私有知识库搭建整理

• 资源: 从零搭建基于本地大语言模型构建的私有知识库系统（下）软件篇_哔哩哔哩_bilibili

一. 私有知识库选型:

• 主程序 + 前端:
  • TestGeneration-WebUI: oobabooga/text-generation-webui: A Gradio web UI for Large Language Models. Supports transformers, GPTQ, llama.cpp (ggml), Llama models. (github.com)
  • Langchain-Chatchat: chatchat-space/Langchain-Chatchat: Langchain-Chatchat (formerly langchain-ChatGLM), local knowledge based LLM (like ChatGLM) QA app with langchain ｜ 基于 Langchain 与 ChatGLM 等语言模型的本地知识库问答 (github.com)
  • Langchan + Gradio UI 开发
• 向量数据库:
  • PGvector
  • Milvus
  • Faiss
• 模型与知识库
  • ChatGLM2-6B-32K THUDM/chatglm2-6b · Hugging Face THUDM/chatglm2-6b-32k · Hugging Face
  • LLaMA2 meta-llama/Llama-2-7b · Hugging Face
  • LLM + 微调
  • 本地知识库

二. 安装笔记

• 下载资源:

  git clone https://huggingface.co/THUDM/chatglm2-6b-32k
  git clone https://huggingface.co/moka-ai/m3e-base
  git clone https://github.com/chatchat-space/Langchain-Chatchat.git 
  cd Langchain-Chatchat

• conda环境

  conda create -n chatchat python=3.10
  conda activate chatchat
  pip install --upgrade pip
  pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
  # pip install -pre torch torchvision torchautio --index-url https://download.pytorch.org/whl/nightly/cu121
  conda install spacy
  pip install cchardet
  pip install accelerate

• chatchat构建

  pip install -r requirements.txt
  cd configs
  cp ./model_config.py.example ./model_config.py
  # embedding_model_dict 中
  # "m3e-base":"D:\Files\projects\chatchat\models\m3e-base"
  # llm_model_dict 中
  # "local_model_path":"D:\Files\projects\chatchat\models\chatglm2-6b-32k"
  cp ./server_config.py.example ./ server_config.py

• 向量数据库配置

  git clone --branch v0.4.4 https://github.com/pgvector/pgvector.git
  cd pgvector
  # Postgresql + PGVector
  # https://www.enterprisedb.com/downloads/postgres-postgresql-downloads
  # 下载并安装Postgresql15
  # cmd中执行以下代码
  set PGROOT=C:\Program Files\PostgreSQL\15
  call "C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvars64.bat"
  nmake /F MakeFile.win
  nmake /F MakeFile.win install

  -- .\psql.exe --username=postgres 登录root账户
  CREATE DATABASE TEST;
  CREATE EXTENSION IF NOT EXISTS vector;

  python -m spacy download en_core_web_sm
  python -m spacy download zh_core_web_sm
  pip install psycopg2 pgvector flask-mysqldb protobuf==3.20 filemagic
  pip install -r requirements.txt
  # pgvector报错处理:
  # 错误: KeyError: 'answer'错误
  # Langchain-Chatchat/server/knowledge_base/km_service/base.py
  # 119行: docs = self.do_search(query, top_k, embeddings)
  python init_database.py
  # python init_database.py --recreate-vs

• 启动运行:

  python startup.py --all-webui