nmt_with_attention no longer works on Colab

[okyanusoz]

Hello,

I got stuck halfway through in the nmt_with_attention Colab:

---------------------------------------------------------------------------

ValueError                                Traceback (most recent call last)

[<ipython-input-34-f0271e0d820d>](https://localhost:8080/#) in <cell line: 1>()
----> 1 logits = decoder(ex_context, ex_tar_in)
      2 
      3 print(f'encoder output shape: (batch, s, units) {ex_context.shape}')
      4 print(f'input target tokens shape: (batch, t) {ex_tar_in.shape}')
      5 print(f'logits shape shape: (batch, target_vocabulary_size) {logits.shape}')

1 frames

[<ipython-input-30-dcee1b38b45f>](https://localhost:8080/#) in call(self, context, x, state, return_state)
     13 
     14   # 2. Process the target sequence.
---> 15   x, state = self.rnn(x, initial_state=state)
     16   shape_checker(x, 'batch t units')
     17 

ValueError: Exception encountered when calling Decoder.call().

too many values to unpack (expected 2)

Arguments received by Decoder.call():
  • context=tf.Tensor(shape=(64, 18, 256), dtype=float32)
  • x=tf.Tensor(shape=(64, 16), dtype=int64)
  • state=None
  • return_state=False

When I first tried this notebook a long time ago, it worked though.

Regards.

[JuanVargas]

I also experienced same problem on May 1, 2024. I wonder why these tutorials are not verified and if not working removed or updated.

[JuanVargas]

This error was originally reported on April 9. I decided to raise a 'new' issue on May 1, because I got the same error when I tried to run the code under Google Collab or locally under a notebook, running TF 2.16.1. The code that produces the error is posted on the TF Tutorials page. Therefore people trying to learn TF will be mislead and disappointed, as I am. Message to the TF Team: Please fix the code or remove it from the tutorials page. Date: May 18, 2024

[Tomjerry12345]

try change :

!pip install "tensorflow-text>=2.11"

to :

!pip install "tensorflow-text==v2.15.0"

[Mamadou-Aw]

this one from @Tomjerry12345 was very helpfull, my issues by using this version