Export

metatrain.cli.export.export_model(path: Path | str, output: Path | str, path_in_repo: str | None = None, extensions: Path | str = 'extensions/', hf_token: str | None = None, metadata: ModelMetadata | None = None, revision: str | None = None) → None[source]

Export a trained model allowing it to make predictions.

This includes predictions within molecular simulation engines. Exported models will be saved with a .pt file ending. If path does not end with this file extensions .pt will be added and a warning emitted.

The user can specify the model in three ways:

  1. Local File:

    mtt export model.ckpt

  2. Hugging Face Repository (GitHub-style):

    mtt export metatensor/metatrain-test model.ckpt

  3. Any URL:

    mtt export https://huggingface.co/metatensor/metatrain-test/resolve/main/model.ckpt
Parameters:

  • path (Path | str) – path to a model file to be exported, or a Hugging Face repo ID

  • output (Path | str) – path to save the model

  • path_in_repo (str | None) – path to the model file within the Hugging Face repository

  • extensions (Path | str) – path to save the extensions

  • hf_token (str | None) – HuggingFace API token to download (private) models from HuggingFace (optional)

  • metadata (ModelMetadata | None) – metadata to be appended to the model

  • revision (str | None) – Revision (branch, tag, or commit) to download from Hugging Face

Return type:

None