This is the command I used for testing the issues mentioned from

the argument errors and quantinize not working

python examples/gralora_finetuning/gralora_finetuning.py \
--base_model hf-internal-testing/tiny-random-LlamaForCausalLM \
--output_dir ./test_output \
--quantize \
--num_epochs 1 \
--batch_size 1 \
--save_step 1000 \
--eval_step 5 \
--device cuda

The log part doesn't cause problem under my test. However, I noticed that no log output appeared until I installed tensorboard.
Also, after adding --quantize, I initially got an error because bitsandbytes wasn’t installed — installing it fixed the problem.