Quickstart#

Publish and retrieve your first versioned repository on MEGA Hub. This path uses a small model card so you can verify authentication, repository creation, upload, revision history, and download before moving large artifacts.

Install the CLI#

Install the published mega command with one of the supported Python tool runners:

1uv tool install megatensors2# or: pipx install megatensors3# or: python -m pip install megatensors

Confirm the command is available:

1mega version2mega --help

Sign in#

Start the browser device flow, then confirm the active identity:

1mega auth login2mega auth whoami

The output shows the account handle you can use as OWNER below. For a headless environment, set MEGA_TOKEN or use mega auth login --token "$MEGA_TOKEN" instead. See Authentication for token scopes and automation guidance.

Create a repository#

Create a public model repository. Replace OWNER with your account or organization handle:

1mega repos create OWNER/hello-mega \2  --type model \3  --public \4  --description "My first MEGA Hub repository" \5  --exist-ok

Use --private instead when the release should only be visible to authorized members and tokens.

Upload a first revision#

Create a minimal repository card locally:

1mkdir -p hello-mega2printf '# Hello MEGA\n\nMy first versioned release.\n' > hello-mega/README.md

Upload the folder to the repository root:

1mega models upload OWNER/hello-mega ./hello-mega . \2  --commit-message "Publish first revision"

Every successful upload creates an immutable commit and advances the selected branch. Uploading a directory does not delete unrelated remote files unless you explicitly pass --sync.

Inspect and download#

Inspect the repository and its current file list:

1mega repos info OWNER/hello-mega2mega models list OWNER/hello-mega --revision main

Download the full snapshot into a clean directory:

1mega models download OWNER/hello-mega \2  --revision main \3  --local-dir ./downloaded-release

For reproducible consumers, replace main with a release tag or commit ID. See Hub Repositories for branches, tags, Git, large uploads, and revision history.

Publish a real artifact#

The Hub stores ordinary repository files as well as MEGA-native model artifacts. A production model release normally includes its card, configuration, tokenizer, model.mega.index.json, and every referenced shard in the same revision.

1. Convert safetensors to MEGA.
2. Follow the Model Repository layout and release checklist.
3. Load the downloaded artifact from Python.
4. Add related models, datasets, Spaces, or papers to a Collection.

Tip Keep this small repository as a connectivity smoke test. It separates authentication and Hub permissions from conversion, GPU, and large-file failures.

Troubleshooting#