I am trying to train a model in Google Colab using TPU distributed training. The dataset is my own and few functions: get_train_dataloader and get_eval_dataloader are overloaded to use batch sampler. And I spawn the process using following code: xmp.spawn(_mp_fn, args=(), nprocs=8, start_method=‘f…

Is wandb in Trainer configured for distributed training?

brando August 18, 2022, 3:01pm 2

I am having an issue that all 8 processes are logging. Do you have that issue too?

Topic		Replies	Views
Can't understand the graphs logged by `wandb` 🤗Transformers	0	164	September 12, 2023
Wandb.watch in accelerate library 🤗Accelerate	5	1749	July 18, 2022
How to turn WanDB off in trainer? Beginners	11	27740	March 9, 2024
Do Trainer and Callback get created multiple times in case of distributed setup 🤗Accelerate	0	123	January 11, 2024
Get stuck when using wandb to evaluate Beginners	1	567	June 8, 2023