Hi there,Is there any specific format for user input in the finetunning and user input should be the descriptive or is it okay to have the short about the assistant.Please let me know.
From the documentation here OpenAI Platform
is this example
{"messages": [{"role": "system", "content": "Marv is a factual chatbot that is also sarcastic."}, {"role": "user", "content": "What's the capital of France?"}, {"role": "assistant", "content": "Paris, as if everyone doesn't know that already."}]}
{"messages": [{"role": "system", "content": "Marv is a factual chatbot that is also sarcastic."}, {"role": "user", "content": "Who wrote 'Romeo and Juliet'?"}, {"role": "assistant", "content": "Oh, just some guy named William Shakespeare. Ever heard of him?"}]}
{"messages": [{"role": "system", "content": "Marv is a factual chatbot that is also sarcastic."}, {"role": "user", "content": "How far is the Moon from Earth?"}, {"role": "assistant", "content": "Around 384,400 kilometers. Give or take a few, like that really matters."}]}
1 Like
my question is about…
Ex- I have fine tunned model with my own data.In that user input is descriptive and after the training i am giving the short prompt.
during fine tunning:-
user : Design a specialized button component known as the “Button”. This button includes specific characteristics:
Caption: “Confirm”
Type: “submit”
Class: “btn-default btn-lg” (styled as “btn-default” and “btn-lg” class)
Margin: “unset” (no specific margin applied)
The “Caption” is the visible text, “Type” signifies the button’s function, and “Class” and “Margin” are utilized for styling purposes.
assistant : '
so after training i am giving the input prompt as the “Generate the Confirm button”
so is it right or what? if not what should be fine tune input and what should be the promt to the finetunned model.
Your training data contains a detailed prompt and does not contain an assistant output, it’s just blank, you need to give the model an example of what a human would say “user” and then what the AI should say as an ideal reply “assistant”. If you do that many times the model will learn to follow along.
1 Like
Okay @Foxabilo.Suppose,
case 1: I finetuned the model with descriptive user prompt and after training i asked with the short prompt.
case 2: I finetuned the model with short user input and after fine tunning i asked the descriptive input in the playground during the testing.
whether 2 cases are possible or what ever i trained on the model with data like that only user input should during the testing?
If the training data was using detailed long user inputs and you test the trained model with terse short ones, you will usually get poor results, you can train the model with a combination of both sort and long inputs and the same output, that will allow the model to perform well for both.
Okay @Foxabilo.
Question1.whether in this system will affect or not.
Question2. What should be the data size for the more accurate result?
If you intend to use a system prompt to set a persona or primary instructions, then yes, include it in the training data.
There is no optimal size, it’s whatever your use-case size will be in the finial product, give a range of sizes to cover all eventualities if you need flexability.