diff options
| author | Benjamin Klimczak <benjamin.klimczak@arm.com> | 2022-06-23 10:42:43 +0100 |
|---|---|---|
| committer | Benjamin Klimczak <benjamin.klimczak@arm.com> | 2022-06-23 10:42:43 +0100 |
| commit | 7faf2c4763f299ee53b1ed100025ba50021c8313 (patch) | |
| tree | 42fdcf5e24033f776153f353394308eff99b06e5 | |
| parent | 9097397073133ab4f61da329010839f638978462 (diff) | |
| download | mlia-7faf2c4763f299ee53b1ed100025ba50021c8313.tar.gz | |
MLIA-545 Make quantization non-strict
Use TFLITE_BUILTINS instead of TFLITE_BUILTINS_INT8 to make the
quantization non-strict.
Note: De facto this does not change the behavior of MLIA because the
TFLITE_BUILTINS_INT8 is not correctly applied with the new quantizer.
See: https://github.com/tensorflow/tensorflow/issues/56535
Change-Id: Ia0782ba22c5e9223fa10fec71c16aee60b79bb57
| -rw-r--r-- | src/mlia/nn/tensorflow/utils.py | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/mlia/nn/tensorflow/utils.py b/src/mlia/nn/tensorflow/utils.py index 4abf6cd..b1034d9 100644 --- a/src/mlia/nn/tensorflow/utils.py +++ b/src/mlia/nn/tensorflow/utils.py @@ -71,7 +71,7 @@ def convert_to_tflite(model: tf.keras.Model, quantized: bool = False) -> Interpr if quantized: converter.optimizations = [tf.lite.Optimize.DEFAULT] converter.representative_dataset = representative_dataset(model) - converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8] + converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS] converter.inference_input_type = tf.int8 converter.inference_output_type = tf.int8 @@ -91,7 +91,7 @@ def convert_tf_to_tflite(model: str, quantized: bool = False) -> Interpreter: if quantized: converter.optimizations = [tf.lite.Optimize.DEFAULT] converter.representative_dataset = representative_tf_dataset(model) - converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8] + converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS] converter.inference_input_type = tf.int8 converter.inference_output_type = tf.int8 |