You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Variable class (like tf.Variable) that supports eager mode #170
I've been looking at adding a tf.Variable like class, that would work in eager and graph mode (the variable op doesn't work in eager), and register itself in the execution environment. It's not terribly useful until we get eager gradient support, but it's not dependent on it either so there's no reason to wait.
Eager mode is easy, just have a field. For graph, we can store a Variable op, assign to it, but return/expose the return of the assign (it's the new value). The assigns can take the last assign as a control dep.
I've got a draft here I can PR, but I'm not sure how much the type system refactor would affect it. At a minimum I don't see a reason to merge before #160.
Activity
rnett commentedon Dec 29, 2020
After looking into this, it's a bit more complicated: we will (I assume) want to use the new Resource APIs, not the
variableop (which should probably be deprecated once this is done). See https://github.com/tensorflow/tensorflow/blob/v2.4.0/tensorflow/python/ops/resource_variable_ops.py. It works in eager and graph mode, and isn't that much harder to do.