modal-labs/llm-finetuning

Can I use train.py on CodeLlama-13b or CodeLlama-34b

luo647 opened this issue · 4 comments

luo647 commented
Can I use train.py on CodeLlama-13b or CodeLlama-34b

I also have the same question.

I haven't personally tried it, but from what I can see in the code you just need to add the other models to BASE_MODELS in common.py and invoke train.py with --base set to whatever new key you added.

timiil commented

I also have the same question, what hardware (min GPU ram required) can be fine tuning CodeLlama-34b ?

We now ship this with several example configuration files demonstrating how to use different models!