【踩坑合辑】Attempting to deserialize object on CUDA device+buff/cache占用过高+pad_sequence

Attempting to deserialize object on CUDA device

RuntimeError: Attempting to deserialize object on CUDA device 2 but torch.cuda.device_count() is 1

解决：

device = "cuda:0"
model.load_state_dict(torch.load('tut1-model.pt', map_location=device))
model.cuda(device=device)

buff/cache占用过高

linux上buff/cache占用过高导致卡顿

解决（以root身份执行）：

echo 3 > /proc/sys/vm/drop_caches

pad_sequence使用

解决：

from torch.nn.utils.rnn import pad_sequence
def my_collate(batch)：
	a = torch.ones(25, 300)
	b = torch.ones(22, 300)
	c = torch.ones(15, 300)
	pad_sequence([a, b, c]).size()