Update app.py

app.py

@@ -76,12 +76,12 @@ def run_huggingface_model(model, messages, max_tokens, temperature, top_p):
 
     model = "https://api-inference.huggingface.co/models/" + model + "/v1/chat/completions";
     response = requests.post(model, headers=headers, json=payload, timeout=30)
-    print(response)
-    response = json.loads(response)
 
-    if hasattr(response, 'error'):
+    if response.status_code != 200:
         print("ERROR: " + response.error)
     else:
+        print(response.content)
+        response = response.content
         result = response.choices[0].message.content
         return result