rank 16 lora of llama 3 70b 4bit bnb, trained locally on 2x 3090 Tis
my first attempt at local multi-gpu QLoRA training where we train a model that does not fit on single GPU
it completed just 15% of the epoch on HESOYAM 0.4 dataset when I left it overnight, loss didn't go down much
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support