checkpoint - chatbots work, but weird

app.py

@@ -5,8 +5,8 @@ import os
 import openai
 
 # Load the Vicuna 7B model and tokenizer
-vicuna_tokenizer = AutoTokenizer.from_pretrained("lmsys/vicuna-7b-v1.5")
-vicuna_model = AutoModelForCausalLM.from_pretrained("lmsys/vicuna-7b-v1.5")
+vicuna_tokenizer = AutoTokenizer.from_pretrained("lmsys/vicuna-7b-v1.3")
+vicuna_model = AutoModelForCausalLM.from_pretrained("lmsys/vicuna-7b-v1.3")
 
 # Load the LLaMA 7b model and tokenizer
 llama_tokenizer = AutoTokenizer.from_pretrained("luodian/llama-7b-hf")
@@ -50,13 +50,13 @@ def gpt_respond(tab_name, message, chat_history, max_convo_length = 10):
     formatted_prompt = format_chat_prompt(message, chat_history, max_convo_length)
     print('Prompt + Context:')
     print(formatted_prompt)
-    bot_message = chat(system_prompt = f'''Generate the output only for the assistant. Output any <{tab_name}> in the following sentence one per line without any additional text.''',
+    bot_message = chat(system_prompt = f'''Generate the output only for the assistant. Please output any <{tab_name}> in the following sentence one per line without any additional text.''',
                        user_prompt = formatted_prompt)
     chat_history.append((message, bot_message))
     return "", chat_history
 
 def vicuna_respond(tab_name, message, chat_history):
-    formatted_prompt = f'''Output any {tab_name} in the following sentence one per line without any additional text: {message}'''
+    formatted_prompt = f'''Generate the output only for the assistant. Please output any {tab_name} in the following sentence one per line without any additional text: {message}'''
     print('Prompt + Context:')
     print(formatted_prompt)
     input_ids = vicuna_tokenizer.encode(formatted_prompt, return_tensors="pt")
@@ -69,7 +69,7 @@ def vicuna_respond(tab_name, message, chat_history):
     return tab_name, "", chat_history
 
 def llama_respond(tab_name, message, chat_history):
-    formatted_prompt = f'''Output any {tab_name} in the following sentence one per line without any additional text: {message}'''
+    formatted_prompt = f'''Generate the output only for the assistant. Please output any {tab_name} in the following sentence one per line without any additional text: {message}'''
     print('Prompt + Context:')
     print(formatted_prompt)
     input_ids = llama_tokenizer.encode(formatted_prompt, return_tensors="pt")