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
Requested fix: when exporting a model through jax2tf, weights must be wrapped in tf.Variable before jax2tf is called.
Relevant jax2tf documentation: https://github.com/google/jax/blob/main/jax/experimental/jax2tf/README.md#saved-model-with-parameters: "Some special care is needed to ensure that the model parameters are not embedded as constants in the graph and are instead saved separately as variables. This is useful for two reasons: the parameters could be very large and exceed the 2GB limits of the GraphDef part of the SavedModel, or you may want to fine-tune the model and change the value of the parameters."
The text was updated successfully, but these errors were encountered:
I don't see any change in the repro Colab. It is still saving all variables as constants in the graph as far as I can tell.
And I did test with keras-nightly. See repro Colab.
Repro colab:
https://colab.research.google.com/drive/1QHg0zpFsJS6qfTDfBwts8KLule7B84RO?usp=sharing
Requested fix: when exporting a model through jax2tf, weights must be wrapped in tf.Variable before jax2tf is called.
Relevant jax2tf documentation:
https://github.com/google/jax/blob/main/jax/experimental/jax2tf/README.md#saved-model-with-parameters:
"Some special care is needed to ensure that the model parameters are not embedded as constants in the graph and are instead saved separately as variables. This is useful for two reasons: the parameters could be very large and exceed the 2GB limits of the GraphDef part of the SavedModel, or you may want to fine-tune the model and change the value of the parameters."
The text was updated successfully, but these errors were encountered: