参考：gihtub代码 https://github.com/imartinez/privateGPT

安装

llama-cpp-python-0.1.48安装报错

 Could not build wheels for llama-cpp-python, , which is required to install pyproject.toml-based projects

搜索（结果较少）：
从文章： 升级gcc解决编译llama-cpp-python错误发现是gcc/g++版本过低所致，但是博主环境是ubuntu.

搜索到文章 centos：安装gcc g++知道默认centos7 yum安装的是4.8.5

总结：

解决方案centos gcc/g++升级到11
参考：centos7 升级gcc到11.2.0

注意：编译安装耗时较久

执行

python ingest.py 报错version `GLIBCXX_3.4.29‘ not found的问题

OSError: xxxx/lib/libstdc++.so.6: version `GLIBCXX_3.4.29' not found (required by xxxx/python3.9/site-packages/llama_cpp/libllama.so)

如何解决version `GLIBCXX_3.4.29‘ not found的问题

默认路径下的libstdc++.so.6缺少GLIBCXX_3.4.29

发现是

版本低了

ll  /xxx/lib/libstdc++.so.6

改为指向6.0.29

提取数据：
python ingest.py

python privateGPT.py执行查询

语法错误

  File "/data1/Projects/privateGPT/privateGPT.py", line 26
    match model_type:
          ^
SyntaxError: invalid syntax

是python3.10语法

参考：https://github.com/imartinez/privateGPT/issues/89

修改代码为：

if model_type == "LlamaCpp":
        llm = LlamaCpp(model_path=model_path, n_ctx=model_n_ctx, callbacks=callbacks, verbose=False)
elif model_type == "GPT4All":
        llm = GPT4All(model=model_path, n_ctx=model_n_ctx, backend='gptj', callbacks=callbacks, verbose=False)
else:
        print(f"Model {model_type} not supported!")
        exit;

query报错：
unknown token 忽视掉 https://github.com/imartinez/privateGPT/issues/77

输出乱码

拉取最新代码后解决

总结

缺点：
1.模型比较局限，只能选择 GPT4All-J 类型的模型，性能高的开源模型暂时无法兼容 https://gpt4all.io/index.html

2.中文文档
embedding模型用中文向量，但是问答结果要还是英文，要么是没有意义的文字

shibing624/text2vec-base-chinese

参考：https://zhuanlan.zhihu.com/p/630223486

3.内存管理有问题，第xx个问题报错内存不足，待解决
ggml_new_tensor_impl: not enough space in the context’s memory pool (needed 8355506480, available 8342642000)

4.仅使用cpu
。。。待解决

完整代码

不持久化文档，完整代码：

import os
 

from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
from langchain.llms import GPT4All, LlamaCpp
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.document_loaders import TextLoader
loader = TextLoader('./source_documents/xxx.txt',encoding='utf-8')

from langchain.indexes import VectorstoreIndexCreator
from langchain.chains import RetrievalQA
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.vectorstores import Chroma
 
from langchain.document_loaders import UnstructuredFileLoader
from langchain.chains.question_answering import load_qa_chain
model_path = 'xxx/ggml-gpt4all-j-v1.3-groovy.bin'
callbacks = [StreamingStdOutCallbackHandler()]
llm = GPT4All(model=model_path, backend='gptj', callbacks=callbacks, verbose=False)
chain = load_qa_chain(llm, chain_type="stuff")

chunk_size = 500
chunk_overlap = 50
text_splitter = RecursiveCharacterTextSplitter(chunk_size=chunk_size, chunk_overlap=chunk_overlap)

documents =  loader.load()
texts = text_splitter.split_documents(documents)

embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2" )

db = Chroma.from_documents(texts, embeddings)
retriever = db.as_retriever()

qa = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=retriever, return_source_documents=True)
query = "what is tame?"
res = qa(query)
answer, docs = res['result'], res['source_documents']