verify-training-pipeline

To validate a machine learning training pipeline end-to-end, this Skill checks the standardized example layout, executes the Docker build, runs the GPU training, and verifies that meaningful metrics and checkpoints are logged successfully.

Verifying a CVlization training example requires a standardized layout including an example.yaml file, a Dockerfile, build.sh, train.sh, and train.py to pass structure validation before executing the build and training checks.

Yes, you can use this to debug training failures in a Docker GPU environment by running a repeatable verification workflow that tests the build, checks the GPU setup, and validates the training execution and metric logging.

Pipeline verification supports CI checks for machine learning templates by providing a repeatable workflow that validates structure, build, training execution, and metric logging outputs for new or modified training examples.

To ensure training logs and metrics are properly captured, the verification process checks that training outputs, including checkpoints and logs, are produced and that meaningful metrics are successfully captured and validated.

Yes, dataset and weight downloads are cached during the training verification workflow, ensuring lazy downloading and caching are utilized to optimize iterations and reuse resources across multiple verification runs.