Closed Xiangxu-0103 closed 1 month ago
yeah pytorch_lighting version 2.0.8 does not support these arguments. I got around it by doing: ` trainer = Trainer( accelerator="gpu", # This specifies that GPUs should be used devices=1, strategy="ddp", # Distributed Data Parallel strategy
# accelerator="ddp",
logger=tb_logger,
max_epochs=cfg.TRAIN.MAX_EPOCH,
callbacks=callbacks,
log_every_n_steps=1,
gradient_clip_val=0.5,
accumulate_grad_batches=cfg.TRAIN.BATCH_ACC,
# resume_from_checkpoint=ckpt,
)
trainer.fit(model, data,ckpt_path = ckpt)
`
Hi! sorry about the delay and the confusion, I checked again and I was using an earlier version of pytorch_lightning: 1.5.9.
I know in the newer version of pytorch_lightning some things changed and some arguments do not work anymore.
However, you can find an easy workaround with the new keyworkds.
resume_from_checkpoint
is now called ckpt_path
and you can specify accelerator='gpu'
and devices=1
.
Hi, thanks for your open source. When I train Mask4D with
pytorch_lightning==2.0.8
, it will raise errors as follows:and
Looking forward to hearing back from you.