Update app.py

app.py

@@ -57,17 +57,16 @@ def respond(
 
     sample = True
 
-    if model_temperature == 0:
+    if temperature == 0:
         sample = False
 
-
     # conduct text completion
     generated_ids = model.generate(
         **model_inputs,
         max_new_tokens=max_tokens,
         do_sample=sample,
         top_p=top_p,
-        temperature=model_temperature
+        temperature=temperature
     )
     output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()