ValueErrors when calling Model.export() for TF SavedModel format on Keras Models with dict inputs #20835
Assignees
Labels
Comments
mattdangerw
added
the
keras-team-review-pending
|
When this is fixed the follow should work (and would make a good unit test on the tf backend). inputs = {
"foo": keras.Input(shape=()),
"bar": keras.Input(shape=()),
}
outputs = keras.layers.Add()([inputs["foo"], inputs["bar"]])
model = keras.Model(inputs, outputs)
model.export("test/", format="tf_saved_model") |
sachinprasadhs
added
stat:awaiting keras-eng
|
Apologies for the delay. I’ve been occupied with another PR and haven’t been able to make progress on this. Since time is of the essence, it’s now open for other contributors to work on. If no one picks it up, I’ll resume once my current PR is wrapped up. Thanks for your patience! |
hertschuh
added a commit
to hertschuh/keras
that referenced
this issue
Mar 27, 2025
Fixes keras-team#20835 Also changed multi-input tests to exercise `model.export()` and its signature inference logic.
hertschuh
added a commit
to hertschuh/keras
that referenced
this issue
Mar 27, 2025
Fixes keras-team#20835 Also changed multi-input tests to exercise `model.export()` and its signature inference logic.
fchollet
pushed a commit
that referenced
this issue
Mar 27, 2025
Fixes #20835 Also changed multi-input tests to exercise `model.export()` and its signature inference logic.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Problem
The
Model.export()API in Keras 3 supports exporting to a TensorFlow SavedModel artifact for inference. When trying to export Gemma 2 and ShieldGemma to TF SavedModel, I ran into two differentValueErrors:input_signatureis provided, aValueErrorwill be thrown related to a structural mismatch between the expected and actual inputs passed to theGemmaCausalLMclass; andinput_signatureis provided as alist[keras.InputSpec], aValueErrorwill be thrown related the the wrong number of values being passed to a TF function.However, if yo uwrap the
dictfrommodel.inputin alist, asinput_signature=[model.input], the export runs to completion.This is not restricted to Gemma models, as shown in this minimal reproducible example.
Thanks to @mattdangerw for helping to isolate this minimal example.
The text was updated successfully, but these errors were encountered: