Update app.py

app.py

@@ -7,13 +7,17 @@ import torch
 
 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 
-from langchain_community.embeddings import HuggingFaceEmbeddings
-from langchain_community.vectorstores import Chroma
+# Embeddings 与 VectorStore 用新的分包
+from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_chroma import Chroma
+
+# LLM 继续用 community 包里的 Pipeline
 from langchain_community.llms import HuggingFacePipeline
+
 from langchain.chains import RetrievalQA
 from langchain.prompts import PromptTemplate
 
-from build_index import main as build_index_if_needed  # 需确保 build_index.py 在同目录
+from build_index import main as build_index_if_needed  # 确保 build_index.py 与 app.py 同目录
 
 logging.basicConfig(level=logging.INFO)
 
@@ -22,7 +26,7 @@ VECTOR_STORE_DIR        = "./vector_store"
 MODEL_NAME              = "uer/gpt2-chinese-cluecorpussmall"
 EMBEDDING_MODEL_NAME    = "sentence-transformers/paraphrase-multilingual-mpnet-base-v2"
 
-# 容器启动时自动构建向量库（如果还没提交 vector_store）
+# 容器启动时自动构建向量库（如果 vector_store 目录为空）
 if not os.path.exists(VECTOR_STORE_DIR) or not os.listdir(VECTOR_STORE_DIR):
     logging.info("向量库不存在，启动自动构建……")
     build_index_if_needed()
@@ -72,9 +76,8 @@ prompt_template = PromptTemplate.from_template(
 # ─── 4. 构建 RAG 问答链（map_reduce） ───────────────────────────
 qa_chain = RetrievalQA.from_chain_type(
     llm=llm,
-    chain_type="map_reduce",                             # map_reduce 避免超长
+    chain_type="map_reduce",               # map_reduce 自动分段、避免超长
     retriever=retriever,
-    chain_type_kwargs={"prompt": prompt_template},
     return_source_documents=True,
 )
 logging.info("✅ RAG 问答链（map_reduce）构建成功。")
@@ -122,3 +125,4 @@ if __name__ == "__main__":
 
 
 
+