更新套件

app.py

@@ -6,11 +6,10 @@ import os, glob, requests
 from langchain.docstore.document import Document
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain.chains import RetrievalQA
-from langchain_huggingface import HuggingFaceEmbeddings
+from langchain_huggingface import HuggingFaceEmbeddings, HuggingFaceEndpoint
 from docx import Document as DocxDocument
 import gradio as gr
 from langchain_community.vectorstores import FAISS
-from langchain_community.llms import HuggingFaceHub
 
 # -------------------------------
 # 2. 環境變數與資料路徑
@@ -21,7 +20,9 @@ os.makedirs(DB_PATH, exist_ok=True)
 
 HF_TOKEN = os.environ.get("HUGGINGFACEHUB_API_TOKEN")
 if not HF_TOKEN:
-    raise ValueError("請在 Hugging Face Space 的 Settings → Repository secrets 設定 HUGGINGFACEHUB_API_TOKEN")
+    raise ValueError(
+        "請在 Hugging Face Space 的 Settings → Repository secrets 設定 HUGGINGFACEHUB_API_TOKEN"
+    )
 
 # -------------------------------
 # 3. 建立或載入向量資料庫
@@ -38,7 +39,9 @@ else:
     docs = []
     for filepath in txt_files:
         with open(filepath, "r", encoding="utf-8") as f:
-            docs.append(Document(page_content=f.read(), metadata={"source": os.path.basename(filepath)}))
+            docs.append(
+                Document(page_content=f.read(), metadata={"source": os.path.basename(filepath)})
+            )
     splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
     split_docs = splitter.split_documents(docs)
     db = FAISS.from_documents(split_docs, embeddings_model)
@@ -47,12 +50,13 @@ else:
 retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": 5})
 
 # -------------------------------
-# 4. LLM 設定（Hugging Face Hub）
+# 4. LLM 設定（Hugging Face Endpoint）
 # -------------------------------
-llm = HuggingFaceHub(
+llm = HuggingFaceEndpoint(
     repo_id="google/flan-t5-large",
+    task="text2text-generation",   # 明確指定 task
+    huggingfacehub_api_token=HF_TOKEN,
     model_kwargs={"temperature": 0.7, "max_new_tokens": 512},
-    huggingfacehub_api_token=HF_TOKEN
 )
 
 qa_chain = RetrievalQA.from_chain_type(
@@ -102,7 +106,6 @@ def generate_article_with_rate(query, segments=5):
     doc.save(docx_file)
     full_text = "\n\n".join(all_text)
 
-    # 取得 API 剩餘次數
     rate_info = get_hf_rate_limit()
     return f"{rate_info}\n\n{full_text}", docx_file
 
@@ -120,9 +123,8 @@ iface = gr.Interface(
         gr.File(label="下載 DOCX")
     ],
     title="佛教經論 RAG 系統 (HF API)",
-    description="使用 Hugging Face Hub LLM + FAISS RAG，生成文章並提示 API 剩餘額度。"
+    description="使用 Hugging Face Endpoint LLM + FAISS RAG，生成文章並提示 API 剩餘額度。"
 )
 
 if __name__ == "__main__":
     iface.launch()
-