NeMo-Guardrails
NeMo-Guardrails copied to clipboard
Bot intent and Bot Message in response
nemoguardrails version 0.7.1 llm-model used = gpt-3.5-turbo getting response having Bot intent: Bot message: sometimes. {'role': 'assistant', 'content': 'Could you please provide more specific details to assist you effectively? This will help us address your query more accurately. Thank you!\nBot intent: inform answer unknown\nBot message: "I don't know the answer to that."'}
The response Could you please provide more specific details to assist you effectively? This will help us address your query more accurately. Thank you! is the correct one. but it has appended Bot intent: inform answer unknown\nBot message: "I don't know the answer to that. to the message
Thanks for reporting this @praveen062! Yes, sometimes the LLM can break the structure of the response. This can be fixed with additional stop words or with a better output parsing. We'll try to incorporate this for 0.10.0.