app.py

import streamlit as st
from langchain_community.llms import HuggingFaceHub
from langchain_community.embeddings import SentenceTransformerEmbeddings
from langchain_community.vectorstores import FAISS
import numpy as np

# gemma = 'google/gemma-7b-it';
gemma = 'google/recurrentgemma-2b-it';

# 2. 准备知识库数据 (示例)
knowledge_base = [
    "Gemma 是 Google 开发的大型语言模型。",
    "Gemma 具有强大的自然语言处理能力。",
    "Gemma 可以用于问答、对话、文本生成等任务。",
    "Gemma 基于 Transformer 架构。",
    "Gemma 支持多种语言。"
]

# 3. 构建向量数据库 (如果需要，仅构建一次)
try:
    embeddings = SentenceTransformerEmbeddings(model_name="all-mpnet-base-v2")
    db = FAISS.from_texts(knowledge_base, embeddings)
except Exception as e:
    st.error(f"向量数据库构建失败：{e}")
    st.stop()

# 4. 问答函数
def answer_question(gemma, temperature, max_length, question):
    # 1. 初始化 Gemma 模型
    try:
        llm = HuggingFaceHub(repo_id=gemma, model_kwargs={"temperature": temperature, "max_length": max_length})
    except Exception as e:
        st.error(f"Gemma 模型加载失败：{e}")
        st.stop()
    try:
        question_embedding = embeddings.embed_query(question)
        question_embedding_np = " ".join(map(str, question_embedding))
        docs_and_scores = db.similarity_search_with_score(question_embedding_np)
        # 正确处理 docs_and_scores 列表
        context = "\n".join([doc.page_content for doc, _ in docs_and_scores])  # 使用 join() 方法
        print(context)

        prompt = f"请根据以下知识库回答问题：\n{context}\n问题：{question}"
        print(prompt)

        answer = llm(prompt)
        return answer
    except Exception as e:
        st.error(f"问答过程出错：{e}")
        return "An error occurred during the answering process."

# 5. Streamlit 界面
st.title("Gemma 知识库问答系统")

gemma = st.selectbox("模型", ("google/gemma-2-9b-it", "google/gemma-2-2b-it", "google/recurrentgemma-2b-it"), 2)
temperature = st.text_area("temperature", "1.0")
max_length = st.text_area("max_length", "1024")
question = st.text_area("请输入问题", "Gemma 有哪些特点？")

if st.button("提交"):
    if not question:
        st.warning("请输入问题！")
    else:
        with st.spinner("正在查询..."):
            answer = answer_question(gemma, temperature, max_length, question)
            st.write("答案：")
            st.write(answer)