experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix for pylint errors.

Browse Source 
Change-Id: Idd96d7a41fd459c86ab0f6fbb63e5d543509145d
This commit is contained in:
Elena Zhelezina 2020-06-09 17:47:16 +01:00
parent dbc7faeecd
commit 507c754931
5 changed files with 28 additions and 18 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
tensorflow/lite/python/convert.py
View File

@ -103,7 +103,8 @@ class OpsSet(enum.Enum):
# significantly, but only slightly increases the model size. # significantly, but only slightly increases the model size.
# WARNING: These ops are currently experimental and have not yet been finalized. # WARNING: These ops are currently experimental and have not yet been finalized.
# They are only compatible with CPU execution, and have not been optimized for production. # They are only compatible with CPU execution, and have not been optimized for production.
EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8 = "EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8" EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8 = \
"EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8"
def __str__(self): def __str__(self):
return self.value return self.value

27
tensorflow/lite/python/lite.py
View File

@ -251,7 +251,8 @@ class QuantizationMode(object):
self.post_training_fp16()) self.post_training_fp16())
def activations_type(self): def activations_type(self):
return constants.INT16 if self._is_int16x8_target_required() else constants.INT8 return constants.INT16 if self._is_int16x8_target_required() \
else constants.INT8
def converter_flags(self, inference_ty=None, inference_input_ty=None): def converter_flags(self, inference_ty=None, inference_input_ty=None):
"""Flags to the converter.""" """Flags to the converter."""
@ -262,7 +263,8 @@ class QuantizationMode(object):
if self.training_time_int8_allow_float(): if self.training_time_int8_allow_float():
return { return {
"inference_type": inference_ty if inference_ty else self.activations_type(), "inference_type": inference_ty if inference_ty else \
self.activations_type(),
"inference_input_type": "inference_input_type":
inference_input_ty if inference_input_ty else constants.FLOAT, inference_input_ty if inference_input_ty else constants.FLOAT,
"post_training_quantize": False, # disable dynamic range quantization "post_training_quantize": False, # disable dynamic range quantization
@ -359,15 +361,15 @@ class QuantizationMode(object):
def _is_int16x8_target_required(self): def _is_int16x8_target_required(self):
return bool( return bool(
set(self._target_spec.supported_ops).intersection([ set(self._target_spec.supported_ops).intersection([
OpsSet.EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8 OpsSet.EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8
])) ]))
def _is_allow_float(self): def _is_allow_float(self):
return bool( return bool(
set(self._target_spec.supported_ops).intersection([ set(self._target_spec.supported_ops).intersection([
OpsSet.TFLITE_BUILTINS OpsSet.TFLITE_BUILTINS
])) ]))
def _any_optimization_enabled(self): def _any_optimization_enabled(self):
return bool( return bool(
@ -441,7 +443,8 @@ class TFLiteConverterBase(object):
return _get_grappler_config(optimizers) return _get_grappler_config(optimizers)
def _calibrate_quantize_model(self, result, inference_input_type, def _calibrate_quantize_model(self, result, inference_input_type,
inference_output_type, activations_type, allow_float): inference_output_type, activations_type,
allow_float):
"""Calibrate and quantize the model.""" """Calibrate and quantize the model."""
if not isinstance(self.representative_dataset, RepresentativeDataset): if not isinstance(self.representative_dataset, RepresentativeDataset):
self.representative_dataset = RepresentativeDataset( self.representative_dataset = RepresentativeDataset(
@ -458,8 +461,8 @@ class TFLiteConverterBase(object):
return _mlir_quantize(calibrated) return _mlir_quantize(calibrated)
else: else:
return calibrate_quantize.calibrate_and_quantize( return calibrate_quantize.calibrate_and_quantize(
self.representative_dataset.input_gen, inference_input_type, self.representative_dataset.input_gen, inference_input_type,
inference_output_type, allow_float, activations_type) inference_output_type, allow_float, activations_type)
def _is_unknown_shapes_allowed(self): def _is_unknown_shapes_allowed(self):
# Unknown dimensions are only allowed with the new converter. # Unknown dimensions are only allowed with the new converter.
@ -1992,7 +1995,7 @@ class TocoConverter(object):
@classmethod @classmethod
@_deprecation.deprecated( @_deprecation.deprecated(
None, "Use `lite.TFLiteConverter.from_keras_model_file` instead.") None, "Use `lite.TFLiteConverter.from_keras_model_file` instead.")
def from_keras_model_file(cls, def from_keras_model_file(cls,
model_file, model_file,
input_arrays=None, input_arrays=None,

10
tensorflow/lite/python/lite_test.py
View File

@ -882,11 +882,15 @@ class FromSessionTest(TestModels, parameterized.TestCase):
@parameterized.named_parameters( @parameterized.named_parameters(
# Quantize model to Int8: with enable mlir # Quantize model to Int8: with enable mlir
('UseTfliteBuiltinsIntEnableMLIR', [lite.OpsSet.TFLITE_BUILTINS_INT8], True), ('UseTfliteBuiltinsIntEnableMLIR',
[lite.OpsSet.TFLITE_BUILTINS_INT8], True),
# Quantize model to Int8: with disable mlir # Quantize model to Int8: with disable mlir
('UseTfliteBuiltinsIntDisableMLIR', [lite.OpsSet.TFLITE_BUILTINS_INT8], False), ('UseTfliteBuiltinsIntDisableMLIR',
[lite.OpsSet.TFLITE_BUILTINS_INT8], False),
# Quantize model to Int16: with disable mlir # Quantize model to Int16: with disable mlir
('UseTfliteBuiltinsInt16DisableMLIR', [lite.OpsSet.EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8], False)) ('UseTfliteBuiltinsInt16DisableMLIR',
[lite.OpsSet.EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8],
False))
def testCalibrateAndQuantizeBuiltinInt(self, supported_ops, enable_mlir): def testCalibrateAndQuantizeBuiltinInt(self, supported_ops, enable_mlir):
with ops.Graph().as_default(): with ops.Graph().as_default():
inp, output, calibration_gen = self._getCalibrationQuantizeModel() inp, output, calibration_gen = self._getCalibrationQuantizeModel()

3
tensorflow/lite/python/optimize/calibrator.py
View File

@ -78,7 +78,8 @@ class Calibrator(object):
computation, useful when targeting an integer-only backend. computation, useful when targeting an integer-only backend.
If False, an error will be thrown if an operation cannot be If False, an error will be thrown if an operation cannot be
quantized, otherwise the model will fallback to float ops. quantized, otherwise the model will fallback to float ops.
activations_type: A tf.dtype representing the desired type for activations. activations_type: A tf.dtype representing the desired type for
activations.
resize_input: A boolean. True if the shape of the sample data is different resize_input: A boolean. True if the shape of the sample data is different
from the input. from the input.
""" """

3
tensorflow/lite/python/optimize/calibrator_test.py
View File

@ -96,7 +96,8 @@ class CalibratorTest(test_util.TensorFlowTestCase, parameterized.TestCase):
('UseActivationTypeInt8 - EnableMlirQuantizer', constants.INT8), ('UseActivationTypeInt8 - EnableMlirQuantizer', constants.INT8),
# Activation type Int16 # Activation type Int16
('UseActivationTypeInt16 - DisableEnableMlirQuantizer', constants.INT16)) ('UseActivationTypeInt16 - DisableEnableMlirQuantizer', constants.INT16))
def test_calibration_with_quantization_multiple_inputs(self, activations_type): def test_calibration_with_quantization_multiple_inputs(self,
activations_type):
# Load multi add model from test data. # Load multi add model from test data.
# This model has 4 inputs of size (1, 8, 8, 3). # This model has 4 inputs of size (1, 8, 8, 3).
model_path = resource_loader.get_path_to_datafile( model_path = resource_loader.get_path_to_datafile(