安装与配置：
pip install tensorboard在代码中导入并初始化SummaryWriter：
from torch.utils.tensorboard import SummaryWriterwriter = SummaryWriter("runs/experiment_1") # 日志保存路径记录指标：
标量数据（损失、准确率等）：for epoch in range(epochs): train_loss = ... val_acc = ... writer.add_scalar("Loss/train", train_loss, epoch) writer.add_scalar("Accuracy/val", val_acc, epoch)学习率记录（结合调度器）：writer.add_scalar("Learning Rate", optimizer.param_groups[0]['lr'], epoch)模型参数分布（直方图）：for name, param in model.named_parameters(): writer.add_histogram(name, param, epoch)启动TensorBoard：
tensorboard --logdir=runs/调试工具与技巧检测NaN/Inf异常：
torch.autograd.set_detect_anomaly(True)with torch.autograd.detect_anomaly(): loss.backward() # 反向传播时检测异常，会显著降低速度，仅调试时使用梯度检查：
注册钩子（Hook）监控梯度：def gradient_hook(grad): if torch.any(torch.isnan(grad)): print("NaN梯度 detected!")for param in model.parameters(): param.register_hook(gradient_hook)统计梯度范数：total_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)writer.add_scalar("Gradient Norm", total_norm, epoch)中间变量调试：
使用pdb设置断点：import pdb; pdb.set_trace() # 在需要检查的位置插入保存中间结果：torch.save({"tensor": intermediate_tensor}, "debug_tensor.pt")数据与模型验证数据加载检查：
# 可视化一个batch的数据（图像示例）images, labels = next(iter(train_loader))grid = torchvision.utils.make_grid(images)writer.add_image("Sample Images", grid, 0)模型结构可视化：
dummy_input = torch.randn(1, 3, 224, 224) # 输入示例writer.add_graph(model, dummy_input)过拟合/欠拟合分析监控训练与验证损失：过拟合：训练损失下降，验证损失上升。欠拟合：两者均高且停滞。调整策略： 增加数据增强、正则化（Dropout、权重衰减），或简化模型结构。学习率调度与记录scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=30, gamma=0.1)for epoch in range(epochs): train(...) scheduler.step() writer.add_scalar("Learning Rate", optimizer.param_groups[0]['lr'], epoch)高级工具：PyTorch Lightning简化日志和调试流程：
import pytorch_lightning as plclass LitModel(pl.LightningModule): def training_step(self, batch, batch_idx): loss = ... self.log("train_loss", loss) # 自动记录到TensorBoard return losstrainer = pl.Trainer()trainer.fit(model, train_loader)
