Tensorflow 2.17.0 breaks embedding workflow

[acmiyaguchi]

I have a piece of code that extracts embeddings. This works perfectly fine tensorflow==2.16.1, but starts to break on tensorflow==2.17.0. I'm not sure why it starts to throw this error. The tensorflow changelog doesn't show anything egregious: https://github.com/tensorflow/tensorflow/releases/tag/v2.17.0

tests/inference/test_birdnet.py:6: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
inference/birdnet.py:87: in predict
    embeddings.append(self._infer(row))
inference/birdnet.py:70: in _infer
    recording.extract_embeddings()
/path/to/.venv/lib/python3.10/site-packages/birdnetlib/main.py:80: in extract_embeddings
    self.analyzer.extract_embeddings_for_recording(self)
/path/to/.venv/lib/python3.10/site-packages/birdnetlib/analyzer.py:381: in extract_embeddings_for_recording
    e = self._return_embeddings(data)[0].tolist()
/path/to/.venv/lib/python3.10/site-packages/birdnetlib/analyzer.py:454: in _return_embeddings
    features = self.interpreter.get_tensor(output_layer_index)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <tensorflow.lite.python.interpreter.Interpreter object at 0xXXXXXXXXXXXX>, tensor_index = 545, subgraph_index = 0

    def get_tensor(self, tensor_index, subgraph_index=0):
      """Gets the value of the output tensor (get a copy).
    
      If you wish to avoid the copy, use `tensor()`. This function cannot be used
      to read intermediate results.
    
      Args:
        tensor_index: Tensor index of tensor to get. This value can be gotten from
          the 'index' field in get_output_details.
        subgraph_index: Index of the subgraph to fetch the tensor. Default value
          is 0, which means to fetch from the primary subgraph.
    
      Returns:
        a numpy array.
      """
>     return self._interpreter.GetTensor(tensor_index, subgraph_index)
E     ValueError: Tensor data is null. Run allocate_tensors() first

/path/to/.venv/lib/python3.10/site-packages/tensorflow/lite/python/interpreter.py:888: ValueError

The workaround is to pin tensorflow to the last known working version.

[donnubio]

The same problem with 2.18.0.
After that, the error disappeared:

"You should load the tensorflow.lite model using:
interpreter = tf.lite.Interpreter(model_path=tflite_model_path, experimental_preserve_all_tensors=True)
..."