Tayo Oguntebi

icon company @google icon location Google.. Tensorflow.. TPUs.. XLA.. Systems. Performance.

Results 2 comments of Tayo Oguntebi

Investigate allowing async training for DLRM

This was discussed in the Special Topics meeting on January 30. There were no objections to allowing this for DLRM. The proposer needs to make the case for this and...

DLRM - custom embedding kernels should be allowed

From my reading, these kernels are not "copying data to the GPU embedding tables". These seem to be embedding lookup kernels, which is often expressed as a sparse_matrix * dense_vector...