Comment by binocarlos

Excellent link thank you - I will make sure we check this out because as the README says:

> Finetune Mistral, Llama 2-5x faster with 70% less memory!

Could be very useful for us!

Disclaimer: I work on Helix