telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1 | // |
| 2 | // Copyright © 2017 Arm Ltd. All rights reserved. |
David Beck | 93e4898 | 2018-09-05 13:05:09 +0100 | [diff] [blame] | 3 | // SPDX-License-Identifier: MIT |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 4 | // |
| 5 | #include "DriverTestHelpers.hpp" |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 6 | #include "OperationsUtils.h" |
| 7 | |
| 8 | #include <boost/array.hpp> |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 9 | #include <boost/test/unit_test.hpp> |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 10 | #include <boost/test/data/test_case.hpp> |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 11 | #include <boost/math/special_functions/relative_difference.hpp> |
| 12 | #include <log/log.h> |
| 13 | |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 14 | #include <cmath> |
| 15 | |
| 16 | BOOST_AUTO_TEST_SUITE(LstmTests) |
| 17 | |
| 18 | using ArmnnDriver = armnn_driver::ArmnnDriver; |
| 19 | using DriverOptions = armnn_driver::DriverOptions; |
| 20 | using namespace driverTestHelpers; |
| 21 | using namespace android::hardware; |
| 22 | |
| 23 | namespace |
| 24 | { |
| 25 | |
| 26 | template<typename T> |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 27 | RequestArgument CreateRequestArgument(const std::vector<T>& value, unsigned int poolIndex) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 28 | { |
| 29 | DataLocation inputInloc = {}; |
| 30 | inputInloc.poolIndex = poolIndex; |
| 31 | inputInloc.offset = 0; |
| 32 | inputInloc.length = value.size() * sizeof(T); |
| 33 | RequestArgument inputRequestArgument = {}; |
| 34 | inputRequestArgument.location = inputInloc; |
| 35 | inputRequestArgument.dimensions = hidl_vec<uint32_t>{}; |
| 36 | return inputRequestArgument; |
| 37 | } |
| 38 | |
| 39 | // Returns true if the relative difference between two float values is less than the tolerance value given. |
| 40 | // This is used because the floating point comparison tolerance (set on each BOOST_AUTO_TEST_CASE) does not work! |
| 41 | bool TolerantCompareEqual(float a, float b, float tolerance = 0.00001f) |
| 42 | { |
| 43 | float rd; |
| 44 | if (a == 0.0f) |
| 45 | { |
| 46 | rd = fabs(b); |
| 47 | } |
| 48 | else if (b == 0.0f) |
| 49 | { |
| 50 | rd = fabs(a); |
| 51 | } |
| 52 | else |
| 53 | { |
| 54 | rd = boost::math::relative_difference(a, b); |
| 55 | } |
| 56 | return rd < tolerance; |
| 57 | } |
| 58 | |
Kevin May | f29a2c5 | 2019-03-14 11:56:32 +0000 | [diff] [blame] | 59 | // Helper function to create an OperandLifeTime::NO_VALUE for testing. |
| 60 | // To be used on optional input operands that have no values - these are valid and should be tested. |
| 61 | OperandLifeTime CreateNoValueLifeTime(const hidl_vec<uint32_t>& dimensions) |
| 62 | { |
| 63 | // Only create a NO_VALUE for optional operands that have no elements |
| 64 | if (dimensions.size() == 0 || dimensions[0] == 0) |
| 65 | { |
| 66 | return OperandLifeTime::NO_VALUE; |
| 67 | } |
| 68 | return OperandLifeTime::CONSTANT_COPY; |
| 69 | } |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 70 | } // anonymous namespace |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 71 | |
| 72 | // Add our own tests here since we fail the lstm tests which Google supplies (because of non-const weights) |
| 73 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 74 | void LstmTestImpl(const hidl_vec<uint32_t>& inputDimensions, |
| 75 | const std::vector<float>& inputValue, |
| 76 | const hidl_vec<uint32_t>& inputToInputWeightsDimensions, |
| 77 | const std::vector<float>& inputToInputWeightsValue, |
| 78 | const hidl_vec<uint32_t>& inputToForgetWeightsDimensions, |
| 79 | const std::vector<float>& inputToForgetWeightsValue, |
| 80 | const hidl_vec<uint32_t>& inputToCellWeightsDimensions, |
| 81 | const std::vector<float>& inputToCellWeightsValue, |
| 82 | const hidl_vec<uint32_t>& inputToOutputWeightsDimensions, |
| 83 | const std::vector<float>& inputToOutputWeightsValue, |
| 84 | const hidl_vec<uint32_t>& recurrentToInputWeightsDimensions, |
| 85 | const std::vector<float>& recurrentToInputWeightsValue, |
| 86 | const hidl_vec<uint32_t>& recurrentToForgetWeightsDimensions, |
| 87 | const std::vector<float>& recurrentToForgetWeightsValue, |
| 88 | const hidl_vec<uint32_t>& recurrentToCellWeightsDimensions, |
| 89 | const std::vector<float>& recurrentToCellWeightsValue, |
| 90 | const hidl_vec<uint32_t>& recurrentToOutputWeightsDimensions, |
| 91 | const std::vector<float>& recurrentToOutputWeightsValue, |
| 92 | const hidl_vec<uint32_t>& cellToInputWeightsDimensions, |
| 93 | const std::vector<float>& cellToInputWeightsValue, |
| 94 | const hidl_vec<uint32_t>& cellToForgetWeightsDimensions, |
| 95 | const std::vector<float>& cellToForgetWeightsValue, |
| 96 | const hidl_vec<uint32_t>& cellToOutputWeightsDimensions, |
| 97 | const std::vector<float>& cellToOutputWeightsValue, |
| 98 | const hidl_vec<uint32_t>& inputGateBiasDimensions, |
| 99 | const std::vector<float>& inputGateBiasValue, |
| 100 | const hidl_vec<uint32_t>& forgetGateBiasDimensions, |
| 101 | const std::vector<float>& forgetGateBiasValue, |
| 102 | const hidl_vec<uint32_t>& cellBiasDimensions, |
| 103 | const std::vector<float>& cellBiasValue, |
| 104 | const hidl_vec<uint32_t>& outputGateBiasDimensions, |
| 105 | const std::vector<float>& outputGateBiasValue, |
| 106 | const hidl_vec<uint32_t>& projectionWeightsDimensions, |
| 107 | const std::vector<float>& projectionWeightsValue, |
| 108 | const hidl_vec<uint32_t>& projectionBiasDimensions, |
| 109 | const std::vector<float>& projectionBiasValue, |
| 110 | const hidl_vec<uint32_t>& outputStateInDimensions, |
| 111 | const std::vector<float>& outputStateInValue, |
| 112 | const hidl_vec<uint32_t>& cellStateInDimensions, |
| 113 | const std::vector<float>& cellStateInValue, |
| 114 | const hidl_vec<uint32_t>& activationFunctionDimensions, |
| 115 | const std::vector<int32_t>& activationFunctionValue, |
| 116 | const hidl_vec<uint32_t>& cellClippingThresholdDimensions, |
| 117 | const std::vector<float>& cellClippingThresholdValue, |
| 118 | const hidl_vec<uint32_t>& projectionClippingThresholdDimensions, |
| 119 | const std::vector<float>& projectionClippingThresholdValue, |
| 120 | const hidl_vec<uint32_t>& scratchBufferDimensions, |
| 121 | const std::vector<float>& scratchBufferValue, |
| 122 | const hidl_vec<uint32_t>& outputStateOutDimensions, |
| 123 | const std::vector<float>& outputStateOutValue, |
| 124 | const hidl_vec<uint32_t>& cellStateOutDimensions, |
| 125 | const std::vector<float>& cellStateOutValue, |
| 126 | const hidl_vec<uint32_t>& outputDimensions, |
| 127 | const std::vector<float>& outputValue, |
| 128 | armnn::Compute compute) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 129 | { |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 130 | auto driver = std::make_unique<ArmnnDriver>(DriverOptions(compute)); |
Matteo Martincigh | 8b287c2 | 2018-09-07 09:25:10 +0100 | [diff] [blame] | 131 | V1_0::Model model = {}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 132 | |
| 133 | // Inputs: |
| 134 | // 00: The input: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, input_size], where |
| 135 | // “batch_size” corresponds to the batching dimension, and “input_size” is the size of the input. |
| 136 | AddInputOperand(model, inputDimensions); |
| 137 | |
| 138 | // 01: The input-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 139 | // [num_units, input_size], where “num_units” corresponds to the number of cell units. |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 140 | AddTensorOperand(model, inputToInputWeightsDimensions, inputToInputWeightsValue, V1_0::OperandType::TENSOR_FLOAT32, |
Kevin May | f29a2c5 | 2019-03-14 11:56:32 +0000 | [diff] [blame] | 141 | CreateNoValueLifeTime(inputToInputWeightsDimensions)); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 142 | // 02: The input-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 143 | // [num_units, input_size]. |
| 144 | AddTensorOperand(model, inputToForgetWeightsDimensions, inputToForgetWeightsValue); |
| 145 | // 03: The input-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size]. |
| 146 | AddTensorOperand(model, inputToCellWeightsDimensions, inputToCellWeightsValue); |
| 147 | // 04: The input-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 148 | // [num_units, input_size]. |
| 149 | AddTensorOperand(model, inputToOutputWeightsDimensions, inputToOutputWeightsValue); |
| 150 | // 05: The recurrent-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 151 | // [num_units, output_size], where “output_size” corresponds to either the number of cell units (i.e., |
| 152 | // “num_units”), or the second dimension of the “projection_weights”, if defined. |
Kevin May | f29a2c5 | 2019-03-14 11:56:32 +0000 | [diff] [blame] | 153 | AddTensorOperand(model, recurrentToInputWeightsDimensions, recurrentToInputWeightsValue, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 154 | V1_0::OperandType::TENSOR_FLOAT32, CreateNoValueLifeTime(recurrentToInputWeightsDimensions)); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 155 | // 06: The recurrent-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 156 | // [num_units, output_size]. |
| 157 | AddTensorOperand(model, recurrentToForgetWeightsDimensions, recurrentToForgetWeightsValue); |
| 158 | // 07: The recurrent-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 159 | // [num_units, output_size]. |
| 160 | AddTensorOperand(model, recurrentToCellWeightsDimensions, recurrentToCellWeightsValue); |
| 161 | // 08: The recurrent-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 162 | // [num_units, output_size]. |
| 163 | AddTensorOperand(model, recurrentToOutputWeightsDimensions, recurrentToOutputWeightsValue); |
| 164 | // 09: The cell-to-input weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Kevin May | f29a2c5 | 2019-03-14 11:56:32 +0000 | [diff] [blame] | 165 | AddTensorOperand(model, cellToInputWeightsDimensions, cellToInputWeightsValue, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 166 | V1_0::OperandType::TENSOR_FLOAT32, CreateNoValueLifeTime(cellToInputWeightsDimensions)); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 167 | // 10: The cell-to-forget weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Kevin May | f29a2c5 | 2019-03-14 11:56:32 +0000 | [diff] [blame] | 168 | AddTensorOperand(model, cellToForgetWeightsDimensions, cellToForgetWeightsValue, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 169 | V1_0::OperandType::TENSOR_FLOAT32, CreateNoValueLifeTime(cellToForgetWeightsDimensions)); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 170 | // 11: The cell-to-output weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Kevin May | f29a2c5 | 2019-03-14 11:56:32 +0000 | [diff] [blame] | 171 | AddTensorOperand(model, cellToOutputWeightsDimensions, cellToOutputWeightsValue, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 172 | V1_0::OperandType::TENSOR_FLOAT32, CreateNoValueLifeTime(cellToOutputWeightsDimensions)); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 173 | // 12: The input gate bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Kevin May | f29a2c5 | 2019-03-14 11:56:32 +0000 | [diff] [blame] | 174 | AddTensorOperand(model, inputGateBiasDimensions, inputGateBiasValue, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 175 | V1_0::OperandType::TENSOR_FLOAT32, CreateNoValueLifeTime(inputGateBiasDimensions)); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 176 | // 13: The forget gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
| 177 | AddTensorOperand(model, forgetGateBiasDimensions, forgetGateBiasValue); |
| 178 | // 14: The cell bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
| 179 | AddTensorOperand(model, cellBiasDimensions, cellBiasValue); |
| 180 | // 15: The output gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
| 181 | AddTensorOperand(model, outputGateBiasDimensions, outputGateBiasValue); |
| 182 | // 16: The projection weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 183 | // [output_size, num_units]. |
Kevin May | f29a2c5 | 2019-03-14 11:56:32 +0000 | [diff] [blame] | 184 | AddTensorOperand(model, projectionWeightsDimensions, projectionWeightsValue, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 185 | V1_0::OperandType::TENSOR_FLOAT32, CreateNoValueLifeTime(projectionWeightsDimensions)); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 186 | // 17: The projection bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [output_size]. |
Kevin May | f29a2c5 | 2019-03-14 11:56:32 +0000 | [diff] [blame] | 187 | AddTensorOperand(model, projectionBiasDimensions, projectionBiasValue, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 188 | V1_0::OperandType::TENSOR_FLOAT32, CreateNoValueLifeTime(projectionBiasDimensions)); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 189 | |
| 190 | // 18: The output state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
| 191 | AddInputOperand(model, outputStateInDimensions); |
| 192 | // 19: The cell state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
| 193 | AddInputOperand(model, cellStateInDimensions); |
| 194 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 195 | // Constant scalar values (the VTS test adds these as tensors of dim {}) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 196 | // 20: The activation function: A value indicating the activation function: |
| 197 | // 0: None; 1: Relu; 3: Relu6; 4: Tanh; 6: Sigmoid. |
| 198 | AddTensorOperand(model, activationFunctionDimensions, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 199 | activationFunctionValue, V1_0::OperandType::INT32); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 200 | // 21: The clipping threshold: for the cell state, such that values are bound within [-cell_clip, cell_clip]. |
| 201 | // If set to 0.0 then clipping is disabled. |
| 202 | AddTensorOperand(model, cellClippingThresholdDimensions, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 203 | cellClippingThresholdValue, V1_0::OperandType::FLOAT32); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 204 | // 22: The clipping threshold: for the output from the projection layer, such that values are bound within |
| 205 | // [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled. |
| 206 | AddTensorOperand(model, projectionClippingThresholdDimensions, |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 207 | projectionClippingThresholdValue, V1_0::OperandType::FLOAT32); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 208 | |
| 209 | // Outputs: |
| 210 | // 0: The scratch buffer: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units * 4] with |
| 211 | // CIFG, or [batch_size, num_units * 3] without CIFG. |
| 212 | AddOutputOperand(model, scratchBufferDimensions); |
| 213 | // 1: The output state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
| 214 | AddOutputOperand(model, outputStateOutDimensions); |
| 215 | // 2: The cell state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
| 216 | AddOutputOperand(model, cellStateOutDimensions); |
| 217 | // 3: The output: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. This is |
| 218 | // effectively the same as the current “output state (out)” value. |
| 219 | AddOutputOperand(model, outputDimensions); |
| 220 | |
| 221 | // make the lstm operation |
| 222 | model.operations.resize(1); |
Matteo Martincigh | 8b287c2 | 2018-09-07 09:25:10 +0100 | [diff] [blame] | 223 | model.operations[0].type = V1_0::OperationType::LSTM; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 224 | model.operations[0].inputs = |
| 225 | hidl_vec<uint32_t> {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22}; |
| 226 | model.operations[0].outputs = hidl_vec<uint32_t> {23, 24, 25, 26}; |
| 227 | |
| 228 | // define the input values |
| 229 | hidl_vec<RequestArgument> inputArguments; |
| 230 | inputArguments.resize(3); |
| 231 | |
| 232 | inputArguments[0] = CreateRequestArgument<float>(inputValue, 0); |
| 233 | inputArguments[1] = CreateRequestArgument<float>(outputStateInValue, 1); |
| 234 | inputArguments[2] = CreateRequestArgument<float>(cellStateInValue, 2); |
| 235 | |
| 236 | // define the expected output values |
| 237 | hidl_vec<RequestArgument> outputArguments; |
| 238 | outputArguments.resize(4); |
| 239 | |
| 240 | outputArguments[0] = CreateRequestArgument<float>(scratchBufferValue, 3); |
| 241 | outputArguments[1] = CreateRequestArgument<float>(outputStateOutValue, 4); |
| 242 | outputArguments[2] = CreateRequestArgument<float>(cellStateOutValue, 5); |
| 243 | outputArguments[3] = CreateRequestArgument<float>(outputValue, 6); |
| 244 | |
| 245 | Request request = {}; |
| 246 | request.inputs = inputArguments; |
| 247 | request.outputs = outputArguments; |
| 248 | |
| 249 | // set the input data |
| 250 | AddPoolAndSetData(inputValue.size(), request, inputValue.data()); |
| 251 | AddPoolAndSetData(outputStateInValue.size(), request, outputStateInValue.data()); |
| 252 | AddPoolAndSetData(cellStateInValue.size(), request, cellStateInValue.data()); |
| 253 | |
| 254 | // add memory for the outputs |
| 255 | AddPoolAndGetData(scratchBufferValue.size(), request); |
| 256 | android::sp<IMemory> outputStateOutMemory = AddPoolAndGetData(outputStateOutValue.size(), request); |
| 257 | float* outputStateOutData = static_cast<float*>(static_cast<void*>(outputStateOutMemory->getPointer())); |
| 258 | android::sp<IMemory> cellStateOutMemory = AddPoolAndGetData(cellStateOutValue.size(), request); |
| 259 | float* cellStateOutData = static_cast<float*>(static_cast<void*>(cellStateOutMemory->getPointer())); |
| 260 | android::sp<IMemory> outputMemory = AddPoolAndGetData(outputValue.size(), request); |
| 261 | float* outputData = static_cast<float*>(static_cast<void*>(outputMemory->getPointer())); |
| 262 | |
| 263 | // make the prepared model and run the execution |
Sadik Armagan | e6e54a8 | 2019-05-08 10:18:05 +0100 | [diff] [blame] | 264 | android::sp<V1_0::IPreparedModel> preparedModel = PrepareModel(model, *driver); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 265 | if (preparedModel.get() != nullptr) |
| 266 | { |
| 267 | Execute(preparedModel, request); |
| 268 | } |
| 269 | |
| 270 | // check the results |
| 271 | for (size_t i = 0; i < outputStateOutValue.size(); ++i) |
| 272 | { |
| 273 | BOOST_TEST(TolerantCompareEqual(outputStateOutValue[i], outputStateOutData[i]), |
| 274 | "outputStateOut[" << i << "]: " << outputStateOutValue[i] << " != " << outputStateOutData[i]); |
| 275 | } |
| 276 | for (size_t i = 0; i < cellStateOutValue.size(); ++i) |
| 277 | { |
| 278 | BOOST_TEST(TolerantCompareEqual(cellStateOutValue[i], cellStateOutData[i]), |
| 279 | "cellStateOut[" << i << "]: " << cellStateOutValue[i] << " != " << cellStateOutData[i]); |
| 280 | } |
| 281 | for (size_t i = 0; i < outputValue.size(); ++i) |
| 282 | { |
| 283 | BOOST_TEST(TolerantCompareEqual(outputValue[i], outputData[i]), |
| 284 | "output[" << i << "]: " << outputValue[i] << " != " << outputData[i]); |
| 285 | } |
| 286 | } |
| 287 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 288 | void LstmNoCifgNoPeepholeNoProjection(armnn::Compute compute) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 289 | { |
| 290 | // This replicates android/frameworks/ml/nn/runtime/test/generated/vts_models/lstm.model.cpp |
| 291 | // with values from android/frameworks/ml/nn/runtime/test/generated/examples/lstm.example.cpp |
| 292 | // and weights, biases and scalars passed as CONSTANT_COPY tensors (instead of MODEL_INPUT tensors). |
| 293 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 294 | uint32_t batchSize = 1; |
| 295 | uint32_t inputSize = 2; |
| 296 | uint32_t numUnits = 4; |
| 297 | uint32_t outputSize = numUnits; |
| 298 | |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 299 | // Inputs: |
| 300 | // 00: The input: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, input_size], where |
| 301 | // “batch_size” corresponds to the batching dimension, and “input_size” is the size of the input. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 302 | hidl_vec<uint32_t> inputDimensions{batchSize, inputSize}; |
| 303 | std::vector<float> inputValue{2.0f, 3.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 304 | |
| 305 | // 01: The input-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 306 | // [num_units, input_size], where “num_units” corresponds to the number of cell units. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 307 | hidl_vec<uint32_t> inputToInputWeightsDimensions{numUnits, inputSize}; |
| 308 | std::vector<float> inputToInputWeightsValue{-0.45018822f, -0.02338299f, |
| 309 | -0.08705890f, -0.34550029f, |
| 310 | 0.04266912f, -0.15680569f, |
| 311 | -0.34856534f, 0.43890524f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 312 | // 02: The input-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 313 | // [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 314 | hidl_vec<uint32_t> inputToForgetWeightsDimensions{numUnits, inputSize}; |
| 315 | std::vector<float> inputToForgetWeightsValue{ 0.09701663f, 0.20334584f, |
| 316 | -0.50592935f, -0.31343272f, |
| 317 | -0.40032279f, 0.44781327f, |
| 318 | 0.01387155f, -0.35593212f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 319 | // 03: The input-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 320 | hidl_vec<uint32_t> inputToCellWeightsDimensions{numUnits, inputSize}; |
| 321 | std::vector<float> inputToCellWeightsValue{-0.50013041f, 0.13702840f, |
| 322 | 0.11810488f, 0.20131630f, |
| 323 | -0.20583314f, 0.44344562f, |
| 324 | 0.22077113f, -0.29909778f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 325 | // 04: The input-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 326 | // [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 327 | hidl_vec<uint32_t> inputToOutputWeightsDimensions{numUnits, inputSize}; |
| 328 | std::vector<float> inputToOutputWeightsValue{-0.25065863f, -0.28290087f, |
| 329 | 0.04613829f, 0.40525138f, |
| 330 | 0.44272184f, 0.03897077f, |
| 331 | -0.15568960f, 0.19487578f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 332 | // 05: The recurrent-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 333 | // [num_units, output_size], where “output_size” corresponds to either the number of cell units (i.e., |
| 334 | // “num_units”), or the second dimension of the “projection_weights”, if defined. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 335 | hidl_vec<uint32_t> recurrentToInputWeightsDimensions{numUnits, outputSize}; |
| 336 | std::vector<float> recurrentToInputWeightsValue{-0.00635350f, -0.20423880f, 0.31454784f, -0.35746509f, |
| 337 | 0.28902304f, 0.08183324f, -0.16555229f, 0.02286911f, |
| 338 | -0.13566875f, 0.03034258f, 0.48091322f, -0.12528998f, |
| 339 | 0.24077177f, -0.51332325f, -0.33502164f, 0.10629296f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 340 | // 06: The recurrent-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 341 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 342 | hidl_vec<uint32_t> recurrentToForgetWeightsDimensions{numUnits, outputSize}; |
| 343 | std::vector<float> recurrentToForgetWeightsValue{-0.48684245f, -0.06655136f, 0.42224967f, 0.21126390f, |
| 344 | 0.27654213f, 0.20864892f, -0.07646349f, 0.45877004f, |
| 345 | 0.00141793f, -0.14609534f, 0.36447752f, 0.09196436f, |
| 346 | 0.28053468f, 0.01560611f, -0.20127171f, -0.01140004f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 347 | // 07: The recurrent-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 348 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 349 | hidl_vec<uint32_t> recurrentToCellWeightsDimensions{numUnits, outputSize}; |
| 350 | std::vector<float> recurrentToCellWeightsValue{-0.34074140f, 0.24443203f, -0.20785320f, 0.26320225f, |
| 351 | 0.05695659f, -0.00123841f, -0.47447860f, -0.35869038f, |
| 352 | -0.06418842f, -0.13502428f, -0.50176400f, 0.22830659f, |
| 353 | -0.46367589f, 0.26016325f, -0.03894562f, -0.16368064f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 354 | // 08: The recurrent-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 355 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 356 | hidl_vec<uint32_t> recurrentToOutputWeightsDimensions{numUnits, outputSize}; |
| 357 | std::vector<float> recurrentToOutputWeightsValue{ 0.43385774f, -0.17194885f, 0.27182370f, 0.09215671f, |
| 358 | 0.24107647f, -0.39835793f, 0.18212086f, 0.01301402f, |
| 359 | 0.48572797f, -0.50656658f, 0.20047462f, -0.20607421f, |
| 360 | -0.51818722f, -0.15390486f, 0.04681480f, 0.39922136f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 361 | // 09: The cell-to-input weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 362 | hidl_vec<uint32_t> cellToInputWeightsDimensions{0}; |
| 363 | std::vector<float> cellToInputWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 364 | // 10: The cell-to-forget weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 365 | hidl_vec<uint32_t> cellToForgetWeightsDimensions{0}; |
| 366 | std::vector<float> cellToForgetWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 367 | // 11: The cell-to-output weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 368 | hidl_vec<uint32_t> cellToOutputWeightsDimensions{0}; |
| 369 | std::vector<float> cellToOutputWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 370 | // 12: The input gate bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 371 | hidl_vec<uint32_t> inputGateBiasDimensions{numUnits}; |
| 372 | std::vector<float> inputGateBiasValue(numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 373 | // 13: The forget gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 374 | hidl_vec<uint32_t> forgetGateBiasDimensions{numUnits}; |
| 375 | std::vector<float> forgetGateBiasValue(numUnits, 1.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 376 | // 14: The cell bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 377 | hidl_vec<uint32_t> cellBiasDimensions{numUnits}; |
| 378 | std::vector<float> cellBiasValue(numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 379 | // 15: The output gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 380 | hidl_vec<uint32_t> outputGateBiasDimensions{numUnits}; |
| 381 | std::vector<float> outputGateBiasValue(numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 382 | // 16: The projection weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 383 | // [output_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 384 | hidl_vec<uint32_t> projectionWeightsDimensions{0}; |
| 385 | std::vector<float> projectionWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 386 | // 17: The projection bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 387 | hidl_vec<uint32_t> projectionBiasDimensions{0}; |
| 388 | std::vector<float> projectionBiasValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 389 | |
| 390 | // 18: The output state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 391 | hidl_vec<uint32_t> outputStateInDimensions{batchSize, outputSize}; |
| 392 | std::vector<float> outputStateInValue(batchSize * outputSize, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 393 | // 19: The cell state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 394 | hidl_vec<uint32_t> cellStateInDimensions{batchSize, numUnits}; |
| 395 | std::vector<float> cellStateInValue(batchSize * numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 396 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 397 | // Constant scalar values (the VTS test adds these as tensors of dim {}) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 398 | // 20: The activation function: A value indicating the activation function: |
| 399 | // 0: None; 1: Relu; 3: Relu6; 4: Tanh; 6: Sigmoid. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 400 | hidl_vec<uint32_t> activationFunctionDimensions{}; |
| 401 | std::vector<int32_t> activationFunctionValue{4}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 402 | // 21: The clipping threshold: for the cell state, such that values are bound within [-cell_clip, cell_clip]. |
| 403 | // If set to 0.0 then clipping is disabled. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 404 | hidl_vec<uint32_t> cellClippingThresholdDimensions{}; |
| 405 | std::vector<float> cellClippingThresholdValue{0.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 406 | // 22: The clipping threshold: for the output from the projection layer, such that values are bound within |
| 407 | // [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 408 | hidl_vec<uint32_t> projectionClippingThresholdDimensions{}; |
| 409 | std::vector<float> projectionClippingThresholdValue{0.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 410 | |
| 411 | // Outputs: |
| 412 | // 0: The scratch buffer: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units * 4] with |
| 413 | // CIFG, or [batch_size, num_units * 3] without CIFG. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 414 | // HOWEVER, by looking at the code, seems that it's the opposite: (cifg ? 3 : 4) * numUnits |
| 415 | // Refer to: android/frameworks/ml/nn/common/operations/LSTM.cpp:319 |
| 416 | // android/frameworks/ml/nn/common/operations/LSTMTest.cpp:114 |
| 417 | // tensorflow/tensorflow/contrib/lite/kernels/lstm.cc:332 |
| 418 | hidl_vec<uint32_t> scratchBufferDimensions{batchSize, numUnits * 4}; |
| 419 | std::vector<float> scratchBufferValue(batchSize * numUnits * 4, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 420 | // 1: The output state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 421 | hidl_vec<uint32_t> outputStateOutDimensions{batchSize, outputSize}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 422 | std::vector<float> outputStateOutValue {-0.0297319f, 0.122947f, 0.208851f, -0.153588f}; |
| 423 | // 2: The cell state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 424 | hidl_vec<uint32_t> cellStateOutDimensions{batchSize, numUnits}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 425 | std::vector<float> cellStateOutValue {-0.145439f, 0.157475f, 0.293663f, -0.277353f}; |
| 426 | // 3: The output: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. This is |
| 427 | // effectively the same as the current “output state (out)” value. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 428 | hidl_vec<uint32_t> outputDimensions{batchSize, outputSize}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 429 | std::vector<float> outputValue {-0.02973187f, 0.1229473f, 0.20885126f, -0.15358765f}; |
| 430 | |
| 431 | LstmTestImpl(inputDimensions, inputValue, |
| 432 | inputToInputWeightsDimensions, inputToInputWeightsValue, |
| 433 | inputToForgetWeightsDimensions, inputToForgetWeightsValue, |
| 434 | inputToCellWeightsDimensions, inputToCellWeightsValue, |
| 435 | inputToOutputWeightsDimensions, inputToOutputWeightsValue, |
| 436 | recurrentToInputWeightsDimensions, recurrentToInputWeightsValue, |
| 437 | recurrentToForgetWeightsDimensions, recurrentToForgetWeightsValue, |
| 438 | recurrentToCellWeightsDimensions, recurrentToCellWeightsValue, |
| 439 | recurrentToOutputWeightsDimensions, recurrentToOutputWeightsValue, |
| 440 | cellToInputWeightsDimensions, cellToInputWeightsValue, |
| 441 | cellToForgetWeightsDimensions, cellToForgetWeightsValue, |
| 442 | cellToOutputWeightsDimensions, cellToOutputWeightsValue, |
| 443 | inputGateBiasDimensions, inputGateBiasValue, |
| 444 | forgetGateBiasDimensions, forgetGateBiasValue, |
| 445 | cellBiasDimensions, cellBiasValue, |
| 446 | outputGateBiasDimensions, outputGateBiasValue, |
| 447 | projectionWeightsDimensions, projectionWeightsValue, |
| 448 | projectionBiasDimensions, projectionBiasValue, |
| 449 | outputStateInDimensions, outputStateInValue, |
| 450 | cellStateInDimensions, cellStateInValue, |
| 451 | activationFunctionDimensions, activationFunctionValue, |
| 452 | cellClippingThresholdDimensions, cellClippingThresholdValue, |
| 453 | projectionClippingThresholdDimensions, projectionClippingThresholdValue, |
| 454 | scratchBufferDimensions, scratchBufferValue, |
| 455 | outputStateOutDimensions, outputStateOutValue, |
| 456 | cellStateOutDimensions, cellStateOutValue, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 457 | outputDimensions, outputValue, |
| 458 | compute); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 459 | } |
| 460 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 461 | void LstmCifgPeepholeNoProjection(armnn::Compute compute) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 462 | { |
| 463 | // This replicates android/frameworks/ml/nn/runtime/test/generated/vts_models/lstm2.model.cpp |
| 464 | // with values from android/frameworks/ml/nn/runtime/test/generated/examples/lstm2.example.cpp |
| 465 | // and weights, biases and scalars passed as CONSTANT_COPY tensors (instead of MODEL_INPUT tensors). |
| 466 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 467 | uint32_t batchSize = 1; |
| 468 | uint32_t inputSize = 2; |
| 469 | uint32_t numUnits = 4; |
| 470 | uint32_t outputSize = numUnits; |
| 471 | |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 472 | // Inputs: |
| 473 | // 00: The input: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, input_size], where |
| 474 | // “batch_size” corresponds to the batching dimension, and “input_size” is the size of the input. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 475 | hidl_vec<uint32_t> inputDimensions{batchSize, inputSize}; |
| 476 | std::vector<float> inputValue{2.0f, 3.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 477 | |
| 478 | // 01: The input-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 479 | // [num_units, input_size], where “num_units” corresponds to the number of cell units. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 480 | hidl_vec<uint32_t> inputToInputWeightsDimensions{0}; |
| 481 | std::vector<float> inputToInputWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 482 | // 02: The input-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 483 | // [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 484 | hidl_vec<uint32_t> inputToForgetWeightsDimensions{numUnits, inputSize}; |
| 485 | std::vector<float> inputToForgetWeightsValue{-0.55291498f, -0.42866567f, |
| 486 | 0.13056988f, -0.36333650f, |
| 487 | -0.22755712f, 0.28253698f, |
| 488 | 0.24407166f, 0.33826375f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 489 | // 03: The input-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 490 | hidl_vec<uint32_t> inputToCellWeightsDimensions{numUnits, inputSize}; |
| 491 | std::vector<float> inputToCellWeightsValue{-0.49770179f, -0.27711356f, |
| 492 | -0.09624726f, 0.05100781f, |
| 493 | 0.04717243f, 0.48944736f, |
| 494 | -0.38535351f, -0.17212132f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 495 | // 04: The input-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 496 | // [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 497 | hidl_vec<uint32_t> inputToOutputWeightsDimensions{numUnits, inputSize}; |
| 498 | std::vector<float> inputToOutputWeightsValue{ 0.10725588f, -0.02335852f, |
| 499 | -0.55932593f, -0.09426838f, |
| 500 | -0.44257352f, 0.54939759f, |
| 501 | 0.01533556f, 0.42751634f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 502 | // 05: The recurrent-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 503 | // [num_units, output_size], where “output_size” corresponds to either the number of cell units (i.e., |
| 504 | // “num_units”), or the second dimension of the “projection_weights”, if defined. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 505 | hidl_vec<uint32_t> recurrentToInputWeightsDimensions{0}; // VTS was {4, 4} -> {0} ? |
| 506 | std::vector<float> recurrentToInputWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 507 | // 06: The recurrent-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 508 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 509 | hidl_vec<uint32_t> recurrentToForgetWeightsDimensions{numUnits, outputSize}; |
| 510 | std::vector<float> recurrentToForgetWeightsValue{-0.13832897f, -0.05151010f, -0.23590070f, -0.16661474f, |
| 511 | -0.14340827f, 0.36986142f, 0.23414481f, 0.55899000f, |
| 512 | 0.10798943f, -0.41174671f, 0.17751795f, -0.34484994f, |
| 513 | -0.35874045f, -0.11352962f, 0.27268326f, 0.54058349f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 514 | // 07: The recurrent-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 515 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 516 | hidl_vec<uint32_t> recurrentToCellWeightsDimensions{numUnits, outputSize}; |
| 517 | std::vector<float> recurrentToCellWeightsValue{ 0.54066205f, -0.32668582f, -0.43562764f, -0.56094903f, |
| 518 | 0.42957711f, 0.01841056f, -0.32764608f, -0.33027974f, |
| 519 | -0.10826075f, 0.20675004f, 0.19069612f, -0.03026325f, |
| 520 | -0.54532051f, 0.33003211f, 0.44901288f, 0.21193194f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 521 | // 08: The recurrent-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 522 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 523 | hidl_vec<uint32_t> recurrentToOutputWeightsDimensions{numUnits, outputSize}; |
| 524 | std::vector<float> recurrentToOutputWeightsValue{0.41613156f, 0.42610586f, -0.16495961f, -0.56638730f, |
| 525 | 0.30579174f, -0.05115908f, -0.33941799f, 0.23364776f, |
| 526 | 0.11178309f, 0.09481031f, -0.26424935f, 0.46261835f, |
| 527 | 0.50248802f, 0.26114327f, -0.43736315f, 0.33149987f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 528 | // 09: The cell-to-input weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 529 | hidl_vec<uint32_t> cellToInputWeightsDimensions{0}; |
| 530 | std::vector<float> cellToInputWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 531 | // 10: The cell-to-forget weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 532 | hidl_vec<uint32_t> cellToForgetWeightsDimensions{4}; |
| 533 | std::vector<float> cellToForgetWeightsValue{0.47485286f, -0.51955009f, -0.24458408f, 0.31544167f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 534 | // 11: The cell-to-output weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 535 | hidl_vec<uint32_t> cellToOutputWeightsDimensions{4}; |
| 536 | std::vector<float> cellToOutputWeightsValue{-0.17135078f, 0.82760304f, 0.85573703f, -0.77109635f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 537 | // 12: The input gate bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 538 | hidl_vec<uint32_t> inputGateBiasDimensions{0}; // VTS was {4} -> {0} ? |
| 539 | std::vector<float> inputGateBiasValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 540 | // 13: The forget gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 541 | hidl_vec<uint32_t> forgetGateBiasDimensions{numUnits}; |
| 542 | std::vector<float> forgetGateBiasValue(numUnits, 1.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 543 | // 14: The cell bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 544 | hidl_vec<uint32_t> cellBiasDimensions{numUnits}; |
| 545 | std::vector<float> cellBiasValue(numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 546 | // 15: The output gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 547 | hidl_vec<uint32_t> outputGateBiasDimensions{numUnits}; |
| 548 | std::vector<float> outputGateBiasValue(numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 549 | // 16: The projection weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 550 | // [output_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 551 | hidl_vec<uint32_t> projectionWeightsDimensions{0}; |
| 552 | std::vector<float> projectionWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 553 | // 17: The projection bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 554 | hidl_vec<uint32_t> projectionBiasDimensions{0}; |
| 555 | std::vector<float> projectionBiasValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 556 | |
| 557 | // 18: The output state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 558 | hidl_vec<uint32_t> outputStateInDimensions{batchSize, outputSize}; |
| 559 | std::vector<float> outputStateInValue(batchSize * outputSize, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 560 | // 19: The cell state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 561 | hidl_vec<uint32_t> cellStateInDimensions{batchSize, numUnits}; |
| 562 | std::vector<float> cellStateInValue(batchSize * numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 563 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 564 | // Constant scalar values (the VTS test adds these as tensors of dim {}) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 565 | // 20: The activation function: A value indicating the activation function: |
| 566 | // 0: None; 1: Relu; 3: Relu6; 4: Tanh; 6: Sigmoid. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 567 | hidl_vec<uint32_t> activationFunctionDimensions{}; |
| 568 | std::vector<int32_t> activationFunctionValue{4}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 569 | // 21: The clipping threshold: for the cell state, such that values are bound within [-cell_clip, cell_clip]. |
| 570 | // If set to 0.0 then clipping is disabled. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 571 | hidl_vec<uint32_t> cellClippingThresholdDimensions{}; |
| 572 | std::vector<float> cellClippingThresholdValue{0.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 573 | // 22: The clipping threshold: for the output from the projection layer, such that values are bound within |
| 574 | // [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 575 | hidl_vec<uint32_t> projectionClippingThresholdDimensions{}; |
| 576 | std::vector<float> projectionClippingThresholdValue{0.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 577 | |
| 578 | // Outputs: |
| 579 | // 0: The scratch buffer: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units * 4] with |
| 580 | // CIFG, or [batch_size, num_units * 3] without CIFG. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 581 | // HOWEVER, by looking at the code, seems that it's the opposite: (cifg ? 3 : 4) * numUnits |
| 582 | // Refer to: android/frameworks/ml/nn/common/operations/LSTM.cpp:319 |
| 583 | // android/frameworks/ml/nn/common/operations/LSTMTest.cpp:114 |
| 584 | // tensorflow/tensorflow/contrib/lite/kernels/lstm.cc:332 |
| 585 | hidl_vec<uint32_t> scratchBufferDimensions{batchSize, numUnits * 3}; |
| 586 | std::vector<float> scratchBufferValue(batchSize * numUnits * 3, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 587 | // 1: The output state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 588 | hidl_vec<uint32_t> outputStateOutDimensions{batchSize, outputSize}; |
| 589 | std::vector<float> outputStateOutValue{-0.364445f, -0.00352185f, 0.128866f, -0.0516365f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 590 | // 2: The cell state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 591 | hidl_vec<uint32_t> cellStateOutDimensions{batchSize, numUnits}; |
| 592 | std::vector<float> cellStateOutValue{-0.760444f, -0.0180416f, 0.182264f, -0.0649371f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 593 | // 3: The output: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. This is |
| 594 | // effectively the same as the current “output state (out)” value. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 595 | hidl_vec<uint32_t> outputDimensions{batchSize, outputSize}; |
| 596 | std::vector<float> outputValue{-0.36444446f, -0.00352185f, 0.12886585f, -0.05163646f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 597 | |
| 598 | LstmTestImpl(inputDimensions, inputValue, |
| 599 | inputToInputWeightsDimensions, inputToInputWeightsValue, |
| 600 | inputToForgetWeightsDimensions, inputToForgetWeightsValue, |
| 601 | inputToCellWeightsDimensions, inputToCellWeightsValue, |
| 602 | inputToOutputWeightsDimensions, inputToOutputWeightsValue, |
| 603 | recurrentToInputWeightsDimensions, recurrentToInputWeightsValue, |
| 604 | recurrentToForgetWeightsDimensions, recurrentToForgetWeightsValue, |
| 605 | recurrentToCellWeightsDimensions, recurrentToCellWeightsValue, |
| 606 | recurrentToOutputWeightsDimensions, recurrentToOutputWeightsValue, |
| 607 | cellToInputWeightsDimensions, cellToInputWeightsValue, |
| 608 | cellToForgetWeightsDimensions, cellToForgetWeightsValue, |
| 609 | cellToOutputWeightsDimensions, cellToOutputWeightsValue, |
| 610 | inputGateBiasDimensions, inputGateBiasValue, |
| 611 | forgetGateBiasDimensions, forgetGateBiasValue, |
| 612 | cellBiasDimensions, cellBiasValue, |
| 613 | outputGateBiasDimensions, outputGateBiasValue, |
| 614 | projectionWeightsDimensions, projectionWeightsValue, |
| 615 | projectionBiasDimensions, projectionBiasValue, |
| 616 | outputStateInDimensions, outputStateInValue, |
| 617 | cellStateInDimensions, cellStateInValue, |
| 618 | activationFunctionDimensions, activationFunctionValue, |
| 619 | cellClippingThresholdDimensions, cellClippingThresholdValue, |
| 620 | projectionClippingThresholdDimensions, projectionClippingThresholdValue, |
| 621 | scratchBufferDimensions, scratchBufferValue, |
| 622 | outputStateOutDimensions, outputStateOutValue, |
| 623 | cellStateOutDimensions, cellStateOutValue, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 624 | outputDimensions, outputValue, |
| 625 | compute); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 626 | } |
| 627 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 628 | void LstmNoCifgPeepholeProjection(armnn::Compute compute) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 629 | { |
| 630 | // This replicates android/frameworks/ml/nn/runtime/test/generated/vts_models/lstm3.model.cpp |
| 631 | // with values from android/frameworks/ml/nn/runtime/test/generated/examples/lstm3.example.cpp |
| 632 | // and weights, biases and scalars passed as CONSTANT_COPY tensors (instead of MODEL_INPUT tensors). |
| 633 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 634 | uint32_t batchSize = 2; |
| 635 | uint32_t inputSize = 5; |
| 636 | uint32_t numUnits = 20; |
| 637 | uint32_t outputSize = 16; |
| 638 | |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 639 | // Inputs: |
| 640 | // 00: The input: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, input_size], where |
| 641 | // “batch_size” corresponds to the batching dimension, and “input_size” is the size of the input. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 642 | hidl_vec<uint32_t> inputDimensions{batchSize, inputSize}; |
| 643 | std::vector<float> inputValue{0.787926f, 0.151646f, 0.071352f, 0.118426f, 0.458058f, |
| 644 | 0.295743f, 0.544053f, 0.690064f, 0.858138f, 0.497181f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 645 | |
| 646 | // 01: The input-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 647 | // [num_units, input_size], where “num_units” corresponds to the number of cell units. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 648 | hidl_vec<uint32_t> inputToInputWeightsDimensions{numUnits, inputSize}; |
| 649 | std::vector<float> inputToInputWeightsValue |
| 650 | { |
| 651 | 0.0213936830f, 0.0612455100f, 0.0469051670f, -0.0146576770f, -0.0314946300f, |
| 652 | 0.0917180300f, 0.1464780100f, 0.1079719300f, -0.0057968358f, 0.0019193048f, |
| 653 | -0.2726754000f, 0.1015402900f, -0.0185398850f, 0.0803498850f, -0.1026238500f, |
| 654 | -0.0225997870f, -0.0912115500f, -0.0086759670f, -0.0452061030f, -0.0821282000f, |
| 655 | -0.0080459520f, 0.0154780810f, 0.0552172470f, 0.0387195870f, 0.0441536270f, |
| 656 | -0.0645324300f, 0.0503182500f, -0.0469351080f, -0.0081644309f, 0.0145742260f, |
| 657 | -0.1671009000f, -0.1551955200f, -0.1681979700f, -0.1397126900f, -0.1195305900f, |
| 658 | 0.2500548700f, -0.2279098300f, 0.0098550870f, -0.0281409580f, -0.1120069800f, |
| 659 | 0.1129540800f, -0.0035217577f, 0.0544850750f, 0.0518469500f, 0.0647112060f, |
| 660 | 0.1098919300f, 0.1167478600f, 0.0349060700f, 0.0772735700f, 0.1139058500f, |
| 661 | -0.1863375000f, -0.1034451000f, -0.1394518900f, -0.0494012270f, -0.1876706300f, |
| 662 | 0.0424839030f, 0.1423355200f, 0.1383258100f, 0.1835016500f, 0.1454560300f, |
| 663 | -0.0285457040f, 0.0249395310f, 0.0509297180f, 0.0076203286f, -0.0029723682f, |
| 664 | -0.0424842240f, -0.1182759600f, -0.0917110400f, -0.1080862800f, -0.1632798800f, |
| 665 | -0.2273378000f, -0.0993647000f, -0.0171551070f, 0.0023917493f, 0.0492727640f, |
| 666 | 0.0038534778f, 0.0547645050f, 0.0897537840f, 0.0694723400f, 0.0801447600f, |
| 667 | -0.0454423400f, -0.0497073000f, -0.0713563100f, -0.0489291060f, -0.0040420120f, |
| 668 | -0.0092840260f, 0.0180420540f, 0.0036860977f, -0.0742730200f, -0.1143460400f, |
| 669 | -0.0189954560f, 0.0314875430f, 0.0128349080f, 0.0199777540f, 0.0442566540f, |
| 670 | -0.3929261300f, -0.1851933400f, -0.1165128100f, -0.0680989200f, 0.0113736770f |
| 671 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 672 | // 02: The input-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 673 | // [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 674 | hidl_vec<uint32_t> inputToForgetWeightsDimensions{numUnits, inputSize}; |
| 675 | std::vector<float> inputToForgetWeightsValue |
| 676 | { |
| 677 | -0.0018401089f, -0.0048522370f, 0.0369842400f, 0.0141817040f, 0.0282732360f, |
| 678 | -0.0167261940f, -0.0524975900f, -0.1020426100f, 0.0086106600f, -0.0409795050f, |
| 679 | -0.0098991870f, 0.0192389200f, -0.0281772690f, -0.0853510300f, -0.1458549500f, |
| 680 | 0.1066256700f, -0.0190973100f, -0.0178835340f, -0.0047269356f, -0.0451033230f, |
| 681 | 0.0030784295f, 0.0767847750f, 0.0746369600f, 0.0945313950f, 0.0814421000f, |
| 682 | -0.1225789900f, -0.0339457580f, -0.0313034650f, 0.0456306260f, 0.0684388700f, |
| 683 | -0.1349294500f, -0.0124800070f, -0.0811829000f, -0.0722449900f, -0.0962879100f, |
| 684 | 0.0451009460f, 0.0012300825f, 0.0139646620f, 0.0993723940f, 0.0254305900f, |
| 685 | 0.0695832400f, 0.0342572960f, 0.0482646000f, 0.0626799700f, 0.0526250680f, |
| 686 | 0.1278466600f, 0.0707789700f, 0.0257259350f, 0.0416500900f, 0.0724190500f, |
| 687 | 0.0186686440f, -0.0373772940f, -0.0627778300f, -0.0883363600f, -0.0401206050f, |
| 688 | -0.0114055860f, -0.0078083350f, -0.0103013860f, -0.0051021670f, 0.0277174640f, |
| 689 | 0.0548342300f, 0.1144911100f, 0.1128965200f, 0.1093983900f, 0.1339650600f, |
| 690 | -0.0840216600f, -0.0190146200f, -0.0446783040f, -0.0772056500f, 0.0143500630f, |
| 691 | -0.1175795800f, -0.0652038000f, -0.0818573300f, -0.0767543240f, -0.0926143750f, |
| 692 | 0.1040549100f, 0.0529603360f, 0.0357558950f, 0.0358393860f, -0.0125405530f, |
| 693 | 0.0368812980f, 0.0291337600f, 0.0342015900f, 0.0544844700f, -0.0545233530f, |
| 694 | 0.0258271500f, 0.0232735500f, -0.0118571790f, -0.0011980024f, -0.0346417170f, |
| 695 | -0.0261250940f, -0.1758261500f, -0.1592365700f, -0.2748677400f, -0.0006143371f, |
| 696 | 0.0001771948f, -8.470171e-05f, 0.0265180700f, 0.0457907650f, 0.069564960f |
| 697 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 698 | // 03: The input-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 699 | hidl_vec<uint32_t> inputToCellWeightsDimensions{numUnits, inputSize}; |
| 700 | std::vector<float> inputToCellWeightsValue |
| 701 | { |
| 702 | -0.0458028300f, -0.0954946200f, -0.0324189850f, -0.0645463300f, -0.0435284530f, |
| 703 | 0.0430185870f, -0.0491523440f, -0.1241814400f, -0.0789854750f, -0.0759688900f, |
| 704 | 0.0194843620f, -0.1143496200f, -0.0074034138f, -0.0631484400f, -0.0929814950f, |
| 705 | 0.0062155537f, -0.0250343380f, -0.0028890965f, 0.0489295270f, 0.0623507500f, |
| 706 | 0.1066591800f, -0.0320367920f, -0.0850591600f, -0.1084335800f, -0.1300243300f, |
| 707 | -0.0368164370f, -0.0213013400f, -0.0165182390f, 0.0047691227f, -0.0025825808f, |
| 708 | 0.0660178660f, 0.0299915340f, -0.1065283600f, -0.1037554000f, -0.1305607100f, |
| 709 | -0.0326664300f, -0.0337024140f, -0.0064734240f, -0.0461169200f, 0.0144193390f, |
| 710 | -0.0251743230f, 0.0396852000f, 0.0817775060f, 0.0615746800f, 0.1021009500f, |
| 711 | -0.0096581940f, 0.0465117170f, 0.0360390600f, 0.0069369148f, 0.0159600950f, |
| 712 | -0.0650766600f, 0.0955159800f, 0.0535688360f, 0.0640871400f, 0.1283566700f, |
| 713 | -0.0087143290f, -0.2021196600f, -0.1209367400f, 0.0294504720f, 0.2849013000f, |
| 714 | -0.0292279010f, 0.1164364000f, -0.0856026300f, 0.0994178600f, -0.0369995650f, |
| 715 | -0.0288426260f, -0.0033637602f, -0.0170129020f, -0.0972086500f, -0.1119335100f, |
| 716 | -0.0291551170f, -0.0179360340f, -0.0097689360f, -0.0422332400f, -0.0361596350f, |
| 717 | 0.0650511200f, -0.0217428920f, -0.0233772120f, -0.0722136400f, -0.0643055200f, |
| 718 | 0.0545386500f, 0.0911498140f, 0.0638733100f, 0.0075183930f, 0.0559609530f, |
| 719 | 0.0697793440f, 0.0464111680f, 0.1050991100f, 0.0746389400f, 0.0075130584f, |
| 720 | 0.0128509820f, 0.0455543100f, 0.0569556880f, 0.0655528500f, 0.0508014560f, |
| 721 | -0.0098626830f, 0.0082677200f, -0.0265556090f, -0.0073611983f, -0.0014897042f |
| 722 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 723 | // 04: The input-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 724 | // [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 725 | hidl_vec<uint32_t> inputToOutputWeightsDimensions{numUnits, inputSize}; |
| 726 | std::vector<float> inputToOutputWeightsValue |
| 727 | { |
| 728 | -0.0998932000f, -0.0720195600f, -0.0528037730f, -0.1562959300f, -0.1500191800f, |
| 729 | -0.0765075100f, 0.0235985500f, -0.0751553550f, -0.0803770900f, -0.1509353400f, |
| 730 | 0.0295175520f, -0.0475139300f, 0.0103505310f, -0.0266485100f, -0.0168397220f, |
| 731 | -0.0231211630f, 0.0077019283f, 0.0128512570f, -0.0504064900f, -0.0129761000f, |
| 732 | -0.0217377470f, -0.0383057930f, -0.0687058600f, -0.0148124700f, -0.0012853940f, |
| 733 | 0.1012423600f, 0.0831228350f, 0.0533130060f, -0.0622356460f, -0.0756371540f, |
| 734 | -0.0278339030f, 0.0297749710f, 0.1130802000f, 0.0921890600f, 0.0950613500f, |
| 735 | -0.0866657640f, -0.0371627060f, -0.0388809140f, -0.0358328450f, -0.0144815640f, |
| 736 | -0.0982500300f, -0.1204856900f, -0.0976655860f, -0.0528763300f, -0.0964047000f, |
| 737 | -0.1136642900f, 0.0357775050f, 0.1356881900f, 0.0524513830f, 0.0506493040f, |
| 738 | 0.0579895100f, -0.0218523350f, -0.0998488440f, 0.0147404750f, -0.0788979460f, |
| 739 | 0.0497469900f, 0.0141604730f, 0.0697393200f, 0.0496494200f, 0.0333646460f, |
| 740 | 0.0819012400f, 0.0255353670f, 0.0508931650f, 0.0485142540f, 0.0694581300f, |
| 741 | -0.0789075640f, -0.0670761600f, -0.1184450800f, -0.0998668800f, -0.0750940300f, |
| 742 | 0.0626322600f, 0.1492558700f, 0.2018843600f, 0.1209845100f, 0.1463941500f, |
| 743 | 0.0015017595f, -0.0142673820f, -0.0341725700f, 0.0127114680f, 0.0028300495f, |
| 744 | -0.0247584820f, -0.0509854800f, -0.0821182000f, 0.0142256720f, 0.0215441580f, |
| 745 | 0.0894972500f, 0.0750526800f, -0.0020780868f, 0.0490825800f, 0.0647629500f, |
| 746 | -0.0229070630f, 0.0275624560f, 0.0401857350f, 0.0195675770f, -0.0155987390f, |
| 747 | -0.0490973030f, -0.0171218660f, -0.0833682340f, -0.0233200200f, -0.084095600f |
| 748 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 749 | // 05: The recurrent-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 750 | // [num_units, output_size], where “output_size” corresponds to either the number of cell units (i.e., |
| 751 | // “num_units”), or the second dimension of the “projection_weights”, if defined. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 752 | hidl_vec<uint32_t> recurrentToInputWeightsDimensions{numUnits, outputSize}; |
| 753 | std::vector<float> recurrentToInputWeightsValue |
| 754 | { |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 755 | -0.001374326f, -0.078856036f, 0.10672688f, 0.029162422f, // 00 |
| 756 | -0.11585556f, 0.02557986f, -0.13446963f, -0.035785314f, |
| 757 | -0.01244275f, 0.025961924f, -0.02337298f, -0.044228926f, |
| 758 | -0.055839065f, -0.046598054f, -0.010546039f, -0.06900766f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 759 | 0.027239809f, 0.022582639f, -0.013296484f, -0.05459212f, // 01 |
| 760 | 0.08981f, -0.045407712f, 0.08682226f, -0.06867011f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 761 | -0.14390695f, -0.02916037f, 0.000996957f, 0.091420636f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 762 | 0.14283475f, -0.07390571f, -0.06402044f, 0.062524505f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 763 | -0.093129106f, 0.04860203f, -0.08364217f, -0.08119002f, // 02 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 764 | 0.009352075f, 0.22920375f, 0.0016303885f, 0.11583097f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 765 | -0.13732095f, 0.012405723f, -0.07551853f, 0.06343048f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 766 | 0.12162708f, -0.031923793f, -0.014335606f, 0.01790974f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 767 | -0.10650317f, -0.0724401f, 0.08554849f, -0.05727212f, // 03 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 768 | 0.06556731f, -0.042729504f, -0.043227166f, 0.011683251f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 769 | -0.013082158f, -0.029302018f, -0.010899579f, -0.062036745f, |
| 770 | -0.022509435f, -0.00964907f, -0.01567329f, 0.04260106f, |
| 771 | -0.07787477f, -0.11576462f, 0.017356863f, 0.048673786f, // 04 |
| 772 | -0.017577527f, -0.05527947f, -0.082487635f, -0.040137455f, |
| 773 | -0.10820036f, -0.04666372f, 0.022746278f, -0.07851417f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 774 | 0.01068115f, 0.032956902f, 0.022433773f, 0.0026891115f, |
| 775 | 0.08944216f, -0.0685835f, 0.010513544f, 0.07228705f, // 05 |
| 776 | 0.02032331f, -0.059686817f, -0.0005566496f, -0.086984694f, |
| 777 | 0.040414046f, -0.1380399f, 0.094208956f, -0.05722982f, |
| 778 | 0.012092817f, -0.04989123f, -0.086576f, -0.003399834f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 779 | -0.04696032f, -0.045747425f, 0.10091314f, 0.048676282f, // 06 |
| 780 | -0.029037097f, 0.031399418f, -0.0040285117f, 0.047237843f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 781 | 0.09504992f, 0.041799378f, -0.049185462f, -0.031518843f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 782 | -0.10516937f, 0.026374253f, 0.10058866f, -0.0033195973f, |
| 783 | -0.041975245f, 0.0073591834f, 0.0033782164f, -0.004325073f, // 07 |
| 784 | -0.10167381f, 0.042500053f, -0.01447153f, 0.06464186f, |
| 785 | -0.017142897f, 0.03312627f, 0.009205989f, 0.024138335f, |
| 786 | -0.011337001f, 0.035530265f, -0.010912711f, 0.0706555f, |
| 787 | -0.005894094f, 0.051841937f, -0.1401738f, -0.02351249f, // 08 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 788 | 0.0365468f, 0.07590991f, 0.08838724f, 0.021681072f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 789 | -0.10086113f, 0.019608743f, -0.06195883f, 0.077335775f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 790 | 0.023646897f, -0.095322326f, 0.02233014f, 0.09756986f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 791 | -0.048691444f, -0.009579111f, 0.07595467f, 0.11480546f, // 09 |
| 792 | -0.09801813f, 0.019894179f, 0.08502348f, 0.004032281f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 793 | 0.037211012f, 0.068537936f, -0.048005626f, -0.091520436f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 794 | -0.028379958f, -0.01556313f, 0.06554592f, -0.045599163f, |
| 795 | -0.01672207f, -0.020169014f, -0.011877351f, -0.20212261f, // 10 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 796 | 0.010889619f, 0.0047078193f, 0.038385306f, 0.08540671f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 797 | -0.017140968f, -0.0035865551f, 0.016678626f, 0.005633034f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 798 | 0.015963363f, 0.00871737f, 0.060130805f, 0.028611384f, |
| 799 | 0.10109069f, -0.015060172f, -0.07894427f, 0.06401885f, // 11 |
| 800 | 0.011584063f, -0.024466386f, 0.0047652307f, -0.09041358f, |
| 801 | 0.030737216f, -0.0046374933f, 0.14215417f, -0.11823516f, |
| 802 | 0.019899689f, 0.006106124f, -0.027092824f, 0.0786356f, |
| 803 | 0.05052217f, -0.058925f, -0.011402121f, -0.024987547f, // 12 |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 804 | -0.0013661642f, -0.06832946f, -0.015667673f, -0.1083353f, |
| 805 | -0.00096863037f, -0.06988685f, -0.053350925f, -0.027275559f, |
| 806 | -0.033664223f, -0.07978348f, -0.025200296f, -0.017207067f, |
| 807 | -0.058403496f, -0.055697463f, 0.005798788f, 0.12965427f, // 13 |
| 808 | -0.062582195f, 0.0013350133f, -0.10482091f, 0.0379771f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 809 | 0.072521195f, -0.0029455067f, -0.13797039f, -0.03628521f, |
| 810 | 0.013806405f, -0.017858358f, -0.01008298f, -0.07700066f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 811 | -0.017081132f, 0.019358726f, 0.0027079724f, 0.004635139f, // 14 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 812 | 0.062634714f, -0.02338735f, -0.039547626f, -0.02050681f, |
| 813 | 0.03385117f, -0.083611414f, 0.002862572f, -0.09421313f, |
| 814 | 0.058618143f, -0.08598433f, 0.00972939f, 0.023867095f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 815 | -0.053934585f, -0.023203006f, 0.07452513f, -0.048767887f, // 15 |
| 816 | -0.07314807f, -0.056307215f, -0.10433547f, -0.06440842f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 817 | 0.04328182f, 0.04389765f, -0.020006588f, -0.09076438f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 818 | -0.11652589f, -0.021705797f, 0.03345259f, -0.010329105f, |
| 819 | -0.025767034f, 0.013057034f, -0.07316461f, -0.10145612f, // 16 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 820 | 0.06358255f, 0.18531723f, 0.07759293f, 0.12006465f, |
| 821 | 0.1305557f, 0.058638252f, -0.03393652f, 0.09622831f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 822 | -0.16253184f, -2.4580743e-06f, 0.079869635f, -0.070196845f, |
| 823 | -0.005644518f, 0.06857898f, -0.12598175f, -0.035084512f, // 17 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 824 | 0.03156317f, -0.12794146f, -0.031963028f, 0.04692781f, |
| 825 | 0.030070418f, 0.0071660685f, -0.095516115f, -0.004643372f, |
| 826 | 0.040170413f, -0.062104587f, -0.0037324072f, 0.0554317f, |
| 827 | 0.08184801f, -0.019164372f, 0.06791302f, 0.034257166f, // 18 |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 828 | -0.10307039f, 0.021943003f, 0.046745934f, 0.0790918f, |
| 829 | -0.0265588f, -0.007824208f, 0.042546265f, -0.00977924f, |
| 830 | -0.0002440307f, -0.017384544f, -0.017990116f, 0.12252321f, |
| 831 | -0.014512694f, -0.08251313f, 0.08861942f, 0.13589665f, // 19 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 832 | 0.026351685f, 0.012641483f, 0.07466548f, 0.044301085f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 833 | -0.045414884f, -0.051112458f, 0.03444247f, -0.08502782f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 834 | -0.04106223f, -0.028126027f, 0.028473156f, 0.10467447f |
| 835 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 836 | // 06: The recurrent-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 837 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 838 | hidl_vec<uint32_t> recurrentToForgetWeightsDimensions{numUnits, outputSize}; |
| 839 | std::vector<float> recurrentToForgetWeightsValue |
| 840 | { |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 841 | -0.057784554f, -0.026057621f, -0.068447545f, -0.022581743f, // 00 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 842 | 0.14811787f, 0.10826372f, 0.09471067f, 0.03987225f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 843 | -0.0039523416f, 0.00030638507f, 0.053185795f, 0.10572994f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 844 | 0.08414449f, -0.022036452f, -0.00066928595f, -0.09203576f, |
| 845 | 0.032950465f, -0.10985798f, -0.023809856f, 0.0021431844f, // 01 |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 846 | -0.02196096f, -0.00326074f, 0.00058621005f, -0.074678116f, |
| 847 | -0.06193199f, 0.055729095f, 0.03736828f, 0.020123724f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 848 | 0.061878487f, -0.04729229f, 0.034919553f, -0.07585433f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 849 | -0.04421272f, -0.044019096f, 0.085488975f, 0.04058006f, // 02 |
| 850 | -0.06890133f, -0.030951202f, -0.024628663f, -0.07672815f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 851 | 0.034293607f, 0.08556707f, -0.05293577f, -0.033561368f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 852 | -0.04899627f, 0.0241671f, 0.015736353f, -0.095442444f, |
| 853 | -0.029564252f, 0.016493602f, -0.035026584f, 0.022337519f, // 03 |
| 854 | -0.026871363f, 0.004780428f, 0.0077918363f, -0.03601621f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 855 | 0.016435321f, -0.03263031f, -0.09543275f, -0.047392778f, |
| 856 | 0.013454138f, 0.028934088f, 0.01685226f, -0.086110644f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 857 | -0.046250615f, -0.01847454f, 0.047608484f, 0.07339695f, // 04 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 858 | 0.034546845f, -0.04881143f, 0.009128804f, -0.08802852f, |
| 859 | 0.03761666f, 0.008096139f, -0.014454086f, 0.014361001f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 860 | -0.023502491f, -0.0011840804f, -0.07607001f, 0.001856849f, |
| 861 | -0.06509276f, -0.006021153f, -0.08570962f, -0.1451793f, // 05 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 862 | 0.060212336f, 0.055259194f, 0.06974018f, 0.049454916f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 863 | -0.027794661f, -0.08077226f, -0.016179763f, 0.1169753f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 864 | 0.17213494f, -0.0056326236f, -0.053934924f, -0.0124349f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 865 | -0.11520337f, 0.05409887f, 0.088759385f, 0.0019655675f, // 06 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 866 | 0.0042065294f, 0.03881498f, 0.019844765f, 0.041858196f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 867 | -0.05695512f, 0.047233116f, 0.038937137f, -0.06542224f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 868 | 0.014429736f, -0.09719407f, 0.13908425f, -0.05379757f, |
| 869 | 0.012321099f, 0.082840554f, -0.029899208f, 0.044217527f, // 07 |
| 870 | 0.059855383f, 0.07711018f, -0.045319796f, 0.0948846f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 871 | -0.011724666f, -0.0033288454f, -0.033542685f, -0.04764985f, |
| 872 | -0.13873616f, 0.040668588f, 0.034832682f, -0.015319203f, |
| 873 | -0.018715994f, 0.046002675f, 0.0599172f, -0.043107376f, // 08 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 874 | 0.0294216f, -0.002314414f, -0.022424703f, 0.0030315618f, |
| 875 | 0.0014641669f, 0.0029166266f, -0.11878115f, 0.013738511f, |
| 876 | 0.12375372f, -0.0006038222f, 0.029104086f, 0.087442465f, |
| 877 | 0.052958444f, 0.07558703f, 0.04817258f, 0.044462286f, // 09 |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 878 | -0.015213451f, -0.08783778f, -0.0561384f, -0.003008196f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 879 | 0.047060397f, -0.002058388f, 0.03429439f, -0.018839769f, |
| 880 | 0.024734668f, 0.024614193f, -0.042046934f, 0.09597743f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 881 | -0.0043254104f, 0.04320769f, 0.0064070094f, -0.0019131786f, // 10 |
| 882 | -0.02558259f, -0.022822596f, -0.023273505f, -0.02464396f, |
| 883 | -0.10991725f, -0.006240552f, 0.0074488563f, 0.024044557f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 884 | 0.04383914f, -0.046476185f, 0.028658995f, 0.060410924f, |
| 885 | 0.050786525f, 0.009452605f, -0.0073054377f, -0.024810238f, // 11 |
| 886 | 0.0052906186f, 0.0066939713f, -0.0020913032f, 0.014515517f, |
| 887 | 0.015898481f, 0.021362653f, -0.030262267f, 0.016587038f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 888 | -0.011442813f, 0.041154444f, -0.007631438f, -0.03423484f, |
| 889 | -0.010977775f, 0.036152758f, 0.0066366293f, 0.11915515f, // 12 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 890 | 0.02318443f, -0.041350313f, 0.021485701f, -0.10906167f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 891 | -0.028218046f, -0.00954771f, 0.020531068f, -0.11995105f, |
| 892 | -0.03672871f, 0.024019798f, 0.014255957f, -0.05221243f, |
| 893 | -0.00661567f, -0.04630967f, 0.033188973f, 0.10107534f, // 13 |
| 894 | -0.014027541f, 0.030796422f, -0.10270911f, -0.035999842f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 895 | 0.15443139f, 0.07684145f, 0.036571592f, -0.035900835f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 896 | -0.0034699554f, 0.06209149f, 0.015920248f, -0.031122351f, |
| 897 | -0.03858649f, 0.01849943f, 0.13872518f, 0.01503974f, // 14 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 898 | 0.069941424f, -0.06948533f, -0.0088794185f, 0.061282158f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 899 | -0.047401894f, 0.03100163f, -0.041533746f, -0.10430945f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 900 | 0.044574402f, -0.01425562f, -0.024290353f, 0.034563623f, |
| 901 | 0.05866852f, 0.023947537f, -0.09445152f, 0.035450947f, // 15 |
| 902 | 0.02247216f, -0.0042998926f, 0.061146557f, -0.10250651f, |
| 903 | 0.020881841f, -0.06747029f, 0.10062043f, -0.0023941975f, |
| 904 | 0.03532124f, -0.016341697f, 0.09685456f, -0.016764693f, |
| 905 | 0.051808182f, 0.05875331f, -0.04536488f, 0.001626336f, // 16 |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 906 | -0.028892258f, -0.01048663f, -0.009793449f, -0.017093895f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 907 | 0.010987891f, 0.02357273f, -0.00010856845f, 0.0099760275f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 908 | -0.001845119f, -0.03551521f, 0.0018358806f, 0.05763657f, |
| 909 | -0.01769146f, 0.040995963f, 0.02235177f, -0.060430344f, // 17 |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 910 | 0.11475477f, -0.023854522f, 0.10071741f, 0.0686208f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 911 | -0.014250481f, 0.034261297f, 0.047418304f, 0.08562733f, |
| 912 | -0.030519066f, 0.0060542435f, 0.014653856f, -0.038836084f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 913 | 0.04096551f, 0.032249358f, -0.08355519f, -0.026823482f, // 18 |
| 914 | 0.056386515f, -0.010401743f, -0.028396193f, 0.08507674f, |
| 915 | 0.014410365f, 0.020995233f, 0.17040324f, 0.11511526f, |
| 916 | 0.02459721f, 0.0066619175f, 0.025853224f, -0.023133837f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 917 | -0.081302024f, 0.017264642f, -0.009585969f, 0.09491168f, // 19 |
| 918 | -0.051313367f, 0.054532815f, -0.014298593f, 0.10657464f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 919 | 0.007076659f, 0.10964551f, 0.0409152f, 0.008275321f, |
| 920 | -0.07283536f, 0.07937492f, 0.04192024f, -0.1075027f |
| 921 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 922 | // 07: The recurrent-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 923 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 924 | hidl_vec<uint32_t> recurrentToCellWeightsDimensions{numUnits, outputSize}; |
| 925 | std::vector<float> recurrentToCellWeightsValue |
| 926 | { |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 927 | -0.037322544f, 0.018592842f, 0.0056175636f, -0.06253426f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 928 | 0.055647098f, -0.05713207f, -0.05626563f, 0.005559383f, |
| 929 | 0.03375411f, -0.025757805f, -0.088049285f, 0.06017052f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 930 | -0.06570978f, 0.007384076f, 0.035123326f, -0.07920549f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 931 | 0.053676967f, 0.044480428f, -0.07663568f, 0.0071805613f, |
| 932 | 0.08089997f, 0.05143358f, 0.038261272f, 0.03339287f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 933 | -0.027673481f, 0.044746667f, 0.028349208f, 0.020090483f, |
| 934 | -0.019443132f, -0.030755889f, -0.0040000007f, 0.04465846f, |
| 935 | -0.021585021f, 0.0031670958f, 0.0053199246f, -0.056117613f, |
| 936 | -0.10893326f, 0.076739706f, -0.08509834f, -0.027997585f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 937 | 0.037871376f, 0.01449768f, -0.09002357f, -0.06111149f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 938 | -0.046195522f, 0.0422062f, -0.005683705f, -0.1253618f, |
| 939 | -0.012925729f, -0.04890792f, 0.06985068f, 0.037654128f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 940 | 0.03398274f, -0.004781977f, 0.007032333f, -0.031787455f, |
| 941 | 0.010868644f, -0.031489216f, 0.09525667f, 0.013939797f, |
| 942 | 0.0058680447f, 0.0167067f, 0.02668468f, -0.04797466f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 943 | -0.048885044f, -0.12722108f, 0.035304096f, 0.06554885f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 944 | 0.00972396f, -0.039238118f, -0.05159735f, -0.11329045f, |
| 945 | 0.1613692f, -0.03750952f, 0.06529313f, -0.071974665f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 946 | -0.11769596f, 0.015524369f, -0.0013754242f, -0.12446318f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 947 | 0.02786344f, -0.014179351f, 0.005264273f, 0.14376344f, |
| 948 | 0.015983658f, 0.03406988f, -0.06939408f, 0.040699873f, |
| 949 | 0.02111075f, 0.09669095f, 0.041345075f, -0.08316494f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 950 | -0.07684199f, -0.045768797f, 0.032298047f, -0.041805092f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 951 | 0.0119405f, 0.0061010392f, 0.12652606f, 0.0064572375f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 952 | -0.024950314f, 0.11574242f, 0.04508852f, -0.04335324f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 953 | 0.06760663f, -0.027437469f, 0.07216407f, 0.06977076f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 954 | -0.05438599f, 0.034033038f, -0.028602652f, 0.05346137f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 955 | 0.043184172f, -0.037189785f, 0.10420091f, 0.00882477f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 956 | -0.054019816f, -0.074273005f, -0.030617684f, -0.0028467078f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 957 | 0.024302477f, -0.0038869337f, 0.005332455f, 0.0013399826f, |
| 958 | 0.04361412f, -0.007001822f, 0.09631092f, -0.06702025f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 959 | -0.042049985f, -0.035070654f, -0.04103342f, -0.10273396f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 960 | 0.0544271f, 0.037184782f, -0.13150354f, -0.0058036847f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 961 | -0.008264958f, 0.042035464f, 0.05891794f, 0.029673764f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 962 | 0.0063542654f, 0.044788733f, 0.054816857f, 0.062257513f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 963 | -0.00093483756f, 0.048938446f, -0.004952862f, -0.007730018f, |
| 964 | -0.04043371f, -0.017094059f, 0.07229206f, -0.023670016f, |
| 965 | -0.052195564f, -0.025616996f, -0.01520939f, 0.045104615f, |
| 966 | -0.007376126f, 0.003533447f, 0.006570588f, 0.056037236f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 967 | 0.12436656f, 0.051817212f, 0.028532185f, -0.08686856f, |
| 968 | 0.11868599f, 0.07663395f, -0.07323171f, 0.03463402f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 969 | -0.050708205f, -0.04458982f, -0.11590894f, 0.021273347f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 970 | 0.1251325f, -0.15313013f, -0.12224372f, 0.17228661f, |
| 971 | 0.023029093f, 0.086124025f, 0.006445803f, -0.03496501f, |
| 972 | 0.028332196f, 0.04449512f, -0.042436164f, -0.026587414f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 973 | -0.006041347f, -0.09292539f, -0.05678812f, 0.03897832f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 974 | 0.09465633f, 0.008115513f, -0.02171956f, 0.08304309f, |
| 975 | 0.071401566f, 0.019622514f, 0.032163795f, -0.004167056f, |
| 976 | 0.02295182f, 0.030739572f, 0.056506045f, 0.004612461f, |
| 977 | 0.06524936f, 0.059999723f, 0.046395954f, -0.0045512207f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 978 | -0.1335546f, -0.030136576f, 0.11584653f, -0.014678886f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 979 | 0.0020118146f, -0.09688814f, -0.0790206f, 0.039770417f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 980 | -0.0329582f, 0.07922767f, 0.029322514f, 0.026405897f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 981 | 0.04207835f, -0.07073373f, 0.063781224f, 0.0859677f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 982 | -0.10925287f, -0.07011058f, 0.048005477f, 0.03438226f, |
| 983 | -0.09606514f, -0.006669445f, -0.043381985f, 0.04240257f, |
| 984 | -0.06955775f, -0.06769346f, 0.043903265f, -0.026784198f, |
| 985 | -0.017840602f, 0.024307009f, -0.040079936f, -0.019946516f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 986 | 0.045318738f, -0.12233574f, 0.026170589f, 0.0074471775f, |
| 987 | 0.15978073f, 0.10185836f, 0.10298046f, -0.015476589f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 988 | -0.039390966f, -0.072174534f, 0.0739445f, -0.1211869f, |
| 989 | -0.0347889f, -0.07943156f, 0.014809798f, -0.12412325f, |
| 990 | -0.0030663363f, 0.039695457f, 0.0647603f, -0.08291318f, |
| 991 | -0.018529687f, -0.004423833f, 0.0037507233f, 0.084633216f, |
| 992 | -0.01514876f, -0.056505352f, -0.012800942f, -0.06994386f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 993 | 0.012962922f, -0.031234352f, 0.07029052f, 0.016418684f, |
| 994 | 0.03618972f, 0.055686004f, -0.08663945f, -0.017404709f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 995 | -0.054761406f, 0.029065743f, 0.052404847f, 0.020238016f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 996 | 0.0048197987f, -0.0214882f, 0.07078733f, 0.013016777f, |
| 997 | 0.06262858f, 0.009184685f, 0.020785125f, -0.043904778f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 998 | -0.0270329f, -0.03299152f, -0.060088247f, -0.015162964f, |
| 999 | -0.001828936f, 0.12642565f, -0.056757294f, 0.013586685f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1000 | 0.09232601f, -0.035886683f, 0.06000002f, 0.05229691f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1001 | -0.052580316f, -0.082029596f, -0.010794592f, 0.012947712f, |
| 1002 | -0.036429964f, -0.085508935f, -0.13127148f, -0.017744139f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1003 | 0.031502828f, 0.036232427f, -0.031581745f, 0.023051167f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1004 | -0.05325106f, -0.03421577f, 0.028793324f, -0.034633752f, |
| 1005 | -0.009881397f, -0.043551125f, -0.018609839f, 0.0019097115f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1006 | -0.008799762f, 0.056595087f, 0.0022273948f, 0.055752404f |
| 1007 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1008 | // 08: The recurrent-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1009 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1010 | hidl_vec<uint32_t> recurrentToOutputWeightsDimensions{numUnits, outputSize}; |
| 1011 | std::vector<float> recurrentToOutputWeightsValue |
| 1012 | { |
| 1013 | 0.025825322f, -0.05813119f, 0.09495884f, -0.045984812f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1014 | -0.01255415f, -0.0026479573f, -0.08196161f, -0.054914974f, |
| 1015 | -0.0046604523f, -0.029587349f, -0.044576716f, -0.07480124f, |
| 1016 | -0.082868785f, 0.023254942f, 0.027502948f, -0.0039728214f, |
| 1017 | -0.08683098f, -0.08116779f, -0.014675607f, -0.037924774f, |
| 1018 | -0.023314456f, -0.007401714f, -0.09255757f, 0.029460307f, |
| 1019 | -0.08829125f, -0.005139627f, -0.08989442f, -0.0555066f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1020 | 0.13596267f, -0.025062224f, -0.048351806f, -0.03850004f, |
| 1021 | 0.07266485f, -0.022414139f, 0.05940088f, 0.075114764f, |
| 1022 | 0.09597592f, -0.010211725f, -0.0049794707f, -0.011523867f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1023 | -0.025980417f, 0.072999895f, 0.11091378f, -0.081685916f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1024 | 0.014416728f, 0.043229222f, 0.034178585f, -0.07530371f, |
| 1025 | 0.035837382f, -0.085607f, -0.007721233f, -0.03287832f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1026 | -0.043848954f, -0.06404588f, -0.06632928f, -0.073643476f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1027 | 0.008214239f, -0.045984086f, 0.039764922f, 0.03474462f, |
| 1028 | 0.060612556f, -0.080590084f, 0.049127717f, 0.04151091f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1029 | -0.030063879f, 0.008801774f, -0.023021035f, -0.019558564f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1030 | 0.05158114f, -0.010947698f, -0.011825728f, 0.0075720972f, |
| 1031 | 0.0699727f, -0.0039981045f, 0.069350146f, 0.08799282f, |
| 1032 | 0.016156472f, 0.035502106f, 0.11695009f, 0.006217345f, |
| 1033 | 0.13392477f, -0.037875112f, 0.025745004f, 0.08940699f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1034 | -0.00924166f, 0.0046702605f, -0.036598757f, -0.08811812f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1035 | 0.10522024f, -0.032441203f, 0.008176899f, -0.04454919f, |
| 1036 | 0.07058152f, 0.0067963637f, 0.039206743f, 0.03259838f, |
| 1037 | 0.03725492f, -0.09515802f, 0.013326398f, -0.052055415f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1038 | -0.025676316f, 0.03198509f, -0.015951829f, -0.058556724f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1039 | 0.036879618f, 0.043357447f, 0.028362012f, -0.05908629f, |
| 1040 | 0.0059240665f, -0.04995891f, -0.019187413f, 0.0276265f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1041 | -0.01628143f, 0.0025863599f, 0.08800015f, 0.035250366f, |
| 1042 | -0.022165963f, -0.07328642f, -0.009415526f, -0.07455109f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1043 | 0.11690406f, 0.0363299f, 0.07411125f, 0.042103454f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1044 | -0.009660886f, 0.019076364f, 0.018299393f, -0.046004917f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1045 | 0.08891175f, 0.0431396f, -0.026327137f, -0.051502608f, |
| 1046 | 0.08979574f, -0.051670972f, 0.04940282f, -0.07491107f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1047 | -0.021240504f, 0.022596184f, -0.034280192f, 0.060163025f, |
| 1048 | -0.058211457f, -0.051837247f, -0.01349775f, -0.04639988f, |
| 1049 | -0.035936575f, -0.011681591f, 0.064818054f, 0.0073146066f, |
| 1050 | -0.021745546f, -0.043124277f, -0.06471268f, -0.07053354f, |
| 1051 | -0.029321948f, -0.05330136f, 0.016933719f, -0.053782392f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1052 | 0.13747959f, -0.1361751f, -0.11569455f, 0.0033329215f, |
| 1053 | 0.05693899f, -0.053219706f, 0.063698f, 0.07977434f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1054 | -0.07924483f, 0.06936997f, 0.0034815092f, -0.007305279f, |
| 1055 | -0.037325785f, -0.07251102f, -0.033633437f, -0.08677009f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1056 | 0.091591336f, -0.14165086f, 0.021752775f, 0.019683983f, |
| 1057 | 0.0011612234f, -0.058154266f, 0.049996935f, 0.0288841f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1058 | -0.0024567875f, -0.14345716f, 0.010955264f, -0.10234828f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1059 | 0.1183656f, -0.0010731248f, -0.023590032f, -0.072285876f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1060 | -0.0724771f, -0.026382286f, -0.0014920527f, 0.042667855f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1061 | 0.0018776858f, 0.02986552f, 0.009814309f, 0.0733756f, |
| 1062 | 0.12289186f, 0.018043943f, -0.0458958f, 0.049412545f, |
| 1063 | 0.033632483f, 0.05495232f, 0.036686596f, -0.013781798f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1064 | -0.010036754f, 0.02576849f, -0.08307328f, 0.010112348f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1065 | 0.042521734f, -0.05869831f, -0.071689695f, 0.03876447f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1066 | -0.13275425f, -0.0352966f, -0.023077697f, 0.10285965f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1067 | 0.084736146f, 0.15568255f, -0.00040734606f, 0.027835453f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1068 | -0.10292561f, -0.032401145f, 0.10053256f, -0.026142767f, |
| 1069 | -0.08271222f, -0.0030240538f, -0.016368777f, 0.1070414f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1070 | 0.042672627f, 0.013456989f, -0.0437609f, -0.022309763f, |
| 1071 | 0.11576483f, 0.04108048f, 0.061026827f, -0.0190714f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1072 | -0.0869359f, 0.037901703f, 0.0610107f, 0.07202949f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1073 | 0.01675338f, 0.086139716f, -0.08795751f, -0.014898893f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1074 | -0.023771819f, -0.01965048f, 0.007955471f, -0.043740474f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1075 | 0.03346837f, -0.10549954f, 0.090567775f, 0.042013682f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1076 | -0.03176985f, 0.12569028f, -0.02421228f, -0.029526481f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1077 | 0.023851605f, 0.031539805f, 0.05292009f, -0.02344001f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1078 | -0.07811758f, -0.08834428f, 0.10094801f, 0.16594367f, |
| 1079 | -0.06861939f, -0.021256343f, -0.041093912f, -0.06669611f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1080 | 0.035498552f, 0.021757556f, -0.09302526f, -0.015403468f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1081 | -0.06614931f, -0.051798206f, -0.013874718f, 0.03630673f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1082 | 0.010412845f, -0.08077351f, 0.046185967f, 0.0035662893f, |
| 1083 | 0.03541868f, -0.094149634f, -0.034814864f, 0.003128424f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1084 | -0.020674974f, -0.03944324f, -0.008110165f, -0.11113267f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1085 | 0.08484226f, 0.043586485f, 0.040582247f, 0.0968012f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1086 | -0.065249965f, -0.028036479f, 0.0050708856f, 0.0017462453f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1087 | 0.0326779f, 0.041296225f, 0.09164146f, -0.047743853f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1088 | -0.015952192f, -0.034451712f, 0.084197424f, -0.05347844f, |
| 1089 | -0.11768019f, 0.085926116f, -0.08251791f, -0.045081906f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1090 | 0.0948852f, 0.068401024f, 0.024856757f, 0.06978981f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1091 | -0.057309967f, -0.012775832f, -0.0032452994f, 0.01977615f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1092 | -0.041040014f, -0.024264973f, 0.063464895f, 0.05431621f |
| 1093 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1094 | // 09: The cell-to-input weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1095 | hidl_vec<uint32_t> cellToInputWeightsDimensions{numUnits}; |
| 1096 | std::vector<float> cellToInputWeightsValue |
| 1097 | { |
| 1098 | 0.040369894f, 0.030746894f, 0.24704495f, 0.018586371f, -0.037586458f, |
| 1099 | -0.15312155f, -0.11812848f, -0.11465643f, 0.20259799f, 0.11418174f, |
| 1100 | -0.10116027f, -0.011334949f, 0.12411352f, -0.076769054f, -0.052169047f, |
| 1101 | 0.21198851f, -0.38871562f, -0.09061183f, -0.09683246f, -0.21929175f |
| 1102 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1103 | // 10: The cell-to-forget weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1104 | hidl_vec<uint32_t> cellToForgetWeightsDimensions{numUnits}; |
| 1105 | std::vector<float> cellToForgetWeightsValue |
| 1106 | { |
| 1107 | -0.01998659f, -0.15568835f, -0.24248174f, -0.012770197f, 0.041331276f, |
| 1108 | -0.072311886f, -0.052123554f, -0.0066330447f, -0.043891653f, 0.036225766f, |
| 1109 | -0.047248036f, 0.021479502f, 0.033189066f, 0.11952997f, -0.020432774f, |
| 1110 | 0.64658105f, -0.06650122f, -0.03467612f, 0.095340036f, 0.23647355f |
| 1111 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1112 | // 11: The cell-to-output weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1113 | hidl_vec<uint32_t> cellToOutputWeightsDimensions{numUnits}; |
| 1114 | std::vector<float> cellToOutputWeightsValue |
| 1115 | { |
| 1116 | 0.08286371f, -0.08261836f, -0.51210177f, 0.002913762f, 0.17764764f, |
| 1117 | -0.5495371f, -0.08460716f, -0.24552552f, 0.030037103f, 0.04123544f, |
| 1118 | -0.11940523f, 0.007358328f, 0.1890978f, 0.4833202f, -0.34441817f, |
| 1119 | 0.36312827f, -0.26375428f, 0.1457655f, -0.19724406f, 0.15548733f |
| 1120 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1121 | // 12: The input gate bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1122 | hidl_vec<uint32_t> inputGateBiasDimensions{numUnits}; |
| 1123 | std::vector<float> inputGateBiasValue |
| 1124 | { |
| 1125 | 0.02234832f, 0.14757581f, 0.18176508f, 0.10380666f, 0.053110216f, |
| 1126 | -0.06928846f, -0.13942584f, -0.11816189f, 0.19483899f, 0.03652339f, |
| 1127 | -0.10250295f, 0.036714908f, -0.18426876f, 0.036065217f, 0.21810818f, |
| 1128 | 0.02383196f, -0.043370757f, 0.08690144f, -0.04444982f, 0.00030581196f |
| 1129 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1130 | // 13: The forget gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1131 | hidl_vec<uint32_t> forgetGateBiasDimensions{numUnits}; |
| 1132 | std::vector<float> forgetGateBiasValue |
| 1133 | { |
| 1134 | 0.035185695f, -0.042891346f, -0.03032477f, 0.23027696f, 0.11098921f, |
| 1135 | 0.15378423f, 0.09263801f, 0.09790885f, 0.09508917f, 0.061199076f, |
| 1136 | 0.07665568f, -0.015443159f, -0.03499149f, 0.046190713f, 0.08895977f, |
| 1137 | 0.10899629f, 0.40694186f, 0.06030037f, 0.012413437f, -0.06108739f |
| 1138 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1139 | // 14: The cell bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1140 | hidl_vec<uint32_t> cellBiasDimensions{numUnits}; |
| 1141 | std::vector<float> cellBiasValue |
| 1142 | { |
| 1143 | -0.024379363f, 0.0055531194f, 0.23377132f, 0.033463873f, -0.1483596f, |
| 1144 | -0.10639995f, -0.091433935f, 0.058573797f, -0.06809782f, -0.07889636f, |
| 1145 | -0.043246906f, -0.09829136f, -0.4279842f, 0.034901652f, 0.18797937f, |
| 1146 | 0.0075234566f, 0.016178843f, 0.1749513f, 0.13975595f, 0.92058027f |
| 1147 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1148 | // 15: The output gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1149 | hidl_vec<uint32_t> outputGateBiasDimensions{numUnits}; |
| 1150 | std::vector<float> outputGateBiasValue |
| 1151 | { |
| 1152 | 0.046159424f, -0.0012809046f, 0.03563469f, 0.12648113f, 0.027195795f, |
| 1153 | 0.35373217f, -0.018957434f, 0.008907322f, -0.0762701f, 0.12018895f, |
| 1154 | 0.04216877f, 0.0022856654f, 0.040952638f, 0.3147856f, 0.08225149f, |
| 1155 | -0.057416286f, -0.14995944f, -0.008040261f, 0.13208859f, 0.029760877f |
| 1156 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1157 | // 16: The projection weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1158 | // [output_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1159 | hidl_vec<uint32_t> projectionWeightsDimensions{outputSize, numUnits}; |
| 1160 | std::vector<float> projectionWeightsValue |
| 1161 | { |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1162 | -0.009802181f, 0.09401916f, 0.0717386f, -0.13895074f, 0.09641832f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1163 | 0.060420845f, 0.08539281f, 0.054285463f, 0.061395317f, 0.034448683f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1164 | -0.042991187f, 0.019801661f, -0.16840284f, -0.015726732f, -0.23041931f, |
| 1165 | -0.024478018f, -0.10959692f, -0.013875541f, 0.18600968f, -0.061274476f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1166 | 0.0138165f, -0.08160894f, -0.07661644f, 0.032372914f, 0.16169067f, |
| 1167 | 0.22465782f, -0.03993472f, -0.004017731f, 0.08633481f, -0.28869787f, |
| 1168 | 0.08682067f, 0.17240396f, 0.014975425f, 0.056431185f, 0.031037588f, |
| 1169 | 0.16702051f, 0.0077946745f, 0.15140012f, 0.29405436f, 0.120285f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1170 | -0.188994f, -0.027265169f, 0.043389652f, -0.022061434f, 0.014777949f, |
| 1171 | -0.20203483f, 0.094781205f, 0.19100232f, 0.13987629f, -0.036132768f, |
| 1172 | -0.06426278f, -0.05108664f, 0.13221376f, 0.009441198f, -0.16715929f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1173 | 0.15859416f, -0.040437475f, 0.050779544f, -0.022187516f, 0.012166504f, |
| 1174 | 0.027685808f, -0.07675938f, -0.0055694645f, -0.09444123f, 0.0046453946f, |
| 1175 | 0.050794356f, 0.10770313f, -0.20790008f, -0.07149004f, -0.11425117f, |
| 1176 | 0.008225835f, -0.035802525f, 0.14374903f, 0.15262283f, 0.048710253f, |
| 1177 | 0.1847461f, -0.007487823f, 0.11000021f, -0.09542012f, 0.22619456f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1178 | -0.029149994f, 0.08527916f, 0.009043713f, 0.0042746216f, 0.016261552f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1179 | 0.022461696f, 0.12689082f, -0.043589946f, -0.12035478f, -0.08361797f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1180 | -0.050666027f, -0.1248618f, -0.1275799f, -0.071875185f, 0.07377272f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1181 | 0.09944291f, -0.18897448f, -0.1593054f, -0.06526116f, -0.040107165f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1182 | -0.004618631f, -0.067624845f, -0.007576253f, 0.10727444f, 0.041546922f, |
| 1183 | -0.20424393f, 0.06907816f, 0.050412357f, 0.00724631f, 0.039827548f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1184 | 0.12449835f, 0.10747581f, 0.13708383f, 0.09134148f, -0.12617786f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1185 | -0.06428341f, 0.09956831f, 0.1208086f, -0.14676677f, -0.0727722f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1186 | 0.1126304f, 0.010139365f, 0.015571211f, -0.038128063f, 0.022913318f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1187 | -0.042050496f, 0.16842307f, -0.060597885f, 0.10531834f, -0.06411776f, |
| 1188 | -0.07451711f, -0.03410368f, -0.13393489f, 0.06534304f, 0.003620307f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1189 | 0.04490757f, 0.05970546f, 0.05197996f, 0.02839995f, 0.10434969f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1190 | -0.013699693f, -0.028353551f, -0.07260381f, 0.047201227f, -0.024575593f, |
| 1191 | -0.036445823f, 0.07155557f, 0.009672501f, -0.02328883f, 0.009533515f, |
| 1192 | -0.03606021f, -0.07421458f, -0.028082801f, -0.2678904f, -0.13221288f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1193 | 0.18419984f, -0.13012612f, -0.014588381f, -0.035059117f, -0.04824723f, |
| 1194 | 0.07830115f, -0.056184657f, 0.03277091f, 0.025466874f, 0.14494097f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1195 | -0.12522776f, -0.098633975f, -0.10766018f, -0.08317623f, 0.08594209f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1196 | 0.07749552f, 0.039474737f, 0.1776665f, -0.07409566f, -0.0477268f, |
| 1197 | 0.29323658f, 0.10801441f, 0.1154011f, 0.013952499f, 0.10739139f, |
| 1198 | 0.10708251f, -0.051456142f, 0.0074137426f, -0.10430189f, 0.10034707f, |
| 1199 | 0.045594677f, 0.0635285f, -0.0715442f, -0.089667566f, -0.10811871f, |
| 1200 | 0.00026344223f, 0.08298446f, -0.009525053f, 0.006585689f, -0.24567553f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1201 | -0.09450807f, 0.09648481f, 0.026996298f, -0.06419476f, -0.04752702f, |
| 1202 | -0.11063944f, -0.23441927f, -0.17608605f, -0.052156363f, 0.067035615f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1203 | 0.19271925f, -0.0032889997f, -0.043264326f, 0.09663576f, -0.057112187f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1204 | -0.10100678f, 0.0628376f, 0.04447668f, 0.017961001f, -0.10094388f, |
| 1205 | -0.10190601f, 0.18335468f, 0.10494553f, -0.052095775f, -0.0026118709f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1206 | 0.10539724f, -0.04383912f, -0.042349473f, 0.08438151f, -0.1947263f, |
| 1207 | 0.02251204f, 0.11216432f, -0.10307853f, 0.17351969f, -0.039091777f, |
| 1208 | 0.08066188f, -0.00561982f, 0.12633002f, 0.11335965f, -0.0088127935f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1209 | -0.019777594f, 0.06864014f, -0.059751723f, 0.016233567f, -0.06894641f, |
| 1210 | -0.28651384f, -0.004228674f, 0.019708522f, -0.16305895f, -0.07468996f, |
| 1211 | -0.0855457f, 0.099339016f, -0.07580735f, -0.13775392f, 0.08434318f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1212 | 0.08330512f, -0.12131499f, 0.031935584f, 0.09180414f, -0.08876437f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1213 | -0.08049874f, 0.008753825f, 0.03498998f, 0.030215185f, 0.03907079f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1214 | 0.089751154f, 0.029194152f, -0.03337423f, -0.019092513f, 0.04331237f, |
| 1215 | 0.04299654f, -0.036394123f, -0.12915532f, 0.09793732f, 0.07512415f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1216 | -0.11319543f, -0.032502122f, 0.15661901f, 0.07671967f, -0.005491124f, |
| 1217 | -0.19379048f, -0.218606f, 0.21448623f, 0.017840758f, 0.1416943f, |
| 1218 | -0.07051762f, 0.19488361f, 0.02664691f, -0.18104725f, -0.09334311f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1219 | 0.15026465f, -0.15493552f, -0.057762887f, -0.11604192f, -0.262013f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1220 | -0.01391798f, 0.012185008f, 0.11156489f, -0.07483202f, 0.06693364f, |
| 1221 | -0.26151478f, 0.046425626f, 0.036540434f, -0.16435726f, 0.17338543f, |
| 1222 | -0.21401681f, -0.11385144f, -0.08283257f, -0.069031075f, 0.030635102f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1223 | 0.010969227f, 0.11109743f, 0.010919218f, 0.027526086f, 0.13519906f, |
| 1224 | 0.01891392f, -0.046839405f, -0.040167913f, 0.017953383f, -0.09700955f, |
| 1225 | 0.0061885654f, -0.07000971f, 0.026893595f, -0.038844477f, 0.14543656f |
| 1226 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1227 | // 17: The projection bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1228 | hidl_vec<uint32_t> projectionBiasDimensions{outputSize}; |
| 1229 | std::vector<float> projectionBiasValue(outputSize, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1230 | |
| 1231 | // 18: The output state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1232 | hidl_vec<uint32_t> outputStateInDimensions{batchSize, outputSize}; |
| 1233 | std::vector<float> outputStateInValue(batchSize * outputSize, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1234 | // 19: The cell state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1235 | hidl_vec<uint32_t> cellStateInDimensions{batchSize, numUnits}; |
| 1236 | std::vector<float> cellStateInValue(batchSize * numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1237 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1238 | // Constant scalar values (the VTS test adds these as tensors of dim {}) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1239 | // 20: The activation function: A value indicating the activation function: |
| 1240 | // 0: None; 1: Relu; 3: Relu6; 4: Tanh; 6: Sigmoid. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1241 | hidl_vec<uint32_t> activationFunctionDimensions{}; |
| 1242 | std::vector<int32_t> activationFunctionValue{4}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1243 | // 21: The clipping threshold: for the cell state, such that values are bound within [-cell_clip, cell_clip]. |
| 1244 | // If set to 0.0 then clipping is disabled. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1245 | hidl_vec<uint32_t> cellClippingThresholdDimensions{}; |
| 1246 | std::vector<float> cellClippingThresholdValue{0.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1247 | // 22: The clipping threshold: for the output from the projection layer, such that values are bound within |
| 1248 | // [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1249 | hidl_vec<uint32_t> projectionClippingThresholdDimensions{}; |
| 1250 | std::vector<float> projectionClippingThresholdValue{0.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1251 | |
| 1252 | // Outputs: |
| 1253 | // 0: The scratch buffer: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units * 4] with |
| 1254 | // CIFG, or [batch_size, num_units * 3] without CIFG. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1255 | // HOWEVER, by looking at the code, seems that it's the opposite: (cifg ? 3 : 4) * numUnits |
| 1256 | // Refer to: android/frameworks/ml/nn/common/operations/LSTM.cpp:319 |
| 1257 | // android/frameworks/ml/nn/common/operations/LSTMTest.cpp:114 |
| 1258 | // tensorflow/tensorflow/contrib/lite/kernels/lstm.cc:332 |
| 1259 | hidl_vec<uint32_t> scratchBufferDimensions{batchSize, numUnits * 4}; |
| 1260 | std::vector<float> scratchBufferValue(batchSize * numUnits * 4, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1261 | // 1: The output state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1262 | hidl_vec<uint32_t> outputStateOutDimensions{batchSize, outputSize}; |
| 1263 | std::vector<float> outputStateOutValue |
| 1264 | { |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1265 | -0.00396806f, 0.029352f, -0.00279226f, 0.0159977f, -0.00835577f, -0.0211779f, 0.0283512f, -0.0114597f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1266 | 0.00907307f, -0.0244004f, -0.0152191f, -0.0259063f, 0.00914318f, 0.00415119f, 0.017147f, 0.0134203f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1267 | -0.013869f, 0.0287268f, -0.00334694f, 0.00733397f, -0.0287926f, -0.0186926f, 0.0193662f, -0.0115437f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1268 | 0.00422612f, -0.0345232f, 0.00223253f, -0.00957321f, 0.0210624f, 0.013331f, 0.0150954f, 0.0216801f |
| 1269 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1270 | // 2: The cell state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1271 | hidl_vec<uint32_t> cellStateOutDimensions{batchSize, numUnits}; |
| 1272 | std::vector<float> cellStateOutValue |
| 1273 | { |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1274 | -0.0531632f, -0.0118138f, 0.0870833f, 0.0347929f, -0.076144f, |
| 1275 | -0.0659219f, -0.0463811f, 0.0141307f, -0.0127706f, -0.03782f, |
| 1276 | -0.00402401f, -0.00571876f, -0.187957f, -0.0247127f, 0.0711425f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1277 | 0.008244f, 0.0492649f, 0.126972f, 0.0933097f, 0.29848f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1278 | -0.0966178f, -0.114417f, 0.0387229f, 0.0453255f, -0.181286f, |
| 1279 | -0.0651251f, -0.0996879f, -0.00276995f, 0.0617558f, -0.0100728f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1280 | 0.056304f, -0.077416f, -0.162858f, -0.0541251f, 0.0571202f, |
| 1281 | -0.0525331f, 0.0724297f, 0.171029f, 0.141738f, 0.295483f |
| 1282 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1283 | // 3: The output: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. This is |
| 1284 | // effectively the same as the current “output state (out)” value. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1285 | hidl_vec<uint32_t> outputDimensions{batchSize, outputSize}; |
| 1286 | std::vector<float> outputValue |
| 1287 | { |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1288 | -0.00396806f, 0.029352f, -0.00279226f, 0.0159977f, -0.00835576f, -0.0211779f, 0.0283512f, -0.0114597f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1289 | 0.00907307f, -0.0244004f, -0.0152191f, -0.0259063f, 0.00914318f, 0.00415118f, 0.017147f, 0.0134203f, |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1290 | -0.013869f, 0.0287268f, -0.00334693f, 0.00733398f, -0.0287926f, -0.0186926f, 0.0193662f, -0.0115437f, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1291 | 0.00422612f, -0.0345232f, 0.00223253f, -0.00957321f, 0.0210624f, 0.013331f, 0.0150954f, 0.02168f |
| 1292 | }; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1293 | |
| 1294 | LstmTestImpl(inputDimensions, inputValue, |
| 1295 | inputToInputWeightsDimensions, inputToInputWeightsValue, |
| 1296 | inputToForgetWeightsDimensions, inputToForgetWeightsValue, |
| 1297 | inputToCellWeightsDimensions, inputToCellWeightsValue, |
| 1298 | inputToOutputWeightsDimensions, inputToOutputWeightsValue, |
| 1299 | recurrentToInputWeightsDimensions, recurrentToInputWeightsValue, |
| 1300 | recurrentToForgetWeightsDimensions, recurrentToForgetWeightsValue, |
| 1301 | recurrentToCellWeightsDimensions, recurrentToCellWeightsValue, |
| 1302 | recurrentToOutputWeightsDimensions, recurrentToOutputWeightsValue, |
| 1303 | cellToInputWeightsDimensions, cellToInputWeightsValue, |
| 1304 | cellToForgetWeightsDimensions, cellToForgetWeightsValue, |
| 1305 | cellToOutputWeightsDimensions, cellToOutputWeightsValue, |
| 1306 | inputGateBiasDimensions, inputGateBiasValue, |
| 1307 | forgetGateBiasDimensions, forgetGateBiasValue, |
| 1308 | cellBiasDimensions, cellBiasValue, |
| 1309 | outputGateBiasDimensions, outputGateBiasValue, |
| 1310 | projectionWeightsDimensions, projectionWeightsValue, |
| 1311 | projectionBiasDimensions, projectionBiasValue, |
| 1312 | outputStateInDimensions, outputStateInValue, |
| 1313 | cellStateInDimensions, cellStateInValue, |
| 1314 | activationFunctionDimensions, activationFunctionValue, |
| 1315 | cellClippingThresholdDimensions, cellClippingThresholdValue, |
| 1316 | projectionClippingThresholdDimensions, projectionClippingThresholdValue, |
| 1317 | scratchBufferDimensions, scratchBufferValue, |
| 1318 | outputStateOutDimensions, outputStateOutValue, |
| 1319 | cellStateOutDimensions, cellStateOutValue, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1320 | outputDimensions, outputValue, |
| 1321 | compute); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1322 | } |
| 1323 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1324 | void LstmCifgPeepholeNoProjectionBatch2(armnn::Compute compute) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1325 | { |
| 1326 | // This replicates android/frameworks/ml/nn/runtime/test/generated/vts_models/lstm2.model.cpp |
| 1327 | // with values from android/frameworks/ml/nn/runtime/test/generated/examples/lstm2.example.cpp |
| 1328 | // and weights, biases and scalars passed as CONSTANT_COPY tensors (instead of MODEL_INPUT tensors). |
| 1329 | // The batch size has been increased to 2 (it was 1 in the VTS test) with appropriate input and output values added. |
| 1330 | |
| 1331 | uint32_t batchSize = 2; |
| 1332 | uint32_t inputSize = 2; |
| 1333 | uint32_t numUnits = 4; |
| 1334 | uint32_t outputSize = numUnits; |
| 1335 | |
| 1336 | // Inputs: |
| 1337 | // 00: The input: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, input_size], where |
| 1338 | // “batch_size” corresponds to the batching dimension, and “input_size” is the size of the input. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1339 | hidl_vec<uint32_t> inputDimensions{batchSize, inputSize}; |
| 1340 | std::vector<float> inputValue{2.0f, 3.0f, 3.0f, 4.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1341 | |
| 1342 | // 01: The input-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1343 | // [num_units, input_size], where “num_units” corresponds to the number of cell units. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1344 | hidl_vec<uint32_t> inputToInputWeightsDimensions{0}; |
| 1345 | std::vector<float> inputToInputWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1346 | // 02: The input-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1347 | // [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1348 | hidl_vec<uint32_t> inputToForgetWeightsDimensions{numUnits, inputSize}; |
| 1349 | std::vector<float> inputToForgetWeightsValue{-0.55291498f, -0.42866567f, |
| 1350 | 0.13056988f, -0.36333650f, |
| 1351 | -0.22755712f, 0.28253698f, |
| 1352 | 0.24407166f, 0.33826375f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1353 | // 03: The input-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1354 | hidl_vec<uint32_t> inputToCellWeightsDimensions{numUnits, inputSize}; |
| 1355 | std::vector<float> inputToCellWeightsValue{-0.49770179f, -0.27711356f, |
| 1356 | -0.09624726f, 0.05100781f, |
| 1357 | 0.04717243f, 0.48944736f, |
| 1358 | -0.38535351f, -0.17212132f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1359 | // 04: The input-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1360 | // [num_units, input_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1361 | hidl_vec<uint32_t> inputToOutputWeightsDimensions{numUnits, inputSize}; |
| 1362 | std::vector<float> inputToOutputWeightsValue{ 0.10725588f, -0.02335852f, |
| 1363 | -0.55932593f, -0.09426838f, |
| 1364 | -0.44257352f, 0.54939759f, |
| 1365 | 0.01533556f, 0.42751634f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1366 | // 05: The recurrent-to-input weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1367 | // [num_units, output_size], where “output_size” corresponds to either the number of cell units (i.e., |
| 1368 | // “num_units”), or the second dimension of the “projection_weights”, if defined. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1369 | hidl_vec<uint32_t> recurrentToInputWeightsDimensions{0}; // VTS was {4, 4} -> {0} ? |
| 1370 | std::vector<float> recurrentToInputWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1371 | // 06: The recurrent-to-forget weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1372 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1373 | hidl_vec<uint32_t> recurrentToForgetWeightsDimensions{numUnits, outputSize}; |
| 1374 | std::vector<float> recurrentToForgetWeightsValue{-0.13832897f, -0.05151010f, -0.23590070f, -0.16661474f, |
| 1375 | -0.14340827f, 0.36986142f, 0.23414481f, 0.55899000f, |
| 1376 | 0.10798943f, -0.41174671f, 0.17751795f, -0.34484994f, |
| 1377 | -0.35874045f, -0.11352962f, 0.27268326f, 0.54058349f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1378 | // 07: The recurrent-to-cell weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1379 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1380 | hidl_vec<uint32_t> recurrentToCellWeightsDimensions{numUnits, outputSize}; |
| 1381 | std::vector<float> recurrentToCellWeightsValue{ 0.54066205f, -0.32668582f, -0.43562764f, -0.56094903f, |
| 1382 | 0.42957711f, 0.01841056f, -0.32764608f, -0.33027974f, |
| 1383 | -0.10826075f, 0.20675004f, 0.19069612f, -0.03026325f, |
| 1384 | -0.54532051f, 0.33003211f, 0.44901288f, 0.21193194f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1385 | // 08: The recurrent-to-output weights: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1386 | // [num_units, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1387 | hidl_vec<uint32_t> recurrentToOutputWeightsDimensions{numUnits, outputSize}; |
| 1388 | std::vector<float> recurrentToOutputWeightsValue{0.41613156f, 0.42610586f, -0.16495961f, -0.56638730f, |
| 1389 | 0.30579174f, -0.05115908f, -0.33941799f, 0.23364776f, |
| 1390 | 0.11178309f, 0.09481031f, -0.26424935f, 0.46261835f, |
| 1391 | 0.50248802f, 0.26114327f, -0.43736315f, 0.33149987f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1392 | // 09: The cell-to-input weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1393 | hidl_vec<uint32_t> cellToInputWeightsDimensions{0}; |
| 1394 | std::vector<float> cellToInputWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1395 | // 10: The cell-to-forget weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1396 | hidl_vec<uint32_t> cellToForgetWeightsDimensions{numUnits}; |
| 1397 | std::vector<float> cellToForgetWeightsValue{0.47485286f, -0.51955009f, -0.24458408f, 0.31544167f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1398 | // 11: The cell-to-output weights: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1399 | hidl_vec<uint32_t> cellToOutputWeightsDimensions{numUnits}; |
| 1400 | std::vector<float> cellToOutputWeightsValue{-0.17135078f, 0.82760304f, 0.85573703f, -0.77109635f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1401 | // 12: The input gate bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1402 | hidl_vec<uint32_t> inputGateBiasDimensions{0}; // VTS was {4} -> {0} ? |
| 1403 | std::vector<float> inputGateBiasValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1404 | // 13: The forget gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1405 | hidl_vec<uint32_t> forgetGateBiasDimensions{numUnits}; |
| 1406 | std::vector<float> forgetGateBiasValue{1.0f, 1.0f, 1.0f, 1.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1407 | // 14: The cell bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1408 | hidl_vec<uint32_t> cellBiasDimensions{numUnits}; |
| 1409 | std::vector<float> cellBiasValue(numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1410 | // 15: The output gate bias: A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1411 | hidl_vec<uint32_t> outputGateBiasDimensions{numUnits}; |
| 1412 | std::vector<float> outputGateBiasValue(numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1413 | // 16: The projection weights: Optional. A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape |
| 1414 | // [output_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1415 | hidl_vec<uint32_t> projectionWeightsDimensions{0}; |
| 1416 | std::vector<float> projectionWeightsValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1417 | // 17: The projection bias: Optional. A 1-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1418 | hidl_vec<uint32_t> projectionBiasDimensions{0}; |
| 1419 | std::vector<float> projectionBiasValue; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1420 | |
| 1421 | // 18: The output state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1422 | hidl_vec<uint32_t> outputStateInDimensions{batchSize, outputSize}; |
| 1423 | std::vector<float> outputStateInValue(batchSize * outputSize, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1424 | // 19: The cell state: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1425 | hidl_vec<uint32_t> cellStateInDimensions{batchSize, numUnits}; |
| 1426 | std::vector<float> cellStateInValue(batchSize * numUnits, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1427 | |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1428 | // Constant scalar values (the VTS test adds these as tensors of dim {}) |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1429 | // 20: The activation function: A value indicating the activation function: |
| 1430 | // 0: None; 1: Relu; 3: Relu6; 4: Tanh; 6: Sigmoid. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1431 | hidl_vec<uint32_t> activationFunctionDimensions{}; |
| 1432 | std::vector<int32_t> activationFunctionValue{4}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1433 | // 21: The clipping threshold: for the cell state, such that values are bound within [-cell_clip, cell_clip]. |
| 1434 | // If set to 0.0 then clipping is disabled. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1435 | hidl_vec<uint32_t> cellClippingThresholdDimensions{}; |
| 1436 | std::vector<float> cellClippingThresholdValue{0.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1437 | // 22: The clipping threshold: for the output from the projection layer, such that values are bound within |
| 1438 | // [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1439 | hidl_vec<uint32_t> projectionClippingThresholdDimensions{}; |
| 1440 | std::vector<float> projectionClippingThresholdValue{0.0f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1441 | |
| 1442 | // Outputs: |
| 1443 | // 0: The scratch buffer: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units * 4] with |
| 1444 | // CIFG, or [batch_size, num_units * 3] without CIFG. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1445 | // HOWEVER, by looking at the code, seems that it's the opposite: (cifg ? 3 : 4) * numUnits |
| 1446 | // Refer to: android/frameworks/ml/nn/common/operations/LSTM.cpp:319 |
| 1447 | // android/frameworks/ml/nn/common/operations/LSTMTest.cpp:114 |
| 1448 | // tensorflow/tensorflow/contrib/lite/kernels/lstm.cc:332 |
| 1449 | hidl_vec<uint32_t> scratchBufferDimensions{batchSize, numUnits * 3}; |
| 1450 | std::vector<float> scratchBufferValue(batchSize * numUnits * 3, 0.0f); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1451 | // 1: The output state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1452 | hidl_vec<uint32_t> outputStateOutDimensions{batchSize, outputSize}; |
| 1453 | std::vector<float> outputStateOutValue{-0.36444446f, -0.00352185f, 0.12886585f, -0.05163646f, |
| 1454 | -0.42734814f, -0.00478661f, 0.13455015f, -0.03560682f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1455 | // 2: The cell state (out): A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, num_units]. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1456 | hidl_vec<uint32_t> cellStateOutDimensions{batchSize, numUnits}; |
| 1457 | std::vector<float> cellStateOutValue{-0.76044439f, -0.01804161f, 0.18226376f, -0.06493707f, |
| 1458 | -0.90477051f, -0.04355603f, 0.18475688f, -0.04158677f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1459 | // 3: The output: A 2-D tensor of ANEURALNETWORKS_TENSOR_FLOAT32, of shape [batch_size, output_size]. This is |
| 1460 | // effectively the same as the current “output state (out)” value. |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1461 | hidl_vec<uint32_t> outputDimensions{batchSize, outputSize}; |
| 1462 | std::vector<float> outputValue{-0.36444446f, -0.00352185f, 0.12886585f, -0.05163646f, |
| 1463 | -0.42734814f, -0.00478661f, 0.13455015f, -0.03560682f}; |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1464 | |
| 1465 | LstmTestImpl(inputDimensions, inputValue, |
| 1466 | inputToInputWeightsDimensions, inputToInputWeightsValue, |
| 1467 | inputToForgetWeightsDimensions, inputToForgetWeightsValue, |
| 1468 | inputToCellWeightsDimensions, inputToCellWeightsValue, |
| 1469 | inputToOutputWeightsDimensions, inputToOutputWeightsValue, |
| 1470 | recurrentToInputWeightsDimensions, recurrentToInputWeightsValue, |
| 1471 | recurrentToForgetWeightsDimensions, recurrentToForgetWeightsValue, |
| 1472 | recurrentToCellWeightsDimensions, recurrentToCellWeightsValue, |
| 1473 | recurrentToOutputWeightsDimensions, recurrentToOutputWeightsValue, |
| 1474 | cellToInputWeightsDimensions, cellToInputWeightsValue, |
| 1475 | cellToForgetWeightsDimensions, cellToForgetWeightsValue, |
| 1476 | cellToOutputWeightsDimensions, cellToOutputWeightsValue, |
| 1477 | inputGateBiasDimensions, inputGateBiasValue, |
| 1478 | forgetGateBiasDimensions, forgetGateBiasValue, |
| 1479 | cellBiasDimensions, cellBiasValue, |
| 1480 | outputGateBiasDimensions, outputGateBiasValue, |
| 1481 | projectionWeightsDimensions, projectionWeightsValue, |
| 1482 | projectionBiasDimensions, projectionBiasValue, |
| 1483 | outputStateInDimensions, outputStateInValue, |
| 1484 | cellStateInDimensions, cellStateInValue, |
| 1485 | activationFunctionDimensions, activationFunctionValue, |
| 1486 | cellClippingThresholdDimensions, cellClippingThresholdValue, |
| 1487 | projectionClippingThresholdDimensions, projectionClippingThresholdValue, |
| 1488 | scratchBufferDimensions, scratchBufferValue, |
| 1489 | outputStateOutDimensions, outputStateOutValue, |
| 1490 | cellStateOutDimensions, cellStateOutValue, |
Matteo Martincigh | c743412 | 2018-11-14 12:27:04 +0000 | [diff] [blame] | 1491 | outputDimensions, outputValue, |
| 1492 | compute); |
| 1493 | } |
| 1494 | |
| 1495 | static const boost::array<armnn::Compute, 2> COMPUTE_DEVICES = {{ armnn::Compute::CpuRef, armnn::Compute::GpuAcc }}; |
| 1496 | |
| 1497 | BOOST_DATA_TEST_CASE(LstmNoCifgNoPeepholeNoProjectionTest, COMPUTE_DEVICES) |
| 1498 | { |
| 1499 | LstmNoCifgNoPeepholeNoProjection(sample); |
| 1500 | } |
| 1501 | |
| 1502 | BOOST_DATA_TEST_CASE(LstmCifgPeepholeNoProjectionTest, COMPUTE_DEVICES) |
| 1503 | { |
| 1504 | LstmCifgPeepholeNoProjection(sample); |
| 1505 | } |
| 1506 | |
| 1507 | BOOST_DATA_TEST_CASE(LstmNoCifgPeepholeProjectionTest, COMPUTE_DEVICES) |
| 1508 | { |
| 1509 | LstmNoCifgPeepholeProjection(sample); |
| 1510 | } |
| 1511 | |
| 1512 | BOOST_DATA_TEST_CASE(LstmCifgPeepholeNoProjectionBatch2Test, COMPUTE_DEVICES) |
| 1513 | { |
| 1514 | LstmCifgPeepholeNoProjectionBatch2(sample); |
telsoa01 | ce3e84a | 2018-08-31 09:31:35 +0100 | [diff] [blame] | 1515 | } |
| 1516 | |
| 1517 | BOOST_AUTO_TEST_SUITE_END() |