【PyTorch】报错场景及解决方案汇总

Backto PyTorch Index

Error	Case	Solution	Reason
RuntimeError: Error(s) in loading state_dict for ResNet:	load_state_dict()	model.load_state_dict( checkpoint[‘state_dict’], strict=False)	version not match
AssertionError: Invalid device id	using GPU	in config.py, set GPU_ID correctly	gpu devices misfound error
Volatile = now has no effect. Use with torch.no_grad(): instead	image = Variable(image, volatile=True)	with torch.no_grad(): image = Variable(image)	removed
“non-inplace resize is deprecated”	tensor.resize()	change to tensor.reshape or torch.reshape	deprecated
RuntimeError: An attempt has been made to start a new process before the current process has finished its bootstrapping phase.	runtime error	1. 去掉num_workers参数 ； 2. (推荐)在跑epoch之前，加上if __name__==__main__:	import module导致重复执行
Torch: not enough memory: you tried to allocate 0GB. Buy new RAM!	error occurs half-way	add swap memory; Don NOT run other memory-cosuming apps, like Chrome/PyCharm, etc	Memory is valuable & limited.
RuntimeError: Expected object of backend CPU but got backend CUDA for argument #2 ‘weight’	testing	with torch.no\_grad():input = input.cuda();net.cuda();net(input)	Mixed usage of CPU/GPU