class LabelSmoothingCrossEntropy(nn.Module):
    def __init__(self):
        super(LabelSmoothingCrossEntropy, self).__init__()
    def forward(self, x, target, smoothing=0.1):
        confidence = 1. - smoothing
        logprobs = F.log_softmax(x, dim=-1)
        nll_loss = -logprobs.gather(dim=-1, index=target.unsqueeze(1))
        nll_loss = nll_loss.squeeze(1)
        smooth_loss = -logprobs.mean(dim=-1)
        loss = confidence * nll_loss + smoothing * smooth_loss
        return loss.mean()

from utils import LabelSmoothingCrossEntropy

criterion = LabelSmoothingCrossEntropy()
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()