We are excited to announce the release of LightOn ML v1.4, now available on PyPi and on LightOn servers on LightOn Cloud!
With this release, you now have at your disposal a lightning-fast scalable linear random projection on any Aurora Nitro OPU. A linear transform opens the path to many interesting applications, for example in randomized linear algebra. You can find a selection of these in our preprint Photonic co-processors in HPC: using LightOnOPUs for Randomized Numerical Linear Algebra. To use the linear transform, you just need to call
opu.linear_transform in place of
opu.transform. If you use one of the
OPUMap classes, the linear transform is activated with the
linear boolean attribute.
The API doesn’t change: you can still use numpy arrays, Pytorch tensors, and everything you have ever used before!