Polyaxon v3 is coming →

Train and fine-tune models with a complete run record

Package your training recipe, choose its compute, and compare the resulting metrics and checkpoints without separating them from the configuration that produced them.

Community Edition supports managed jobs and core tracking. Shared queues, managed sweeps, and model versioning use commercial capabilities.

Compare two fine-tuning configurations

An ML engineer runs the same model and dataset with different training settings. The team needs to inspect both results, retain checkpoints, and identify the exact configuration behind the selected candidate.

From training recipe to model candidate

  1. Define the recipe

    Record the model, data, code, and image revisions.

  2. Submit the job

    Select the resources and required connections.

  3. Compare runs

    Review logged metrics and checkpoint outputs.

  4. Select a candidate

    Keep the chosen artifact linked to its source run.

Illustrative workflow. The training framework produces the model; Polyaxon manages the operation and the results your code records.

Keep training choices separate from cluster settings

Start with one job, then choose how to distribute it

Compare results and keep the useful checkpoints

Make the selected output traceable

When this workflow fits