Update app.py

app.py

@@ -45,7 +45,8 @@ def generate_with_llama_chat(my_config):
                        repetition_penalty=repetition_penalty,
                        #max_length=2000,
                        max_new_tokens=max_new_tokens, 
-                       model_kwargs={"device_map": "auto","load_in_8bit": True})  #, "src_lang": "en", "tgt_lang": "nl"})  does not work!
+                       #model_kwargs={"device_map": "auto","load_in_8bit": True}
+                      )  #, "src_lang": "en", "tgt_lang": "nl"})  does not work!
     end_time = time.time()
     elapsed_time = end_time - start_time
     print(f"Loading the model: {elapsed_time} seconds")