train with multi-gpus

[pokameng]

@lllyasviel @williamyang1991 @eltociear @camenduru hello bro, Is it possible to train the tutorial_train_sd21.py with multi-GPUs? I have modified the trainer = pl.Trainer(gpus=1, precision=32, callbacks=[logger]) to trainer = pl.Trainer(gpus=4, accelerator='dp,precision=32, callbacks=[logger]) but it doesn't work

Can someone help me?

[sethupavan12]

What was the error?

[XavierCHEN34]

I get the same problem

[soon-yau]

You'll need to add the distributed strategy to avoid multiple gpus accessing the same files. I used the following arguments and it works for me: trainer = pl.Trainer(strategy="ddp", accelerator="gpu", devices=2, precision=32, callbacks=[logger])

[ningbende]

solve my problem， thx！

You'll need to add the distributed strategy to avoid multiple gpus accessing the same files. I used the following arguments and it works for me: trainer = pl.Trainer(strategy="ddp", accelerator="gpu", devices=2, precision=32, callbacks=[logger])