Correct format for fine-tuning

#121
by engrzulqarnain - opened

i want to finetune mistralai/Mistral-7B-Instruct-v0.2 what is the correct format of input sequence to the model: in my custom data set have two columns: "Instructions" and "Answer" the instructions contain such as queries "write a function to add two numbers" and the Answer contain the answer of the given query like: def add(num1,num2): return num1+num2

i used this one format is its correct or not

def create_prompt(sample):
bos_token = ""
original_system_message = "Below is an instruction that describes a Pine Script task. Write a response in Pine Script language that appropriately completes the request"
input = sample["Instructions"]
eos_token = "
"

full_prompt = ""
full_prompt += bos_token
full_prompt += "### Instruction:"
full_prompt += "\n" + original_system_message
full_prompt += "\n\n### Input:"
full_prompt += "\n" + input
full_prompt += "\n\n### Response:"
full_prompt += "\n" + sample["code"]
full_prompt += eos_token

return full_prompt

now I want how to format it for fine-tuning @arthurmensch ,@teven

Sign up or log in to comment