Polyaxon v3 is coming →

Automate training without losing the review step

Connect data preparation, training, and evaluation in a repeatable workflow. Define which results can move forward and where a person must approve the next operation.

Uses the commercial workflow engine, schedules, and model-version capabilities in Polyaxon Cloud or self-hosted Enterprise.

A recurring model refresh with a quality gate

An ML team trains on an updated dataset and evaluates against a fixed validation set. Only candidates that meet the team's condition proceed to a review operation; unsuccessful candidates retain their results for investigation.

Training and evaluation with a controlled handoff

  1. Prepare

    Produce and record the dataset reference.

  2. Train

    Save the candidate model and checkpoint outputs.

  3. Evaluate

    Run the team's evaluation code and log its result.

Evaluation succeeded and the condition is met?

The downstream operation uses the recorded score and the configured dependency rules.

Yes — wait for review

Require approval before running the registration operation.

No — do not promote

Inspect the score or failure and keep the previous selected version.

Illustrative configuration, not an executed pipeline. A completed evaluation job and an acceptable model score are separate checks.

Define what each operation produces

Distinguish a successful job from an acceptable model

Require approval before the handoff

Decide how repeats and failures should behave

Automate an established process first