Hugging Face to ModelScope Transfer

简体中文

A simple command-line tool to transfer models and datasets from the Hugging Face Hub to the ModelScope Hub.

Features

• Supports both models and datasets.
• Transfer multiple repositories in a single command.
• Option to specify a custom repository name on ModelScope.
• Automatic repository naming with collision handling (appends a timestamp if the name exists).
• Configure credentials using a .env file or command-line arguments.
• Transfer into public or private repositories.

Installation

pip install hf-ms-transfer

Usage

1. Configuration

Create a .env file in your project directory and add your ModelScope token and username. You can get your token from the ModelScope user center.

# .env
ms_token="your_modelscope_token"
ms_name="your_modelscope_username"

Alternatively, you can provide the token and username directly as command-line arguments.

2. Run the Command

hf-ms-transfer <hf_repo_ids> [options]

Arguments

• <hf_repo_ids>: (Required) One or more Hugging Face repository IDs, separated by commas (e.g., bert-base-uncased,stanfordnlp/imdb).
• --ms-token <token>: (Optional) Your ModelScope access token. Overrides the value in the .env file.
• --ms-name <username>: (Optional) Your ModelScope username. Overrides the value in the .env file.
• --ms-repo-name <name>: (Optional) A custom name for the repository on ModelScope.

  Note: This option cannot be used when transferring multiple repositories.

• --private: (Optional) Create the ModelScope repository as private. Defaults to public.
• --keep-dataset-infos: (Optional) For datasets, upload the dataset_infos.json file. By default, this file is excluded as it can cause issues when using load_datastes function.