How to use Transformer Trainer Training Arguments report_to
method in Accelerator? Do I need to make manually calculate each data like loss etc. and send to tensorboard or wandb?
Hello @Indramal, yes, you are correct. Please refer to the *no_trainer.py
examples here transformers/examples/pytorch at main · huggingface/transformers · GitHub.
For example, run_glue_no_trainer.py
has arg report_to
transformers/run_glue_no_trainer.py at main · huggingface/transformers · GitHub. You can log to the trackers via accelerator.log
as shown here transformers/run_glue_no_trainer.py at main · huggingface/transformers · GitHub.
Hope this helps. Let us know if this resolves your query.
1 Like
Thank you very much i will try it and get back to you