update sth. at 2024-11-21 18:18:17

app.py

@@ -272,15 +272,15 @@ def generate_response(user_prompt, system_instructions=sysx, model="microsoft/Ph
             max_new_tokens=100,
             do_sample=True,
         )
-
         formatted_prompt = f"[SYSTEM] {system_instructions}[QUESTION]{user_prompt}[ANSWER]"
-        stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True,
-                                        return_full_text=False)
+        stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=True)
         output = ""
+
         for response in stream:
             output += response.token.text
             if "[QUESTION]" in output:
                 break
+        print("output:",output)
         # 移除 [QUESTION] 及其之后的内容
         if "[QUESTION]" in output:
             output = output[:output.index("[QUESTION]")].strip()
@@ -294,6 +294,10 @@ def home():
     return "hello, work~~~"
 
 # /completions 路径
+@app.route("/api/v1/chat/completions", methods=["POST"])
+@app.route("/hf/v1/chat/completions", methods=["POST"])
+@app.route("/chat/completions", methods=["POST"])
+@app.route("/chat", methods=["POST"])
 @app.route("/completions", methods=["POST"])
 def completions():
     data = request.json
@@ -306,11 +310,26 @@ def completions():
 
     # 生成响应
     response_text = generate_response(user_message, system_instructions, model)
-    response_data = {"choices": [{"message": {"role": "assistant", "content": response_text}}]}
+    # 构建响应数据
+    response_data = {
+        "choices": [
+            {
+                "message": {
+                    "role": "assistant",
+                    "content": response_text
+                }
+            }
+        ]
+    }
     print(response_data)
-    # 返回 JSON 响应
-    # return jsonify(response_data)
-    return response_data
+    # 将数据转换为JSON字符串，确保非ASCII字符正确显示
+    response_json = json.dumps(response_data, ensure_ascii=False)
+
+    # 打印调试信息
+    print(response_json)
+
+    # 返回 JSON 响应，确保内容类型为 application/json
+    return app.response_class(response=response_json, mimetype='application/json')
 
 if __name__ == "__main__":
     # 启动 Flask 应用