Fine tuning LLM
# supportf
floatingpaper
02/09/2025, 9:16 AMHey guys
I am fine tuning an OpenAI LLM.
If the fine tuning .jsonl file only has system, user and assistant prompts,
eg
{
"messages": [
{
"role": "assistant",
"content": "Let me check the current weather for you. Please hold on..."
}
Would the VAPI voice agent understand over time to call the tool when that is said based on the description we set on the tools in VAPI?
Would that work even though I'm not using this data in the fine tuning process? (Btw, should I be using this during fine tuning?):
{
"role": "assistant",
"content": "",
"function": {
"name": "get_weather",
"parameters": {
"location": "London"
},
"description": "Retrieves the current weather for the specified location."
}
}
],
"async": false,
"server": {
"url": "https://n8n.example.com/weather"
}
Would love some insight if you may
c
Chiranjeet Mishra
02/10/2025, 5:11 AMHey, I would suggest you directly write the description in the prompt and use the GPT-4o model. The reason I am suggesting you do this instead of fine-tuning the model is that, at the pace at which AI is evolving, you will have new models in a couple of months that will be far better than this, and it won't be that useful.
f
floatingpaper
02/10/2025, 3:29 PMYou're saying add this in the system prompt?
floatingpaper
02/10/2025, 3:29 PMeg "You are a booking agent... When the user says that they'd like to book an appointment. call This tool:
floatingpaper
02/10/2025, 3:29 PM{
"role": "assistant",
"content": "",
"function": {
"name": "get_weather",
"parameters": {
"location": "London"
},
"description": "Retrieves the current weather for the specified location."
}
}
],
"async": false,
"server": {
"url": "https://n8n.example.com/weather"
}
floatingpaper
02/10/2025, 3:29 PM...etc
floatingpaper
02/10/2025, 3:29 PMIs this what you're saying?
c
Chiranjeet Mishra
02/10/2025, 5:25 PMjust say call this tool get_weather and pass the parameter location.
2 Views