Tensorflow - Converting a pruned model to a lower quantization with tflite

Question

I'm trying to quantize a pruned tensorflow model using TFLite. Both the pruning and quantization work independently, but when I try to prune and then quant I get this error: ValueError: Input 0 of node sequential/prune_low_magnitude_dense/AssignVariableOp was passed float from sequential/prune_low_magnitude_dense/Mul/ReadVariableOp/resource:0 incompatible with expected resource. and when I try to quant and then prune I get this error: ValueError: 'prune_low_magnitude' can only prune an object of the following types: tf.keras.models.Sequential, tf.keras functional model, tf.keras.layers.Layer, list of tf.keras.layers.Layer. You passed an object of type: bytes.

What should I do to make these operations compatible? The model is of type tf.keras.models.Sequential, and for testing purposes is only one layer deep.

Python 3.6, Tensorflow 2.1, sparsity function comes from tensorflow_model_optimization which is at version 0.2.1. Windows 10 if that matters. — Sig, Mar 08 '20 at 02:22
I am also encountering this issue. Looking for an answer now. — Christian Westbrook, Oct 08 '20 at 21:29

score 0 · Answer 1 · answered May 17 '21 at 01:40

0

As shown in this example:

https://github.com/tensorflow/model-optimization/blob/master/tensorflow_model_optimization/python/examples/quantization_with_sparsity/keras/mnist_cnn.py

Seems like you should strip_pruning.

answered May 17 '21 at 01:40

daverim

101
3

Tensorflow - Converting a pruned model to a lower quantization with tflite

1 Answers1