build_vector_store.py

import os
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain_community.vectorstores import Chroma

# 1. 设置路径
BASE_DIR = os.path.dirname(os.path.abspath(__file__))  # 当前脚本所在路径
PERSIST_DIR = os.path.abspath(os.path.join(BASE_DIR, "../vector_store"))  # 向量库存储路径
SOURCE_DIR = BASE_DIR  # 你的 .md 文件就在当前 vector_build/ 目录下

# 2. 加载 Embedding 模型
embed_model = HuggingFaceEmbeddings(
    model_name="sentence-transformers/paraphrase-multilingual-mpnet-base-v2"
)

# 3. 加载 Markdown 文件 & 切分为小段
text_splitter = RecursiveCharacterTextSplitter(
    chunk_size=500, chunk_overlap=50
)

docs = []
for fname in os.listdir(SOURCE_DIR):
    if fname.endswith(".md"):
        with open(os.path.join(SOURCE_DIR, fname), "r", encoding="utf-8") as f:
            raw_text = f.read()
        chunks = text_splitter.split_text(raw_text)
        for chunk in chunks:
            docs.append({
                "text": chunk,
                "source": fname
            })

print(f"🐣 共切分出 {len(docs)} 个文本块，准备向量化...")

# 4. 创建 Chroma 向量库
texts = [d["text"] for d in docs]
metas = [{"source": d["source"]} for d in docs]

vectordb = Chroma.from_texts(
    texts=texts,
    embedding=embed_model,
    metadatas=metas,
    persist_directory=PERSIST_DIR
)
vectordb.persist()

print(f"🎉 向量库生成完毕，已保存在：{PERSIST_DIR}")