Is there a way to use f32 format for Math ops in TFLite while keeping other ops in quantized format during Keras model conversion?

Asked May 24 '23 at 13:22

Active May 24 '23 at 13:22

Viewed 27 times

Possible way to skip quantization for certain layers / Ops in tflite converter

Hello Everyone,

Is there a possible way to skip quantization for certain layers / Ops when converting a keras model to tflite model? specifically for Math.ops supported under tflite (which are available in both quantized int8 or f32 formats), so that the model can have f32 for Math.ops, but other ops in quantized format?

Thanks!

asked May 24 '23 at 13:22

user3288779

Please provide enough code so others can better understand or reproduce the problem. – Community May 24 '23 at 14:33

Is there a way to use f32 format for Math ops in TFLite while keeping other ops in quantized format during Keras model conversion?

0 Answers0