Why train dataloader is not prepared by Accelerator

[Jiaxin-Wen]

🐛 Describe the bug

trlx/trlx/trainer/accelerate_ppo_trainer.py

Line 231 in 3340c2f

return self.store.create_loader(self.config.train.batch_size, shuffle=True)

the train dataloader used in PPO is not prepared by Accelerator.

Then in the training phase (
(

trlx/trlx/trainer/accelerate_base_trainer.py

Line 563 in 3340c2f

for minibatch in MiniBatchIterator(train_dataloader, self.mb_size, self.num_mb):

)), will the minibatch_size be used as the batch size per device or not?

Which trlX version are you using?

No response

Additional system and package information

No response