Option 1 - factor out all tf code currently in TensorBoardCallback to pfl.internal.ops or pfl.internal.bridge, but there is a lot of tf code so Option 2 may be better and simpler.
Option 2 - Make a new callback (named e.g. TensorBoardWithPyTorch) specifically for PyTorch.
We currently have
TensorBoardCallback
for tensorflow https://apple.github.io/pfl-research/reference/callback.html#pfl.callback.TensorBoardCallback. We should also have a callback for TensorBoard to use with pytorch models.Option 1 - factor out all tf code currently in
TensorBoardCallback
topfl.internal.ops
orpfl.internal.bridge
, but there is a lot of tf code so Option 2 may be better and simpler. Option 2 - Make a new callback (named e.g.TensorBoardWithPyTorch
) specifically for PyTorch.(102061658)