This PR merges the backward into forward-gpu branch.
It makes the branch name meaningless, as now the
forward-gpu contains both the work on the forward
and the work on the backward. However, it the
forward-gpu branch is being used as a staging branch
for all of my (chris') changes.
Almost all of these changes relate to the forward and
backward process on the device.
This PR merges the backward into forward-gpu branch.
It makes the branch name meaningless, as now the forward-gpu contains both the work on the forward and the work on the backward. However, it the forward-gpu branch is being used as a staging branch for all of my (chris') changes.
Almost all of these changes relate to the forward and backward process on the device.