app.py

import os, gradio as gr, torch, logging
from langchain_chroma import Chroma
from langchain_community.embeddings import HuggingFaceEmbeddings   # ← 新路径
from langchain_community.llms import HuggingFacePipeline
from langchain.chains import RetrievalQA
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

logging.basicConfig(level=logging.INFO)

VECTOR_STORE_DIR = "./vector_store"
MODEL_NAME = "uer/gpt2-chinese-cluecorpussmall"
EMBEDDING_MODEL_NAME = "sentence-transformers/paraphrase-multilingual-mpnet-base-v2"

# ─── 1. 加载 LLM ───
print("🔧 加载生成模型…")
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto",
)
gen_pipe = pipeline(
    task="text-generation",
    model=model,
    tokenizer=tokenizer,
    max_new_tokens=256,
    temperature=0.5,
    top_p=0.9,
    do_sample=True,
)
llm = HuggingFacePipeline(pipeline=gen_pipe)

# ─── 2. 加载向量库 ───
print("📚 加载向量库…")
embeddings = HuggingFaceEmbeddings(model_name=EMBEDDING_MODEL_NAME)
vectordb = Chroma(persist_directory=VECTOR_STORE_DIR, embedding_function=embeddings)

# ─── 3. 构建 RAG 问答链 ───
retriever = vectordb.as_retriever(search_kwargs={"k": 3})
qa_chain = RetrievalQA.from_chain_type(
    llm=llm,
    chain_type="stuff",
    retriever=retriever,
    return_source_documents=True,
)

# ─── 4. 业务函数 ───
def qa_fn(query: str):
    if not query.strip():
        return "❌ 请输入问题内容。"
    result = qa_chain({"query": query})
    answer = result["result"]
    sources = result.get("source_documents", [])
    sources_text = "\n\n".join(
        [f"【片段 {i+1}】\n{doc.page_content}" for i, doc in enumerate(sources)]
    )
    return f"📌 回答：{answer.strip()}\n\n📚 参考：\n{sources_text}"

# ─── 5. Gradio UI ───
with gr.Blocks(title="数学知识问答助手") as demo:
    gr.Markdown("## 📘 数学知识问答助手\n输入教材相关问题，例如：“什么是函数的定义域？”")
    with gr.Row():
        query = gr.Textbox(label="问题", placeholder="请输入你的问题", lines=2)
        answer = gr.Textbox(label="回答", lines=15)
    gr.Button("提问").click(qa_fn, inputs=query, outputs=answer)
    gr.Markdown("---\n模型：gpt2-chinese-cluecorpus + Chroma RAG\nPowered by Hugging Face Spaces")

if __name__ == "__main__":
    demo.launch()