| /external/tensorflow/tensorflow/lite/tools/optimize/ |
| D | modify_model_interface_test.cc | 93 tensor_1->quantization = absl::make_unique<QuantizationParametersT>(); in CreateQuantizedModelSingleInputOutput()
94 tensor_1->quantization->scale.push_back(0.35); in CreateQuantizedModelSingleInputOutput()
95 tensor_1->quantization->zero_point.push_back(28); in CreateQuantizedModelSingleInputOutput()
101 tensor_2->quantization = absl::make_unique<QuantizationParametersT>(); in CreateQuantizedModelSingleInputOutput()
102 tensor_2->quantization->scale.push_back(0.12); in CreateQuantizedModelSingleInputOutput()
103 tensor_2->quantization->zero_point.push_back(50); in CreateQuantizedModelSingleInputOutput()
203 tensor_2->quantization = absl::make_unique<QuantizationParametersT>(); in CreateQuantizedModelMultipleInputOutput()
204 tensor_2->quantization->scale.push_back(0.35); in CreateQuantizedModelMultipleInputOutput()
205 tensor_2->quantization->zero_point.push_back(28); in CreateQuantizedModelMultipleInputOutput()
211 tensor_3->quantization = absl::make_unique<QuantizationParametersT>(); in CreateQuantizedModelMultipleInputOutput()
[all …]
|
| D | quantize_model_test.cc | 99 const auto quantization_params = tensor->quantization.get(); in ExpectSameModels()
101 expected_tensor->quantization.get(); in ExpectSameModels()
363 EXPECT_EQ(subgraph->tensors[input_idx]->quantization->scale.size(), 1); in TEST_P()
364 EXPECT_FLOAT_EQ(subgraph->tensors[input_idx]->quantization->scale[0], in TEST_P()
366 EXPECT_EQ(subgraph->tensors[input_idx]->quantization->zero_point.size(), 1); in TEST_P()
367 EXPECT_EQ(subgraph->tensors[input_idx]->quantization->zero_point[0], 0); in TEST_P()
370 EXPECT_EQ(subgraph->tensors[output_idx]->quantization->scale.size(), 1); in TEST_P()
371 EXPECT_FLOAT_EQ(subgraph->tensors[output_idx]->quantization->scale[0], in TEST_P()
373 EXPECT_EQ(subgraph->tensors[output_idx]->quantization->zero_point.size(), in TEST_P()
375 EXPECT_EQ(subgraph->tensors[output_idx]->quantization->zero_point[0], 0); in TEST_P()
[all …]
|
| D | model_utils_test.cc | 35 tensor.quantization = absl::make_unique<QuantizationParametersT>(); in TEST()
36 tensor.quantization->scale.push_back(0.5); in TEST()
37 tensor.quantization->scale.push_back(1.5); in TEST()
39 tensor.quantization->zero_point.push_back(1); in TEST()
40 tensor.quantization->zero_point.push_back(-1); in TEST()
60 tensor.quantization = absl::make_unique<QuantizationParametersT>(); in TEST()
61 tensor.quantization->min.push_back(0.5); in TEST()
63 tensor.quantization->max.push_back(1.5); in TEST()
|
| D | quantize_model.cc | 152 TF_LITE_ENSURE(error_reporter, weight_tensor->quantization); in QuantizeBias()
153 std::vector<float> weight_scales = weight_tensor->quantization->scale; in QuantizeBias()
163 if (!input_tensor->quantization || in QuantizeBias()
164 input_tensor->quantization->scale.size() != 1) { in QuantizeBias()
178 model, bias_tensor, input_tensor->quantization->scale[0], in QuantizeBias()
182 model, bias_tensor, input_tensor->quantization->scale[0], in QuantizeBias()
196 input_tensor->quantization->scale[0] * weight_scales[0], in QuantizeBias()
201 input_tensor->quantization->scale[0] * weight_scales[0], in QuantizeBias()
214 !tensor->quantization->scale.empty(); in TensorTypeChangeRequired()
217 !tensor->quantization->scale.empty(); in TensorTypeChangeRequired()
[all …]
|
| D | quantization_utils.cc | 107 tensor->quantization->min[0], tensor->quantization->max[0], in GetQuantizationParams()
112 GetSymmetricQuantizationParams(tensor->quantization->min[0], in GetQuantizationParams()
113 tensor->quantization->max[0], in GetQuantizationParams()
303 if (tensor->quantization == nullptr) { in SymmetricPerChannelQuantization()
304 tensor->quantization = absl::make_unique<QuantizationParametersT>(); in SymmetricPerChannelQuantization()
309 tensor->quantization.get(), error_reporter)); in SymmetricPerChannelQuantization()
317 std::max(std::abs(tensor->quantization->min[channel_idx]), in SymmetricPerChannelQuantization()
318 std::abs(tensor->quantization->max[channel_idx])); in SymmetricPerChannelQuantization()
417 if (tensor->quantization->min.size() != 1 || in SymmetricQuantizeTensorFromMinMax()
418 tensor->quantization->max.size() != 1) { in SymmetricQuantizeTensorFromMinMax()
[all …]
|
| D | model_utils.cc | 106 (*tensor)->quantization = absl::make_unique<QuantizationParametersT>(); in MakeTensorWithQuantParam()
107 (*tensor)->quantization->scale.push_back(scale); in MakeTensorWithQuantParam()
108 (*tensor)->quantization->zero_point.push_back(zero_point); in MakeTensorWithQuantParam()
112 return tensor->quantization != nullptr && in QuantizationParametersExist()
113 !tensor->quantization->scale.empty() && in QuantizationParametersExist()
114 !tensor->quantization->zero_point.empty(); in QuantizationParametersExist()
128 return tensor->quantization && !tensor->quantization->min.empty() && in HasMinMax()
129 !tensor->quantization->max.empty(); in HasMinMax()
|
| D | quantization_utils_test.cc | 204 tensor.quantization = nullptr; in TEST_F()
235 tensor.quantization = absl::make_unique<QuantizationParametersT>(); in TEST_F()
238 tensor.quantization.get(), &error_reporter_); in TEST_F()
243 EXPECT_THAT(tensor.quantization->min, ElementsAreArray(expected_mins)); in TEST_F()
244 EXPECT_THAT(tensor.quantization->max, ElementsAreArray(expected_maxs)); in TEST_F()
359 auto quantization = absl::make_unique<QuantizationParametersT>(); in TEST_F() local
360 quantization->min = {-0.00001, -7.0, -2.0}; in TEST_F()
361 quantization->max = {0.00001, 1.0, -1.0}; in TEST_F()
362 std::vector<float> scales = std::vector<float>(quantization->min.size()); in TEST_F()
364 GetSymmetricScalesFromMaxMin(quantization.get(), &scales, &error_reporter_); in TEST_F()
[all …]
|
| /external/tensorflow/tensorflow/compiler/mlir/lite/tests/mlir2flatbuffer/ |
| D | lstm.mlir | 16 // CHECK-NEXT: quantization: {
23 // CHECK-NEXT: quantization: {
30 // CHECK-NEXT: quantization: {
37 // CHECK-NEXT: quantization: {
44 // CHECK-NEXT: quantization: {
51 // CHECK-NEXT: quantization: {
58 // CHECK-NEXT: quantization: {
65 // CHECK-NEXT: quantization: {
72 // CHECK-NEXT: quantization: {
79 // CHECK-NEXT: quantization: {
[all …]
|
| D | unidirectional_sequence_lstm.mlir | 16 // CHECK-NEXT: quantization: {
23 // CHECK-NEXT: quantization: {
30 // CHECK-NEXT: quantization: {
37 // CHECK-NEXT: quantization: {
44 // CHECK-NEXT: quantization: {
51 // CHECK-NEXT: quantization: {
58 // CHECK-NEXT: quantization: {
65 // CHECK-NEXT: quantization: {
72 // CHECK-NEXT: quantization: {
79 // CHECK-NEXT: quantization: {
[all …]
|
| D | lstm_quantized.mlir | 20 // CHECK-NEXT: quantization: {
29 // CHECK-NEXT: quantization: {
38 // CHECK-NEXT: quantization: {
47 // CHECK-NEXT: quantization: {
56 // CHECK-NEXT: quantization: {
65 // CHECK-NEXT: quantization: {
74 // CHECK-NEXT: quantization: {
83 // CHECK-NEXT: quantization: {
92 // CHECK-NEXT: quantization: {
101 // CHECK-NEXT: quantization: {
[all …]
|
| D | basic_lstm.mlir | 16 // CHECK-NEXT: quantization: {
23 // CHECK-NEXT: quantization: {
30 // CHECK-NEXT: quantization: {
37 // CHECK-NEXT: quantization: {
44 // CHECK-NEXT: quantization: {
51 // CHECK-NEXT: quantization: {
58 // CHECK-NEXT: quantization: {
65 // CHECK-NEXT: quantization: {
72 // CHECK-NEXT: quantization: {
|
| D | while_op.mlir | 26 // CHECK-NEXT: quantization: {
33 // CHECK-NEXT: quantization: {
41 // CHECK-NEXT: quantization: {
48 // CHECK-NEXT: quantization: {
70 // CHECK-NEXT: quantization: {
77 // CHECK-NEXT: quantization: {
85 // CHECK-NEXT: quantization: {
93 // CHECK-NEXT: quantization: {
111 // CHECK-NEXT: quantization: {
118 // CHECK-NEXT: quantization: {
[all …]
|
| D | tfl_while_op.mlir | 26 // CHECK-NEXT: quantization: {
33 // CHECK-NEXT: quantization: {
41 // CHECK-NEXT: quantization: {
48 // CHECK-NEXT: quantization: {
70 // CHECK-NEXT: quantization: {
77 // CHECK-NEXT: quantization: {
85 // CHECK-NEXT: quantization: {
93 // CHECK-NEXT: quantization: {
111 // CHECK-NEXT: quantization: {
118 // CHECK-NEXT: quantization: {
[all …]
|
| D | if_op.mlir | 26 // CHECK-NEXT: quantization: {
33 // CHECK-NEXT: quantization: {
41 // CHECK-NEXT: quantization: {
48 // CHECK-NEXT: quantization: {
73 // CHECK-NEXT: quantization: {
80 // CHECK-NEXT: quantization: {
87 // CHECK-NEXT: quantization: {
108 // CHECK-NEXT: quantization: {
115 // CHECK-NEXT: quantization: {
122 // CHECK-NEXT: quantization: {
|
| /external/tensorflow/tensorflow/compiler/mlir/lite/tests/flatbuffer2mlir/ |
| D | importer_test_min_max.cc | 123 input_tensor->quantization->scale.clear(); in InjectStatsToFullyConnected()
124 input_tensor->quantization->zero_point.clear(); in InjectStatsToFullyConnected()
125 input_tensor->quantization->min.push_back(-1.0); in InjectStatsToFullyConnected()
126 input_tensor->quantization->max.push_back(1.0); in InjectStatsToFullyConnected()
130 output_tensor->quantization->scale.clear(); in InjectStatsToFullyConnected()
131 output_tensor->quantization->zero_point.clear(); in InjectStatsToFullyConnected()
133 output_tensor->quantization->min.push_back(-1.0 * i); in InjectStatsToFullyConnected()
134 output_tensor->quantization->max.push_back(1.0 * i); in InjectStatsToFullyConnected()
136 output_tensor->quantization->quantized_dimension = shape.size() - 1; in InjectStatsToFullyConnected()
|
| /external/tensorflow/tensorflow/lite/toco/tflite/ |
| D | import.cc | 79 auto quantization = input_tensor->quantization(); in ImportTensors() local
80 if (quantization) { in ImportTensors()
83 if (quantization->min() && quantization->max()) { in ImportTensors()
84 CHECK_EQ(1, quantization->min()->Length()); in ImportTensors()
85 CHECK_EQ(1, quantization->max()->Length()); in ImportTensors()
87 minmax.min = quantization->min()->Get(0); in ImportTensors()
88 minmax.max = quantization->max()->Get(0); in ImportTensors()
90 if (quantization->scale() && quantization->zero_point()) { in ImportTensors()
91 CHECK_EQ(1, quantization->scale()->Length()); in ImportTensors()
92 CHECK_EQ(1, quantization->zero_point()->Length()); in ImportTensors()
[all …]
|
| /external/tensorflow/tensorflow/lite/g3doc/performance/ |
| D | post_training_quantization.md | 1 # Post-training quantization
3 Post-training quantization is a conversion technique that can reduce model size
13 There are several post-training quantization options to choose from. Here is a
19 : quantization : : :
21 : quantization : : Microcontrollers :
22 | Float16 quantization | 2x smaller, GPU | CPU, GPU |
25 The following decision tree can help determine which post-training quantization
30 ### Dynamic range quantization
32 The simplest form of post-training quantization statically quantizes only the
53 ### Full integer quantization
[all …]
|
| D | model_optimization.md | 46 Currently, quantization can be used to reduce latency by simplifying the
73 TensorFlow Lite currently supports optimization via quantization, pruning and
83 [Quantization](https://www.tensorflow.org/model_optimization/guide/quantization/post_training)
88 The following types of quantization are available in TensorFlow Lite:
92 [Post-training float16 quantization](post_training_float16_quant.ipynb) …
93 [Post-training dynamic range quantization](post_training_quant.ipynb) …
94 [Post-training integer quantization](post_training_integer_quant.ipynb) …
95 [Quantization-aware training](http://www.tensorflow.org/model_optimization/guide/quantization/train…
97 Below are the latency and accuracy results for post-training quantization and
98 quantization-aware training on a few models. All latency numbers are measured on
[all …]
|
| /external/tensorflow/tensorflow/lite/kernels/ |
| D | kernel_util_test.cc | 279 input.quantization.type = kTfLiteAffineQuantization; in TEST_F()
286 input.quantization.params = reinterpret_cast<void*>(input_params); in TEST_F()
299 filter.quantization.type = kTfLiteAffineQuantization; in TEST_F()
311 filter.quantization.params = reinterpret_cast<void*>(filter_params); in TEST_F()
320 bias.quantization.type = kTfLiteAffineQuantization; in TEST_F()
331 bias.quantization.params = reinterpret_cast<void*>(bias_params); in TEST_F()
340 output.quantization.type = kTfLiteAffineQuantization; in TEST_F()
347 output.quantization.params = reinterpret_cast<void*>(output_params); in TEST_F()
385 input.quantization.type = kTfLiteAffineQuantization; in TEST_F()
392 input.quantization.params = reinterpret_cast<void*>(input_params); in TEST_F()
[all …]
|
| /external/tensorflow/tensorflow/lite/c/ |
| D | common.c | 91 void TfLiteQuantizationFree(TfLiteQuantization* quantization) { in TfLiteQuantizationFree() argument
92 if (quantization->type == kTfLiteAffineQuantization) { in TfLiteQuantizationFree()
94 (TfLiteAffineQuantization*)(quantization->params); in TfLiteQuantizationFree()
105 quantization->params = NULL; in TfLiteQuantizationFree()
106 quantization->type = kTfLiteNoQuantization; in TfLiteQuantizationFree()
152 TfLiteQuantizationFree(&t->quantization); in TfLiteTensorFree()
158 TfLiteQuantizationParams quantization, char* buffer, in TfLiteTensorReset() argument
166 tensor->params = quantization; in TfLiteTensorReset()
173 tensor->quantization.type = kTfLiteNoQuantization; in TfLiteTensorReset()
174 tensor->quantization.params = NULL; in TfLiteTensorReset()
|
| /external/tensorflow/tensorflow/lite/delegates/gpu/common/ |
| D | model_builder_test.cc | 238 TfLiteQuantization quantization; in InterpreterFp16() local
239 quantization.type = kTfLiteNoQuantization; in InterpreterFp16()
242 0, TfLiteType::kTfLiteFloat16, "t0", dims, quantization, false), in InterpreterFp16()
246 2, TfLiteType::kTfLiteFloat16, "t2", dims, quantization, false), in InterpreterFp16()
259 1, TfLiteType::kTfLiteFloat32, "t1", dims, quantization, false), in InterpreterFp16()
263 3, TfLiteType::kTfLiteFloat32, "t3", dims, quantization, false), in InterpreterFp16()
504 TfLiteQuantization quantization; in InterpreterFp32() local
505 quantization.type = kTfLiteNoQuantization; in InterpreterFp32()
507 0, TfLiteType::kTfLiteUInt8, "t0", dims, quantization, false), in InterpreterFp32()
511 1, TfLiteType::kTfLiteFloat32, "t1", dims, quantization, false), in InterpreterFp32()
[all …]
|
| /external/tensorflow/tensorflow/compiler/mlir/lite/quantization/ |
| D | quantization_info.proto | 7 // Represents the quantization parameters for a list of named tensors.
11 // quantization specification.
39 // The quantized axis index if it is per-axis quantization.
48 // The quantization parameters for a named tensor.
62 // The quantization parameters for the tensor. If it is for per-axis, the
67 // Metadata about the quantization parameters.
71 // List of quantization parameters for tensors.
|
| D | quantization.td | 16 // This is the quantization definition file for TensorFlow.
60 // TFL native op traits (for quantization).
84 // are used to generate the op quantization specs.
101 [{Returns quantization dim for the affine operand.}],
141 // and also the quantization dimension if per-axis quantization is support.
142 // If the quantization dimension is -1, per-axis quantization isn't supported.
149 // apply quantization on this op.
|
| /external/tensorflow/tensorflow/lite/ |
| D | interpreter.cc | 62 TfLiteQuantization quantization; in GetQuantizationFromLegacy() local
63 quantization.type = kTfLiteAffineQuantization; in GetQuantizationFromLegacy()
70 quantization.params = affine_quantization; in GetQuantizationFromLegacy()
72 return quantization; in GetQuantizationFromLegacy()
309 const std::vector<int>& dims, TfLiteQuantization quantization, in SetTensorParametersReadOnly() argument
312 tensor_index, type, name, dims.size(), dims.data(), quantization, buffer, in SetTensorParametersReadOnly()
318 const std::vector<int>& dims, TfLiteQuantization quantization, in SetTensorParametersReadWrite() argument
321 tensor_index, type, name, dims.size(), dims.data(), quantization, in SetTensorParametersReadWrite()
327 const int* dims, TfLiteQuantizationParams quantization, const char* buffer, in SetTensorParametersReadOnly() argument
329 TfLiteQuantization new_quantization = GetQuantizationFromLegacy(quantization); in SetTensorParametersReadOnly()
[all …]
|
| /external/tensorflow/tensorflow/lite/tools/optimize/calibration/ |
| D | calibration_reader.cc | 48 if (tensor->quantization) { in AddCalibrationToModel()
49 const float existing_min = tensor->quantization->min[0]; in AddCalibrationToModel()
50 const float existing_max = tensor->quantization->max[0]; in AddCalibrationToModel()
58 subgraph->tensors[tensorid_stat.first]->quantization = in AddCalibrationToModel()
|