pytorch_ lightning. utilities. exceptions. MisconfigurationException: You requested GPUs: [1] But...

Report errors

pytorch_lightning.utilities.exceptions.MisconfigurationException: You requested GPUs: [1] But your machine only has: [0]

The author's error reporting code ：

trainer = Trainer(
                        max_epochs= config.max_epochs,  
                        gpus= [fix_config.hparams.gpus],   
                        distributed_backend= fix_config.hparams.distributed_backend,  
                        benchmark= fix_config.hparams.benchmark,  
                        deterministic= fix_config.hparams.deterministic,  
                        logger=logger,
                        callbacks=[
                            early_stop_callback,
                            checkpoint_callback
                        ]
                      )

if __name__ == "__main__":
    #  Create a parser 
    parser = ArgumentParser()  # Writing a command line interface 
	parser.add_argument("--gpus", default=1, type=int)  #  Error source 
	...

This is the use of pytorch_lightning Error encountered in framework . Error reporting theory , What do you want? GPUs The number of [1], But your machine only has [0] individual .

To see this , I believe most people's first reaction is ：

• hey ！ Report the mistake that I did not gpu！
• Then go and check cuda
  
  notes ： If it shows cuda by Flase, Means you Not loaded cuda perhaps torch It 's loaded with cpu edition , It needs to be reinstalled . The author is in This note Configuration is introduced in cuda And corresponding gpu Version of pytorch、torchvision、torchaudio The bitterness of life , As a reference .
• cuda It can be used ！ So why does it say I only have [0] Well ?

At first, the author couldn't understand , After looking up the meagre data , Find out The author's conjecture It may be right ： in other words ,gpu The quantity is from 0 Start counting ！

thus , Re translate the mistakes ： Do you want 2 individual GPU, But your machine only has 1 individual GPU.

terms of settlement

Put the above code in ,gpu Change the quantity to 0.

...
parser.add_argument("--gpus", default=0, type=int)