Update api.py

api.py

@@ -8,29 +8,39 @@ app = Flask(__name__)
 def chat_completions():
     data = request.json
     messages = data.get('messages', [])
-    max_tokens = data.get('max_tokens', 1080)
+    max_tokens = data.get('max_tokens', 560)
     temperature = data.get('temperature', 0.8)
     
     if not messages:
         return jsonify({"error": "messages array is required"}), 400
     
     try:
+        start_time = time.time()
         result = generate_chat_completion(
             messages=messages,
             max_tokens=max_tokens,
             temperature=temperature
         )
+        elapsed = time.time() - start_time
+        
         return jsonify({
-            "model": "DS-R1-Qwen3-8B",
+            "model": "Mistral-7B-Instruct",
             "choices": [{
                 "message": {
                     "role": "assistant",
                     "content": result
                 }
-            }]
+            }],
+            "usage": {
+                "generation_time": round(elapsed, 2)
+            }
         })
     except Exception as e:
         return jsonify({"error": str(e)}), 500
 
+@app.route('/')
+def health_check():
+    return "API is running", 200
+
 if __name__ == '__main__':
     app.run(host='0.0.0.0', port=8081)