I am using accelerate
for a training loop followed by an evaluation loop. before these loops, I instantiate a wandb
tracker through accelerate
. After the evaluation loop, I log the results to the wandb
tracker. Is it recommended to call accelerator.end_training()
after logging everything from the evaluation loop to wandb
or directly after the training loop?
It should be the last thing done in your script as it will close all the experiment trackers
1 Like
@muellerzr perfect! Thank you!
1 Like