Seems like my last comment(question) in this issue was overseen.
As stated I would like to use reverse-mode automatic differentiation in the Haskell Tensorflow bindings. In my tests in TensorFlow 1.14 the GradientTape was usable and did not require eager mode execution.
Maybe you can explain in a little more detail why this is not usable via Haskell?
Can I be of help in implementing support for this feature?
Seems like my last comment(question) in this issue was overseen.
As stated I would like to use reverse-mode automatic differentiation in the Haskell Tensorflow bindings. In my tests in TensorFlow 1.14 the GradientTape was usable and did not require eager mode execution.
Maybe you can explain in a little more detail why this is not usable via Haskell? Can I be of help in implementing support for this feature?