  • 优化了DenseRetriever,支持索引构建,增量追加以及索引保存,保存内容包括文档、向量以及索引
  • 添加ReRank的BGE排序、Rewriter的HyDE
  • 添加Judge的BgeJudge,判断文章是否有用 20240711

🚀快速上手

🛠️ 安装

方法1:使用pip安装

  1. 创建conda环境(可选)
conda create -n trustrag python=3.9
conda activate trustrag
  1. 使用pip安装依赖
pip install trustrag   

方法2:源码安装

  1. 下载源码
git clone https://github.com/gomate-community/TrustRAG.git
  1. 安装依赖
pip install -e . 

🚀 快速上手

1 模块介绍📝

├── applications
├── modules
|      ├── citation:答案与证据引用
|      ├── document:文档解析与切块,支持多种文档类型
|      ├── generator:生成器
|      ├── judger:文档选择
|      ├── prompt:提示语
|      ├── refiner:信息总结
|      ├── reranker:排序模块
|      ├── retrieval:检索模块
|      └── rewriter:改写模块

2 导入模块

import pickle
import pandas as pd
from tqdm import tqdm

from trustrag.modules.document.chunk import TextChunker
from trustrag.modules.document.txt_parser import TextParser
from trustrag.modules.document.utils import PROJECT_BASE
from trustrag.modules.generator.llm import GLM4Chat
from trustrag.modules.reranker.bge_reranker import BgeRerankerConfig, BgeReranker
from trustrag.modules.retrieval.bm25s_retriever import BM25RetrieverConfig
from trustrag.modules.retrieval.dense_retriever import DenseRetrieverConfig
from trustrag.modules.retrieval.hybrid_retriever import HybridRetriever, HybridRetrieverConfig

3 文档解析以及切片

def generate_chunks():
    tp = TextParser()# 代表txt格式解析
    tc = TextChunker()
    paragraphs = tp.parse(r'H:/2024-Xfyun-RAG/data/corpus.txt', encoding="utf-8")
    print(len(paragraphs))
    chunks = []
    for content in tqdm(paragraphs):
        chunk = tc.chunk_sentences([content], chunk_size=1024)
        chunks.append(chunk)

    with open(f'{PROJECT_BASE}/output/chunks.pkl', 'wb') as f:
        pickle.dump(chunks, f)

corpus.txt每行为一段新闻,可以自行选取paragraph读取的逻辑,语料来自大模型RAG智能问答挑战赛

TextChunker为文本块切块程序,主要特点使用InfiniFlow/huqie作为文本检索的分词器,适合RAG场景。

4 构建检索器

配置检索器:

下面是一个混合检索器HybridRetriever配置参考,其中HybridRetrieverConfig需要由BM25RetrieverConfigDenseRetrieverConfig配置构成。

# BM25 and Dense Retriever configurations
bm25_config = BM25RetrieverConfig(
    method='lucene',
    index_path='indexs/description_bm25.index',
    k1=1.6,
    b=0.7
)
bm25_config.validate()
print(bm25_config.log_config())
dense_config = DenseRetrieverConfig(
    model_name_or_path=embedding_model_path,
    dim=1024,
    index_path='indexs/dense_cache'
)
config_info = dense_config.log_config()
print(config_info)
# Hybrid Retriever configuration
# 由于分数框架不在同一维度,建议可以合并
hybrid_config = HybridRetrieverConfig(
    bm25_config=bm25_config,
    dense_config=dense_config,
    bm25_weight=0.7,  # bm25检索结果权重
    dense_weight=0.3  # dense检索结果权重
)
hybrid_retriever = HybridRetriever(config=hybrid_config)

构建索引:

# 构建索引
hybrid_retriever.build_from_texts(corpus)
# 保存索引
hybrid_retriever.save_index()

如果构建好索引之后,可以多次使用,直接跳过上面步骤,加载索引

hybrid_retriever.load_index()

检索测试:

query = "支付宝"
results = hybrid_retriever.retrieve(query, top_k=10)
print(len(results))
# Output results
for result in results:
    print(f"Text: {result['text']}, Score: {result['score']}")

5 排序模型

reranker_config = BgeRerankerConfig(
    model_name_or_path=reranker_model_path
)
bge_reranker = BgeReranker(reranker_config)

6 生成器配置

glm4_chat = GLM4Chat(llm_model_path)

6 检索问答

# ====================检索问答=========================
test = pd.read_csv(test_path)
answers = []
for question in tqdm(test['question'], total=len(test)):
    search_docs = hybrid_retriever.retrieve(question, top_k=10)
    search_docs = bge_reranker.rerank(
        query=question,
        documents=[doc['text'] for idx, doc in enumerate(search_docs)]
    )
    # print(search_docs)
    content = '\n'.join([f'信息[{idx}]:' + doc['text'] for idx, doc in enumerate(search_docs)])
    answer = glm4_chat.chat(prompt=question, content=content)
    answers.append(answer[0])
    print(question)
    print(answer[0])
    print("************************************/n")
test['answer'] = answers

test[['answer']].to_csv(f'{PROJECT_BASE}/output/gomate_baseline.csv', index=False)

🔧定制化RAG

构建自定义的RAG应用

import os

from trustrag.modules.document.common_parser import CommonParser
from trustrag.modules.generator.llm import GLMChat
from trustrag.modules.reranker.bge_reranker import BgeReranker
from trustrag.modules.retrieval.dense_retriever import DenseRetriever


class RagApplication():
    def __init__(self, config):
        pass

    def init_vector_store(self):
        pass

    def load_vector_store(self):
        pass

    def add_document(self, file_path):
        pass

    def chat(self, question: str = '', topk: int = 5):
        pass

模块可见rag.py

🌐体验RAG效果

可以配置本地模型路径

# 修改成自己的配置!!!
app_config = ApplicationConfig()
app_config.docs_path = "./docs/"
app_config.llm_model_path = "/data/users/searchgpt/pretrained_models/chatglm3-6b/"

retriever_config = DenseRetrieverConfig(
    model_name_or_path="/data/users/searchgpt/pretrained_models/bge-large-zh-v1.5",
    dim=1024,
    index_dir='/data/users/searchgpt/yq/TrustRAG/examples/retrievers/dense_cache'
)
rerank_config = BgeRerankerConfig(
    model_name_or_path="/data/users/searchgpt/pretrained_models/bge-reranker-large"
)

app_config.retriever_config = retriever_config
app_config.rerank_config = rerank_config
application = RagApplication(app_config)
application.init_vector_store()
python app.py

浏览器访问:127.0.0.1:7860 trustrag_demo.png

app后台日志: app_logging3.png

⭐️ Star History

Star History Chart

研究与开发团队

本项目由网络数据科学与技术重点实验室GoMate团队完成,团队指导老师为郭嘉丰、范意兴研究员。

技术交流群

欢迎多提建议、Bad cases,欢迎进群及时交流,也欢迎大家多提PR

群满或者合作交流可以联系:

致谢

About

TrustRAG:The RAG Framework within Reliable input,Trusted output

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published