A class to handle (retrieve and apply) model weights.

Parameters:

name : str

the location of the model weights specified in the form hf-hub:path/architecture_name@revision if available in Hugging Face hub or ns-hub:path/architecture_name if available in the NeuroSpin hub or a path if avaiable in your local machine.

data_dir : pathlib.Path or str

path where data should be downloaded.

filepath : str

the path of the file in the repo.

classmethod hf_download(data_dir: str | Path, hf_id: str, filepath: str, hf_revision: str | None = None, force_download: bool = False) → Path[source]¶

Download a given file if not already present.

Downloads always resume when possible. If you want to force a new download, use force_download=True.

Parameters:

data_dir : pathlib.Path or str

path where data should be downloaded.

hf_id : str

the id of the repository.

filepath : str

the path of the file in the repo.

hf_revision : str, default=None

the revision of the repository (a tag, or a commit hash).

force_download : bool, default=False

whether the file should be downloaded even if it already exists in the local cache.

Returns:

weight_file : Path

local path to the model weights.

classmethod hub_split(hub_name: str) → tuple[str, str | None][source]¶

Interpret the input hub name specified in the form hf-hub:path/architecture_name@revision or ns-hub:path/architecture_name.

Parameters:

hub_name : str

name of the repository.

Returns:

hub_id : str

the id of the repository.

hub_revision : str

the revision of the repository.

load_pretrained(model: Module)[source]¶

Load the model weights.

Parameters:

model : torch.nn.Module

an input model with a load_pretrained method decalred.

classmethod ns_download(data_dir: str | Path, ns_id: str, filepath: str, force_download: bool = False) → Path[source]¶

Download a given file if not already present.

Downloads always resume when possible. If you want to force a new download, use force_download=True.

Parameters:

data_dir : pathlib.Path or str

path where data should be downloaded.

ns_id : str

the id of the repository.

filepath : str

the path of the file in the repo.

force_download : bool, default=False

whether the file should be downloaded even if it already exists in the local cache.

Returns:

weight_file : Path

local path to the model weights.