Trials

A trial is one parameter combination within a sweep. Each trial is a single training job with a specific set of hyperparameters. The sweep orchestrator creates trials, runs them, and tracks the objective metric to find the best configuration.

Preview

The Sweeps tab is in early access. Trial browsing is available in the UI, but trial creation is managed by the sweep orchestrator.

The trials list

The list view shows every trial across all sweeps. Each row displays:

• Name — the trial's identifier.
• Sweep — which sweep this trial belongs to.
• Parameters — the hyperparameter values for this trial.
• Metric — the objective metric value (if the trial completed).
• Status — current phase (Running, Succeeded, Failed).
• Created — when the trial was created.

Click a trial to see its full parameter set, metric history, and logs.

Best trial

The sweep detail page highlights the best trial — the one with the highest (or lowest, depending on the objective) metric value. This is the configuration you'd use for your production model.

Next steps

• Sweeps — the parent experiments
• Algorithms — how trials are generated