Skip to content

Package

The goal of this stage is to hand off ML model from ML engineer to MLOps engineer by documenting the pre-processing, training/validation, testing, inference, and post-processing steps. For example, converting an output artifact using package.py to the format below:

output/<run-id>/
├── model.pt              # weights/state dict
├── metadata.json         # model name, version, created_at, framework, git sha, etc.
├── preprocessing.json    # image_size, mean, std
├── metrics.json          # accuracy, baselines, per-class metrics
└── class_to_idx.json     # class mapping