学习LLamaIndex第一个程序

一、先安装

pip install llama-index

pip install llama-index-llms-ollama
pip install llama-index-embeddings-ollama

二、开发环境

开一个写代码的工具：

jupyter notebook

也可以使用Pycharm来使用conda环境进行代码的编写：

设置/项目/python 解释器，添加解释器的时候选择conda环境，并选择一个已经存在的环境或者新建环境。

也可以使用VSCode：

# https://code.visualstudio.com/Download
sudo apt install ./xxxx.deb

三、基本概念

LLamaIndex核心概念：

  1. 数据加载：DataLoaders
  2. 数据分割：NodeParsers
  3. 向量索引：VectorIndex
  4. 集成引擎：QueryEngine

向量数据库：

  • ChromaDB
  • Mivlus

2.1 ChromaDB

使用十分简单：

import chromadb

client = chromadb.Client()
collection = client.create_collection("docs")
collection.add(ids=["id1"], documents=["This is a document"])
results = collection.query(query_texts=["Find similar docs"], n_results=1)

2.2 Embedding

Ollama

ollama serve

// 使用Ollama原生的REST API
curl http://localhost:11434/api/embed -d '{
  "model": "bge-m3",
  "input": "Does Ollama support embedding models?"
}'
// 返回
{"model":"bge-m3","embeddings":[[-0.026588926,-0.013804924,-0.032380637,.....,0.018297898,-0.034554675,0.0075376146]],"total_duration":1060074189,"load_duration":955242593,"prompt_eval_count":9}


// 使用OpenAI兼容的REST API
curl http://localhost:11434/v1/embeddings -d '{
  "model": "bge-m3",
  "input": "Does Ollama support embedding models?"
}'
// 返回
{"object":"list","data":[{"object":"embedding","embedding":[-0.026588926,-0.013804924,-0.032380637,....,0.026003407,0.018297898,-0.034554675,0.0075376146],"index":0}],"model":"bge-m3","usage":{"prompt_tokens":9,"total_tokens":9}}

四、样例代码

4.1 读文件

from llama_index.core import SimpleDirectoryReader

documents = SimpleDirectoryReader("./data").load_data()

print(f"文档数量: {len(documents)}")
print(documents[0].text[:500]) # 预览前 500 个字符

4.2 构建向量索引并保存

Settings.embed_model = OllamaEmbedding(model_name="bge-m3:latest")
index = VectorStoreIndex.from_documents(
    documents,
    # we can optionally override the embed_model here
    # embed_model=Settings.embed_model,
)
# Save the index
index.storage_context.persist("storage")

$ ll storage/
total 624
-rw-r--r-- 1 xxxx xxxx 481486 Jun  2 21:23 default__vector_store.json
-rw-r--r-- 1 xxxx xxxx 141154 Jun  2 21:23 docstore.json
-rw-r--r-- 1 xxxx xxxx     18 Jun  2 21:23 graph_store.json
-rw-r--r-- 1 xxxx xxxx     72 Jun  2 21:23 image__vector_store.json
-rw-r--r-- 1 xxxx xxxx   2095 Jun  2 21:23 index_store.json

4.3 加载向量，根据大模型生成查询引擎

from llama_index.core import StorageContext, load_index_from_storage

Settings.llm = Ollama(model="llama3.1", request_timeout=360.0)

storage_context = StorageContext.from_defaults(persist_dir="storage")
index = load_index_from_storage(storage_context)

query_engine = index.as_query_engine(
    # we can optionally override the llm here
    # llm=Settings.llm,
)

4.4 构建查询的workflow

def multiply(a: float, b: float) -> float:
    """Useful for multiplying two numbers."""
    return a * b


async def search_documents(query: str) -> str:
    """Useful for answering natural language questions about an personal essay written by Paul Graham."""
    response = await query_engine.aquery(query)
    return str(response)


# Create an enhanced workflow with both tools
agent = AgentWorkflow.from_tools_or_functions(
    [multiply, search_documents],
    llm=Settings.llm,
    system_prompt="""You are a helpful assistant that can perform calculations
    and search through documents to answer questions.""",
)

4.5 调用workflow进行查询

async def main():
    response = await agent.run(
        "What did the author do in college? Also, what's 7 * 8?"
    )
    print(response)

if __name__ == "__main__":
    asyncio.run(main())

$ python ./run.py

The author attended Harvard, where they took a painting class with Idelle Weber, an early photorealist. After high school. The result of 7 * 8 is 56.

五、调用的DeepSeek情况

![[Pasted_image_20250602213511.png]]