Agent实践三:基于Chroma的RAG检索

embeding.py

python 复制代码
import function
import chromadb
from google import genai

google_client = genai.Client()
EMBEDDING_MODEL = "gemini-embedding-exp-03-07"
LLM_MODEL = "gemini-2.5-flash-preview-05-20"

chromadb_client = chromadb.PersistentClient("./chroma.db")
chromadb_collection = chromadb_client.get_or_create_collection("question")

def embed(text: str, store: bool) -> list[float]:
    result = google_client.models.embed_content(
        model=EMBEDDING_MODEL,
        contents=text,
        config={
            "task_type": "RETRIEVAL_DOCUMENT" if store else "RETRIEVAL_QUERY"
        }
    )

    assert result.embeddings
    assert result.embeddings[0].values
    return result.embeddings[0].values

def create_db() -> None:
    for idx, c in enumerate(function.get_chunks()):
        print(f"Process: {c}")
        embedding = embed(c, store=True)
        chromadb_collection.upsert(
            ids=str(idx),
            documents=c,
            embeddings=embedding
        )

def query_db(question: str) -> list[str]:
    question_embedding = embed(question, store=False)
    result = chromadb_collection.query(
        query_embeddings=question_embedding,
        n_results=5
    )
    assert result["documents"]
    return result["documents"][0]


if __name__ == '__main__':
    question = "question"
    # create_db()
    chunks = query_db(question)
    prompt = "Please answer user's question according to context\n"
    prompt += f"Question: {question}\n"
    prompt += "Context:\n"
    for c in chunks:
        prompt += f"{c}\n"
        prompt += "-------------\n"
    
    result = google_client.models.generate_content(
        model=LLM_MODEL,
        contents=prompt
    )
    print(result)

function.py

python 复制代码
def read_data() -> str:
    with open("data.md", "r", encoding="utf-8") as f:
        return f.read()
 
def get_chunks() -> list[str]:
    content = read_data()
    chunks = content.split('\n\n')
    
    result = []
    header = ""
    for c in chunks:
        if c.startswith("#"):
            header += f"{c}\n"
        else:
            result.append(f"{header}{c}")
            header = ""

    return result

if __name__ == '__main__':
    chunks = get_chunks()
    for c in chunks:
        print(c)
        print("--------------")
相关推荐
TomEval1 天前
【测AI】第26篇:线上模型监控与告警体系 — 测试不是终点
人工智能·python·功能测试·ai
prince_zxill1 天前
PEP 751 pylock.toml 实测:pip 26.3 装 6 个包版本全钉死,但 [web] 选不中任何东西
python·pip·依赖管理·pep 751
Thanks_ks1 天前
Python 的垃圾回收——引用计数撑起日常,分代回收收拾残局
python·内存管理·垃圾回收·循环引用·引用计数·gc 模块·weakref
CyrusYiVPP1 天前
实测 uv 0.13.7:同一套包,pip 装 94 秒,它只用 19 秒
人工智能·python
学无止境小袁同学1 天前
LangGraph 深度指南
语言模型·框架·agent·并发·上下文·langgraph·长期记忆
yiangL_1 天前
Python 学习笔记:递归、turtle绘图、字符串与函数
笔记·python·学习
Cosolar1 天前
2026 年 Java AI Agent 生态全景盘点:50+ Java 开源项目的分类、实测与选型判断
人工智能·架构·agent
lula8881 天前
PVC-U、PVC-C、PVC-O 差一个字母,差的是三种增强路线
python
张一凡931 天前
你的 AGENTS.md,真的管到模型写代码了吗?
前端·后端·agent
qinglongcn1 天前
怎么让 200 轮的长任务不崩——三层保全策略
agent·ai编程