Installation

TorchJD can be installed directly with pip:

pip install "torchjd[quadprog_projector]"

Note that torchjd requires Python 3.10, 3.11, 3.12, 3.13 or 3.14 and torch>=2.0.

Some aggregators have additional dependencies that are not included by default when installing torchjd. The following table lists the optional dependency groups and the aggregators they enable:

Group	Classes	Dependencies	Install command
quadprog_projector	QuadprogProjector (used in UPGrad and DualProj)	numpy, quadprog, qpsolvers	pip install "torchjd[quadprog_projector]"
cagrad	CAGrad	numpy, cvxpy	pip install "torchjd[cagrad]"
nash_mtl	NashMTL	numpy, cvxpy, ecos	pip install "torchjd[nash_mtl]"

To install torchjd with all of its optional dependencies, you can also use:

pip install "torchjd[full]"