以 pytorch cifar-10 代码为例

for epoch in range(5):  # 循环遍历数据集多次
    for i, data in enumerate(trainloader, 0):
        # 得到输入数据
        inputs, labels = data
        # 包装数据
        inputs, labels = Variable(inputs), Variable(labels) 
        # 梯度清零
        optimizer.zero_grad()
        # net()为网络模型，通过模型得到输出
        outputs = net(inputs)
        # 计算损失
        loss = criterion(outputs, labels)
        # 反向传播
        loss.backward()
        # 参数优化
        optimizer.step()
        loss_value = loss.data[0]
    # 预设batchsize=100    
    outputs = torch.cat((outputs.data, torch.ones(len(outputs), 1)), 1)
    inputs = inputs.to(torch.device("cpu"))
    # outputs.size():[100,11]///labels.size():[100]///inputs.size():[100,3,32,32]
    writer.add_embedding(outputs, metadata=labels.data, label_img=inputs.data, global_step=epoch)
    writer.add_embedding(outputs, metadata=labels.data, global_step=epoch)
    writer.add_embedding(outputs, label_img=inputs.data, global_step=epoch)
    writer.add_scalar('loss_value',loss_value,epoch)
writer.add_graph(net,(inputs,))
print('Finished Training')
writer.close()

writer.add_embedding(outputs, metadata=labels.data, label_img=inputs.data, global_step=epoch)

writer.add_embedding(outputs, metadata=labels.data, global_step=epoch)

writer.add_embedding(outputs, label_img=inputs.data, global_step=epoch)

writer.add_scalar(‘loss_value’,loss_value,epoch)

writer.add_graph(net,(inputs,))

writer.add_*源代码：https://github.com/lanpa/tensorboard-pytorch/blob/master/tensorboardX/writer.py
tensorboardX官方指导文档：http://tensorboard-pytorch.readthedocs.io/en/latest/tutorial_zh.html
tensorboardX官方github：https://github.com/lanpa/tensorboard-pytorch
参考代码段详见：