Exploiting unstructured sparsity via tiramisu_pytorch

[Msabih]

Hi,

If I have a model that has a number of weights set to zero via unstructured pruning, its it possible to utilize that from tiramisu_pytorch ?

[rbaghdadi]

Hi @Msabih , at this stage we still do not support sparse DNNs in this integration. If you are interested in helping with this please let us know. @IHIaadj is leading the integration of Tiramisu in Pytorch.

In order to have full support of sparse DNNs in tiramisu_pytorch we need to:

• Include the sparse DNN operators in the tiramisu_pytorch package (we already have the sparse operators but they are not included in the package).
• We need to add support for any data layout transformation that might be necessary between dense and sparse operators (if some operators do not have a sparse equivalent).

Feel to reach out by email, we can setup a zoom call to give you more details.