[RNN] TFLite does not appear to be using the UnidirectionalSequenceLSTM

[gaikwadrahul8]

1. System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): macOS
• TensorFlow installation (pip package or built from source): from pip
• TensorFlow library (version, if pip package or github SHA, if built from source): keras-nightly-3.4.1.dev2024080603 tb-nightly-2.18.0a20240806 tf-nightly-2.18.0.dev20240806

2. Code

1) https://colab.research.google.com/gist/eric/f00f071e527f9fa7b2ed39f8d482fbb4/tensorflow-datasets.ipynb 2) https://colab.research.google.com/gist/eric/a292799568831371b7686a2b8cefcd0b/tensorflow-lite-debugger-colab.ipynb

model.zip

3. Failure after conversion

I've found that any Keras LSTM causes this issue.

I expected TFLite to use the UnidirectionalSequenceLSTM op, but instead it seems to be doing something else that then requires the use of flex ops, which I would like to avoid trying to get to work with my tflite deployment situation.

[gaikwadrahul8]

This issue originally reported by @eric has been moved to this dedicated repository for LiteRT to enhance issue tracking and prioritization. To ensure continuity, we have created this new issue on your behalf.

We appreciate your understanding and look forward to your continued involvement.