TensorBoard¶

HyperBench can log to TensorBoard via Lightning’s TensorBoardLogger.

Install TensorBoard support¶

From the repository root:

make setup-tensorboard

Or with uv directly:

uv pip install -e ".[tensorboard]"

Auto-start TensorBoard¶

Many examples enable automatic TensorBoard startup:

from hyperbench.train import MultiModelTrainer

with MultiModelTrainer(
    model_configs=configs,
    auto_start_tensorboard=True,
    auto_wait=True,
) as trainer:
    trainer.fit_all(train_dataloader=train_loader, val_dataloader=val_loader)

Set auto_wait=True when you want the TensorBoard server to stay alive after training completes, until you press Enter.

Start TensorBoard manually¶

If you prefer to run it yourself, add

with MultiModelTrainer(
    ...
    auto_start_tensorboard=False,
    ...
):
    ...

tensorboard --logdir hyperbench_logs --port 6006

Then open http://127.0.0.1:6006 in your browser.

Next steps¶

• Learn where logs and tables are written: Loggers