hf-ms-transfer 0.4.0

A tool to transfer models/datasets from Hugging Face to ModelScope

Hugging Face to ModelScope Transfer

简体中文

A tool to transfer models/datasets from the Hugging Face Hub to the ModelScope platform.

By default, when transferring a repository from Hugging Face (e.g., user/package), it will be created on ModelScope as package. If a repository with the same name already exists, a timestamp will be appended to the name (e.g., package_20250708160640). You can also specify a custom name using the --ms-repo-name argument. Note: You cannot provide a custom ModelScope repository name when transferring multiple Hugging Face repositories.

By default, the dataset_infos.json file is not uploaded to ModelScope for datasets, as it can sometimes cause issues on the platform. You can choose to keep this file by using the --keep-dataset-infos flag.

Installation

pip install hf-ms-transfer

Usage

Create a .env file in the directory where you will run the command and add your ModelScope token and username:

ms_token="your_modelscope_token"
ms_name="your_modelscope_username"
# Optional: Specify a Hugging Face Hub endpoint
# HF_ENDPOINT="https://huggingface.co"

Alternatively, you can provide these directly via command-line arguments.

hf-ms-transfer <hf_repo_ids> [--private] [--ms-token <token>] [--ms-name <username>] [--ms-repo-name <name>] [--keep-dataset-infos]

All parameters:

• <hf_repo_ids>: The Hugging Face model or dataset IDs, separated by commas (e.g., 'bert-base-uncased,stanfordnlp/imdb').
• --private: (Optional) Make the ModelScope repository private (default is public).
• --ms-token <token>: (Optional) Your ModelScope access token. Overrides the value in .env. If not provided, a ValueError will be raised.
• --ms-name <username>: (Optional) Your ModelScope username. Overrides the value in .env. If not provided, a ValueError will be raised.
• --ms-repo-name <name>: (Optional) The desired repository name on ModelScope. If not provided, the Hugging Face repository name (without the username) will be used.
• --keep-dataset-infos: (Optional) Keep the dataset_infos.json file for datasets.