LSTM Model Structure with Low Arena Size for TensorflowLite

Question

I'm working on my project with Time Series Forecasting on ESP32. Seem like my model arena size is too big for ESP32 to predict the value.

I have tried optimize it based on the TensorflowLite document to lower the size. However, I still can't predict with my ESP32. It keeps printing 0.

Here is my colab notebook: https://colab.research.google.com/drive/17Ip0dvavjme9c4oDF7lzte6-uRqRlzq9?usp=sharing

For ESP32. I use EloquentTinyML library to deploy the model

I am really new to this stuff so if you have any solution please let me know. Thank you!