---

name: monitor description: View training progress in real-time. Use when checking on running training, viewing live metrics, or opening dashboards.

Live Training Monitor

View training progress in real-time.

Step 1: Identify Logging Backend

File-based, TensorBoard, W&B, or terminal output

Step 2: Launch Monitor

Terminal Dashboard (ASCII)

Live updating with step progress, loss, LR, gradient norm, ASCII chart

TensorBoard

Launch with tensorboard --logdir ./logs --port 6006

Weights & Biases

Get run URL from wandb.init() or list recent runs

Real-time Plot

Matplotlib animation with auto-refresh

Step 3: Set Up Alerts

Detect loss spikes, NaN, gradient explosion, overfitting

Quick Commands

• tail -f training.log
• watch -n 1 nvidia-smi
• tmux combined view