一种利用 langchain 思想实现的基于本地知识库的问答应用，目标期望建立一套对中文场景与开源模型支持友好、可离线运行的知识库问答解决方案。
1. 下载Langchain-chatchat
git clone https://github.com/chatchat-space/Langchain-Chatchat/
2. 下载大模型和embedding模型
大模型： git clone https://huggingface.co/THUDM/chatglm3-6b-32k
embedding 模型: git clone https://huggingface.co/BAAI/bge-large-zh
如果在下载过程中遇到huggingface.co is not conneted, 可以使用以下方法解决： 完美解决huggingface问题：Failed to connect to huggingface.co port 443_failed to connect to huggingface.co port 443 after-CSDN博客
pip install -U huggingface_hub
export HF_ENDPOINT=https://hf-mirror.com
huggingface-cli download --resume-download --local-dir-use-symlinks False THUDM/chatglm3-6b --local-dir chatglm3-6b
huggingface-cli download --resume-download --local-dir-use-symlinks False BAAI/bge-large-zh --local-dir bge-large-zh3. 下载需要的包 pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install spacy
pip install cchardet
pip install accelerate
pip install --upgrade pip
pip install -r requirements.txt
4. 修改配置文件
生成配置文件
python copy_config_example.py
cd configs
vim model_config.py修改
EMBEDDING_MODEL = "bge-large-zh"
LLM_MODELS = ["chatglm3-6b-32k"]
5. 安装相应的向量数据库
1）安装postgresql数据库
PostgreSQL: Linux downloads (Ubuntu)
启动数据库
service postgresql start
查看是否启动成功
pg_lsclusters
2）安装向量数据库
GitHub - pgvector/pgvector: Open-source vector similarity search for Postgres
通过以下的命令解决：
pg_config --includedir-server
sudo apt install postgresql-server-dev-all
su - postgres
psql修改kb_config.py
SQLALCHEMY_DATABASE_URI = f"postgresql://langchain:langchain@127.0.0.1:5432/langchain_chatchat"
DEFAULT_VS_TYPE = "pg"
创建新的user
CREATE USER langchain WITH PASSWORD 'langchain';
创建数据库langchain_chatchat
CREATE DATABASE langchain_chatchat;
ALTER USER "langchain" WITH login superuser createrole createdb replication bypassrls;
6. 安装插件
python -m spacy download en_core_web_sm
python -m spacy download zh_core_web_sm
pip install psycopg2
pip install pgvector
pip install flask-mysqldb
pip install protobuf==3.20
pip install filemagic
7. 启动服务的代码：
python startup.py -a
postgres sql操作
查询所有的数据库：\l
查看所有的用户：
SELECT usename AS username,
usesysid AS user_id,
usecreatedb AS can_create_db,
usesuper AS is_superuser,
userepl AS can_replicate,
passwd AS password,
valuntil AS password_expiration
FROM pg_catalog.pg_user;
GRANT CONNECT ON DATABASE langchain_chatchat TO langchain;
GRANT USAGE ON SCHEMA public TO langchain;
GRANT SELECT, INSERT, UPDATE, DELETE ON ALL TABLES IN SCHEMA public TO langchain;
ALTER USER "langchain" WITH login superuser createrole createdb replication bypassrls;
