What is the best LLM for finetuning with specific repetetive data?

Hello,
I am new to running LLMs localy and I would like to Finetune LLM for processing social media posts.
The “instruction” will always be the same. only the “input” will change based on the social media post.

Instruction will always be to summarize the post in less than 3 words and give rating of the post. from 1 worse to 5 best.

With that in mind I am trying to find the best LLM to finetune.
I ve read a lot of articles where they say to use lower parameter models that could excel at such tasks and aditionaly require less training data than bigger models.

But the answers are not clear

How many examples of training data do I need and what model should i choose?
Some websites say 100k while GPT says you need 100 examples to finetune (theyr paid api not local LLM).

My first picks were llama 3.1 8b or 3b

If someone knows it would be a big help <3

Thank you for your answers

1 Like