Update app.py

app.py

@@ -167,11 +167,12 @@ def talk(prompt, history):
   #  print("check11")
     # start a thread
     outputs = []  
-    print(model_input)
-    print(model.tokenize(messages))
-    tokens = model.tokenize(messages)
-    for token in model.generate(tokens):
-       print(model.detokenize([token]))
+    outputs = model_input
+    return outputs
+#    print(model.tokenize(messages))
+#    tokens = model.tokenize(messages)
+#    for token in model.generate(tokens):
+#       print(model.detokenize([token]))
  #   input_ids = tokenizer(*messages)
 #    print(model.generate(tensor([[    1,   529, 29989,  5205, 29989]])))
 #    start = time.time()