TensorFlow 2's default eager execution mode does not have a whole static
graph view that is necessary for the standalone tensorflow-large-model-support's
static graph modification strategy.
A non-graph based implementation of large model support is necessary to
support eager execution.
A new implementation of TensorFlow Large Model Support has been written
which is compiled into TensorFlow. It provides similar tensor swapping
capability and has easier usage than the separate graph-base module.
TensorFlow 2's default eager execution mode does not have a whole static graph view that is necessary for the standalone tensorflow-large-model-support's static graph modification strategy.
A non-graph based implementation of large model support is necessary to support eager execution.
A new implementation of TensorFlow Large Model Support has been written which is compiled into TensorFlow. It provides similar tensor swapping capability and has easier usage than the separate graph-base module.