Unfortunately the TFLite framework has some limitations - particularly with the type of quantization modes available. The MCT is more capable, so we would recommend using a framework (e.g. TF or Keras or Pytorch) with floating-point models.
Would you be able to convert your model back to floating point and pass into the MCT perhaps?
Would you be able to convert your model back to floating point and pass into the MCT perhaps?
Statistics: Posted by naushir — Mon Nov 11, 2024 10:33 am