Creating integer only models

Nils is it possible to create an integer only models so this could run on accelerators or frameworks such as ArmNN? https://www.tensorflow.org/lite/performance/post_training_quantization#full_integer_quantization

I always get confused at how to implement the representative_dataset()?

import tensorflow as tf
converter = tf.lite.TFLiteConverter.from_saved_model(saved_model_dir)
converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.representative_dataset = representative_dataset
converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
converter.inference_input_type = tf.int8  # or tf.uint8
converter.inference_output_type = tf.int8  # or tf.uint8
tflite_quant_model = converter.convert()

Has anyone done this and got an example or even better the tflite models?

[JorgeRuizDev]

A few months ago I managed to quantize this LSTM model and run it on a Coral Edge TPU https://colab.research.google.com/github/google-coral/tutorials/blob/master/train_lstm_timeseries_ptq_tf2.ipynb

The example has been broken since TF 2.7…

[StuartIanNaylor]

A few months ago I managed to quantize this LSTM model and run it on a Coral Edge TPU https://colab.research.google.com/github/google-coral/tutorials/blob/master/train_lstm_timeseries_ptq_tf2.ipynb

The example has been broken since TF 2.7…

Yeah its confusing as post-training quantization of recurrent layers does seem to be broken, dunno.