atria_cli is the command-line entry point for Atria Hub operations. It exposes authentication, dataset publishing, and model snapshot management as CLI commands — each backed by the same Python API used programmatically.

Commands

atria
├── sign_in          ← Authenticate with Atria Hub (email + password)
├── sign_out         ← Remove stored credentials
├── sign_up          ← Create a new hub account
├── datasets
│   ├── prepare_and_upload  ← Prepare a registered dataset and push to hub
│   └── download            ← Pull a dataset artifact from hub
└── models
    ├── upload       ← Upload a model snapshot directory to hub
    └── download     ← Pull a model snapshot from hub

Implementation

atria_cli uses Python Fire to turn the command dict into a CLI. Each command is a plain Python function that receives CLI arguments:

fire.Fire({
    "sign_in": sign_in,
    "sign_out": sign_out,
    "sign_up": sign_up,
    "datasets": {"prepare_and_upload": ..., "download": ...},
    "models": {"upload": ..., "download": ...},
})

This keeps commands as regular functions testable in isolation, with Fire handling argument parsing.

Training and evaluation

Training and evaluation are not CLI commands — they are run directly via Python using Trainer, Evaluator, and ModelExplainer with the appropriate TaskConfig. The CLI handles only hub operations (authentication, artifact push/pull).