Update app.py

app.py

@@ -1,3 +1,4 @@
+# coding: utf-8
 from openai import OpenAI
 from docx import Document
 import numpy as np
@@ -20,7 +21,7 @@ client = OpenAI(
 embedder = SentenceTransformer(EMBEDDING_MODEL)
 
 def process_word_document():
-    """处理Word文档并分块（保持原有实现不变）"""
+    """处理Word文档并分块"""
     doc = Document(WORD_DOC_PATH)
     chunks = []
     current_chunk = []
@@ -42,7 +43,7 @@ def process_word_document():
     return chunks
 
 def create_vector_store():
-    """创建并保存向量存储（保持原有实现不变）"""
+    """创建并保存向量存储"""
     if os.path.exists(VECTOR_INDEX_PATH):
         return
 
@@ -58,7 +59,7 @@ def create_vector_store():
     np.save(TEXT_DATA_PATH, np.array(chunks))
 
 def search_knowledge(query, top_k=3):
-    """知识检索（保持原有实现不变）"""
+    """知识检索"""
     index = faiss.read_index(VECTOR_INDEX_PATH)
     text_data = np.load(TEXT_DATA_PATH, allow_pickle=True)
 
@@ -68,15 +69,15 @@ def search_knowledge(query, top_k=3):
     distances, indices = index.search(query_embedding, top_k)
     return "\n".join([text_data[i] for i in indices[0]])
 
-def respond(message, history, max_tokens, temperature, top_p):
+def respond(message, history, max_tokens, temperature, top_p, user_input):
     """Gradio响应函数"""
     # 检索相关知识
-    context = search_knowledge(message)
+    context = search_knowledge(user_input)
     
     # 构建对话消息
     messages = [
         {"role": "system", "content": f"基于以下知识回答问题，如果不知道就说不知道：\n{context}"},
-        {"role": "user", "content": message}
+        {"role": "user", "content": user_input}
     ]
 
     # 流式生成响应
@@ -90,34 +91,23 @@ def respond(message, history, max_tokens, temperature, top_p):
         top_p=top_p
     )
 
+    done_reasoning = False
     for chunk in response:
-    # 从当前分块中获取推理内容和回答内容
         reasoning_chunk = chunk.choices[0].delta.reasoning_content or ""
         answer_chunk = chunk.choices[0].delta.content or ""
         
-        # 累加推理内容到全局变量
         if reasoning_chunk:
-            full_reasoning += reasoning_chunk
-            # 当前分块是推理内容，更新并返回完整的推理内容
-            yield full_reasoning + "\n\n=== 最终答案 ===\n" + full_response  # 提前构建完整输出
-    
-        # 累加回答内容到全局变量
+            full_response += reasoning_chunk
+            print(reasoning_chunk, end='', flush=True)
         elif answer_chunk:
-            full_response += answer_chunk
-    
-            # 如果还没有进入答案输出阶段
             if not done_reasoning:
-                done_reasoning = True  # 标记推理阶段结束
-                # 立即更新并返回，带上推理内容和“最终答案”提示
-                yield full_reasoning + "\n\n=== 最终答案 ===\n" + full_response
+                print('\n\n=== 最终答案 ===\n')
+                done_reasoning = True
+            print(answer_chunk, end='', flush=True)
     
-            # 如果已经在答案输出阶段，则继续累加答案内容并返回更新
-            else:
-                yield full_reasoning + "\n\n=== 最终答案 ===\n" + full_response
-        yield full_response, full_reasoning
+    print("\n" + "="*50)
+    return full_response
 
-    # Finally, yield the complete response and reasoning after all chunks are processed
-    yield full_response, full_reasoning
 # 初始化向量存储
 create_vector_store()
 
@@ -125,10 +115,12 @@ create_vector_store()
 demo = gr.ChatInterface(
     fn=respond,
     additional_inputs=[
+        gr.Textbox(label="用户提问"),
         gr.Slider(512, 2048, value=512, step=1, label="最大Token数"),
         gr.Slider(0.1, 2.0, value=0.7, step=0.1, label="温度参数"),
         gr.Slider(0.1, 1.0, value=0.95, step=0.05, label="Top-p采样"),
     ],
+    theme=gr.themes.Soft(),
     title="制度文档问答系统",
     description="输入关于广西警察学院制度的问题进行问答"
 )