Installation

Prerequisites

For local development, it is generally recommended to install mmlearn in a non-global environment (e.g. venv or conda). This will allow you to use different versions of mmlearn for different projects.

You can create a virtual environment using venv with the following command:

python3 -m venv /path/to/new/virtual/environment
source /path/to/new/virtual/environment/bin/activate

Installing from PyPI

mmlearn is published on the Python Package Index and can be installed using pip.

Run the following command to install the library:

python3 -m pip install mmlearn

Note

mmlearn has several optional dependencies that are used for specific functionality. For example, the peft library for parameter-efficient finetuning. Hence, peft can be installed using:

python3 -m pip install mmlearn[peft]

Specific sets of dependencies are listed below.

Dependency	pip extra	Notes
torchvision, timm, opencv-python	vision	Allows use of computer vision models and image processing functionality
torchaudio	audio	Allows use of audio processing and audio model functionality
peft	peft	Allows use of parameter-efficient fine-tuning methods

Installing from source

You can install mmlearn directly from a clone of the Git repository. This can be done either by cloning the repo and installing from the local clone, or simply installing directly via git.

git clone https://github.com/VectorInstitute/mmlearn.git
cd mmlearn
python3 -m pip install -e .

pip install git+https://github.com/VectorInstitute/mmlearn.git