1. tensorboard

from torch.utils.tensorboard import SummaryWriter# TensorBoard
writer = SummaryWriter('runs/mnist_experiment_1')
...if i % 100 == 99:    # 每 100 个 batch 打印一次并记录到 TensorBoardprint('[%d, %5d] loss: %.3f' %(epoch + 1, i + 1, running_loss / 100))writer.add_scalar('Training Loss',running_loss / 100,epoch * len(trainloader) + i)writer.add_scalar('Accuracy',correct / total,epoch * len(trainloader) + i)running_loss = 0.0total = 0correct = 0# 不要忘了关闭 SummaryWriter
writer.close()然后终端启动命令： tensorboard --logdir=runs

2. F.cross_entropy(input_tensor, target) = F.log_softmax() + F.nll_loss()

两者 计算的loss 相同

import torch
import torch.nn.functional as F
import torch.nn as nn# 定义一个输入张量，形状为 (B, C)
input_tensor = torch.randn(3, 5)  # 假设批量大小为 3，类别数量为 5target = torch.LongTensor([1, 0, 2])# 使用 log_softmax 函数计算对数softmax值
output_tensor = F.log_softmax(input_tensor, dim=1)# print("Input Tensor:")
# print(input_tensor)print("\nOutput Tensor (Log-Softmax):")
print(output_tensor)# 计算 NLL Loss
loss = F.nll_loss(output_tensor, target)
print("NLL Loss:", loss.item())# 计算交叉熵损失
loss = F.cross_entropy(input_tensor, target)print("Cross-Entropy Loss:", loss.item())