docker run -p 9200:9200 \-e "discovery.type=single-node" \-e "xpack.security.enabled=false" \-e "xpack.license.self_generated.type=trial" \docker.elastic.co/elasticsearch/elasticsearch:8.13.2

# 向量数据库: Elasticsearch
from llama_index.vector_stores.elasticsearch import ElasticsearchStorefrom llama_index.vector_stores.elasticsearch import AsyncDenseVectorStrategy
ES_URI = "http://localhost:9200"
es_vector_store = ElasticsearchStore(es_url=ES_URI,index_name="my_index",retrieval_strategy=AsyncDenseVectorStrategy(hybrid=True), # 使用混合检索)

# 文档存储: MongoDB
from llama_index.storage.docstore.mongodb import MongoDocumentStorefrom llama_index.storage.index_store.mongodb import MongoIndexStorefrom llama_index.core import StorageContext
MONGO_URI = "mongodb://localhost:27017"
mongo_doc_store = MongoDocumentStore.from_uri(uri=MONGO_URI)mongo_index_store = MongoIndexStore.from_uri(uri=MONGO_URI)
storage_context = StorageContext.from_defaults(docstore=mongo_doc_store,index_store=mongo_index_store,vector_store=es_vector_store,)

# Ollama本地LLM: gemma:2b
from llama_index.llms.ollama import Ollamallm_ollama = Ollama(model="gemma:2b", request_timeout=600.0)

# 本地嵌入模型: bge-small-zh-v1.5
from llama_index.embeddings.huggingface import HuggingFaceEmbeddingbge_embed_model = HuggingFaceEmbedding(model_name="BAAI/bge-small-zh-v1.5")

export HF_ENDPOINT=https://hf-mirror.com

# LlamaIndex全局配置
from llama_index.core import SettingsSettings.llm = llm_ollamaSettings.embed_model = bge_embed_model

# 文本分割器: SpacyTextSplitter
from llama_index.core.node_parser import LangchainNodeParserfrom langchain.text_splitter import SpacyTextSplitterspacy_text_splitter = LangchainNodeParser(SpacyTextSplitter(pipeline="zh_core_web_sm", chunk_size = 512,chunk_overlap = 128))

# 加载文件信息: SimpleDirectoryReader
from llama_index.core import SimpleDirectoryReaderdocuments = SimpleDirectoryReader(input_dir="./data", recursive=True).load_data() print(f"Loaded {len(documents)} Files")

# 加载网页信息: SimpleWebPageReader
from llama_index.readers.web import SimpleWebPageReader
pages = ["https://mp.weixin.qq.com/s/prnDzOQ8HjUmonNp0jhRfw",]documents = SimpleWebPageReader(html_to_text=True).load_data(pages)print(f"Loaded {len(documents)} Web Pages")

# 数据转换管道
from llama_index.core.ingestion import IngestionPipelinepipeline = IngestionPipeline(transformations=[spacy_text_splitter,bge_embed_model,],docstore=mongo_doc_store,vector_store=es_vector_store,)

# 生成索引存入向量数据库
nodes = pipeline.run(documents=documents)print(f"Ingested {len(nodes)} Nodes")print(f"Load {len(pipeline.docstore.docs)} documents into docstore")

# 创建向量存储索引 
from llama_index.core import VectorStoreIndexindex = VectorStoreIndex(nodes, storage_context=storage_context)

# 定制中文Prompt
from llama_index.core import PromptTemplate
text_qa_template_str = ("以下为上下文信息\n""---------------------\n""{context_str}\n""---------------------\n""请根据上下文信息回答我的问题或回复我的指令。前面的上下文信息可能有用，也可能没用，你需要从我给出的上下文信息中选出与我的问题最相关的那些，来为你的回答提供依据。回答一定要忠于原文，简洁但不丢信息，不要胡乱编造。我的问题或指令是什么语种，你就用什么语种回复。\n""问题：{query_str}\n""你的回复：")

# 创建查询引擎并查询
query_engine = index.as_query_engine(text_qa_template = text_qa_template,top_k=1,)
your_question = "什么是流程？"print(f"问题：{your_question}")response = query_engine.query(your_question)print(f"回答：{response}")

# 使用Streamlit构建Web UI
import streamlit as st
TITLE = "本地大模型知识库问答"st.set_page_config(page_title=TITLE, page_icon="?", layout="centered", initial_sidebar_state="auto", menu_items=None)st.header(TITLE)st.info("By 大卫", icon="?")
if "messages" not in st.session_state.keys(): # 初始化聊天历史记录st.session_state.messages = [{"role": "assistant", "content": "关于文档里的内容，请随便问"}]
# 提示用户输入问题，并将问题添加到消息历史记录if prompt := st.chat_input("Your question"): st.session_state.messages.append({"role": "user", "content": prompt})
# 显示此前的问答记录for message in st.session_state.messages: with st.chat_message(message["role"]):st.write(message["content"])
# 生成回答if st.session_state.messages[-1]["role"] != "assistant": with st.chat_message("assistant"):with st.spinner("Thinking..."):response = query_engine.query(prompt)st.write(response.response)message = {"role": "assistant", "content": response.response}st.session_state.messages.append(message)

streamlit run app.py