Get build error for: decoder = keras.Model([decoder_inputs, encoded_seq_inputs], decoder_outputs) (keras Version: 3.10.0) #2175
Description
An issue with English-to-Spanish translation with a sequence-to-sequence Transformer:
https://keras.io/examples/nlp/neural_machine_translation_with_transformer
Section:
Next, we assemble the end-to-end model.
Statement:
decoder = keras.Model([decoder_inputs, encoded_seq_inputs], decoder_outputs)
build error:
ValueError: inputs not connected to outputs
I had built and tested English-to-Spanish translation successfully with no change except for the location of the input files
Then I used the latest version of TF and keras:
TensorFlow Version: 2.19.0
keras Version: 3.10.0
And got the error above
My fix:
I just removed the encoded_seq_inputs arguments from the keras.Model in the above for now!:
Define the decoder model with only the decoder_inputs
decoder = keras.Model(decoder_inputs, decoder_outputs)
Seems to work
Regards
Example of exception when building:
ValueError Traceback (most recent call last)
/tmp/ipython-input-2933432692.py in <cell line: 0>()
16 #AR got error:encoded_seq_inputs was not connected to the outputs
17 #decoder = keras.Model([decoder_inputs, encoded_seq_inputs], decoder_outputs)
---> 18 decoder = keras.Model([decoder_inputs,encoded_seq_inputs], decoder_outputs)
19 print(decoder.summary())
20
2 frames
/usr/local/lib/python3.12/dist-packages/keras/src/ops/function.py in init(self, inputs, outputs, name)
87 and not input._keras_history.operation._outbound_nodes
88 ):
---> 89 raise ValueError("inputs not connected to outputs")
90
91 @Property
ValueError: inputs not connected to outputs