#concept-pamphlet #todo: pytorch terminology is important
x.shape
x.dtype
.sum
torch.Generator().manual_seed(234235235)
torch.multinomial(p, num_samples=1, replacement=True, generator=g).item()
torch.tensor(x)
import torch.nn.functional as F xenc = F.one_hot(xs, num_classes=27).float()
torch.randn((27, 1))
torch.zeros(5)
W.grad
loss= -probs[torch.arange(5), ys].log().mean() loss.backward()