stuck at epoch 0 ,I try to use one or two gpus on single machine, then stucl

[kongzijian]

trainging set is : dataloader = DataLoader(dataset1, num_workers=0, batch_size=batch_size, shuffle=True) from pytorch_lightning.callbacks.progress import TQDMProgressBar trainer = pl.Trainer(gpus=n_gpus, strategy="ddp", precision=16, accelerator="gpu", callbacks=[TQDMProgressBar(refresh_rate=1)], accumulate_grad_batches=accumulate_grad_batches,log_every_n_steps=20)

[XavierCHEN34]

There may be some bugs for your dataloader. We add "try except" in our dataset base.py to skip some samples. You could remove the try except to debug.