instruction fine tuning template
hi team, I would like to know which prompt template i should use? I wanted to do instruction fine-tuning I'm different fine-tuning template in the internet.
context = "Virgin Australia, the trading name of Virgin Australia Airlines Pty Ltd, is an Australian-based airline. It is the largest airline by fleet size to use the Virgin brand. It commenced services on 31 August 2000 as Virgin Blue, with two aircraft on a single route. It suddenly found itself as a major airline in Australia's domestic market after the collapse of Ansett Australia in September 2001. The airline has since grown to directly serve 32 cities in Australia, from hubs in Brisbane, Melbourne and Sydney."
input_question = "When did Virgin Australia start operating?"
instruction = "You're a helpful chatbot. Please read the give question, and context and answer the question precisely. If you there is no answer present in the context please do not make up on your own, just say I don't know."
response = "Virgin Australia commenced services on 31 August 2000 as Virgin Blue, with two aircraft on a single route"
##First Prompt
text = f"### Instruction: {instruction}\n ### Context: {context}\n ### Question: {input_question}\n ### Answer: {response}"`
and
##second Prompt
template = {
"prompt": "Below is an instruction that describes a task, paired with an input that provides further context. "
"Write a response that appropriately completes the request.\n\n"
"### Instruction:\n{instruction}\n\n### Input:\n{context}\n\n",
"completion": " {response}",
}
and
##third Prompt
"<s>[INST]{instruction}\n Here is some context: {context}.\n [/INST] {response}</s>"
which one is the suitable prompt for instruction fine tuning?
and another questions is, as I wanted to do instruction fine-tuning which mistral model should i pick? should I pick base model(mistralai/Mistral-7B-v0.1) or instrcution model(mistralai/Mistral-7B-Instruct-v0.2)?
@devendrachaplot
@arthurmensch
I have the same question about how to fine tuning by instruction prompt. I try fine-tuning by using the First Prompt, but the effort is not good.
@devendrachaplot
@arthurmensch
3rd prompt works for me, you can also use alpaca format which works great for fine-tuning the base model