You can use WandB to track your training runs to gain useful insights into metrics like the eval loss and VRAM usage.

wandb login

Each TrainArgs child class, such as GENTrainArgs has three parameters that are used to control the WandB tracking.

Parameter Default Meaning
report_to () Platforms where metrics will be saved to
project_name “happy-transformer” Project name within WandB
run_name “test” run name within WandB
from happytransformer import GENTrainArgs

args = GENTrainArgs(report_to=tuple(["wandb"]), 
                    )