Accelerations

Allegro provides several acceleration options to optimize performance for both training and inference on GPUs. Present acceleration options include our CuEquivariance integration or our custom Triton kernels, which can be activated to accelerate the Allegro tensor product.

Feature	CuEquivariance	Triton
Training	✅ Yes	❌ No
Inference	✅ Yes	✅ Yes
Platforms	NVIDIA GPUs	NVIDIA and AMD GPUs