A minimal demonstration of LoRA on GPT2.

Examples:

Basic:

python run.py -w --iterations-to-eval 5 --num-training-steps 10 --quick-eval-one-batch

With downloaded model & logging on wandb.ai:

python run.py -n path\to\model -k \path\to\tokenizer -u [wandb-username] --iterations-to-eval 5 --num-training-steps 10 --quick-eval-one-batch