Workflows

Task-recipe guides organised around "I want to…" intentions. Workflows are terse and action-oriented — if you know what you're trying to do and need the exact incantation, this is the section.

If you're new and need step-by-step learning instead, start with Tutorials.

Build a pipeline

• Build Pipeline (Python)

  ---

  Author a pipeline programmatically with the Python API.

• Build Pipeline (YAML)

  ---

  Define a pipeline declaratively in YAML.

Train a pipeline

• Statistical Training

  ---

  Fit a pipeline using StatisticalTrainer — accumulate background moments, no gradient steps.

• Gradient Training

  ---

  Fit a pipeline using GradientTrainer — optimizer, scheduler, callbacks.

Run, monitor, and profile

• Restore Pipeline

  ---

  Replay a saved pipeline with restore-pipeline or restore-trainrun.

• Model Weights

  ---

  Pretrained plugin weights from the public Cubert mirrors: list, provision, run offline.

• Monitoring & Visualization

  ---

  TensorBoard, metric callbacks, and runtime visualisation of pipeline state.

• Profiling

  ---

  Measure where a pipeline spends its time and memory.