Currently, we have the model outputs saved in the path of results/<exp_name>/predictions_step_0_rank_0.jsonl. But if we run the experiment with the same name, it will be overridden. So it's better to save it under the run name given by wandb.
Ideally, we want to save the checkpoints in the wandb folder as well
Currently, we have the model outputs saved in the path of
results/<exp_name>/predictions_step_0_rank_0.jsonl. But if we run the experiment with the same name, it will be overridden. So it's better to save it under the run name given by wandb.Ideally, we want to save the checkpoints in the wandb folder as well