Download models with the modelscope CLI

Install

Install the modelscope package:

pip install modelscope

On macOS, the default download location is ~/.cache/modelscope/hub. To change it, set the MODELSCOPE_CACHE environment variable in ~/.zshrc:

export MODELSCOPE_CACHE=/path/to/your/cache

Notes

• If the path given to local_dir does not exist, the tool creates it automatically.

• local_dir must be an absolute path. Do not use ~ for the home directory: Python's open() does not expand it, and the command fails with No such file or directory.

• The quick start in the official docs uses the form --model="Qwen/Qwen2.5-0.5B-Instruct". Prefer the spaced form --model <model_repo_id> instead.

• <model_repo_id> is the model repository ID from the model page. It is made of the organization name and the model name, for example Qwen/Qwen-Image-2.1.

• Downloading a private model requires you to log in.

Download specific files to a directory

modelscope download --model '<model_repo_id>' '<file_path>' --local_dir '<local_dir>'
 
# Example: download one weight file to a given directory
modelscope download --model 'Qwen/Qwen-Image-2.1-Turbo' 'transformer/diffusion_pytorch_model-00001-of-00002.safetensors' --local_dir '/Users/oceanz/MyFiles/models/Qwen/Qwen-Image-2.1-Turbo'

• <file_path> must keep the file's relative path inside the repository.

• Separate multiple files with spaces.

Download a full repository to the default cache

modelscope download --model '<model_repo_id>'

Download a full repository to a given directory

modelscope download --model '<model_repo_id>' --local_dir '<local_dir>'
 
# Example
modelscope download --model 'Qwen/Qwen-Image-2.1-Turbo' --local_dir '/Users/oceanz/MyFiles/models/Qwen/Qwen-Image-2.1-Turbo'

Parameters