TensorBoard
A local dashboard for visualizing training metrics, experiment comparisons, and model diagnostics.
Published by TensorFlow Project
How to use it
Write metrics such as loss, validation scores, and learning rate to separate runs. Compare aligned steps or time periods so training schedules and evaluation intervals stay comparable.
Run TensorBoard against the output log directory. Use the same validation dataset and metric names across runs when checking whether a change improved training.
- Compatibility
- Python, TensorFlow, PyTorch
- Monitoring areas
- Training monitoring, Scalar metrics, Experiment comparison
- Access
- Local Python package and dashboard.