Sine op custom operator

TensorFlow General Discussion
,
1

Hi,

I’m following 自定义算子  |  TensorFlow Lite to create a custom operator (to use it on a micocontroller) which can be supported by Tensorflow Lite. For the sine operator the tutorial uses

converter = tf.lite.TFLiteConverter.from_concrete_functions([sin.get_concrete_function(x)])
converter.allow_custom_ops = True
tflite_model = converter.convert()

Does this snippet of code allow you to convert the entire model which uses sine op, right?

3

Hi @Gianni_Rossi ,
For the function from_concrete_functions, along with sin.get_concrete_function(x) you need to pass defined sine function. Please see the modified code snippet for conversion .

# Convert the model to TensorFlow Lite
converter = tf.lite.TFLiteConverter.from_concrete_functions([my_function.get_concrete_function(tf.constant(x, dtype=tf.float32))], my_function)
converter.allow_custom_ops = True
tflite_model = converter.convert()

Also find the complete code in the gist for custom sine operator conversion