# # Copyright (C) 2017 The Android Open Source Project # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # # LSTM Test: No Cifg, No Peephole, No Projection, and No Clipping. model = Model() n_batch = 2 n_input = 2 n_cell = 4 n_output = n_cell input_ = Input("input", ("TENSOR_QUANT8_ASYMM", (n_batch, n_input), 1 / 128, 128)) weights_scale = 0.00408021 weights_zero_point = 100 input_to_input_weights = Input("inputToInputWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_input), weights_scale, weights_zero_point)) input_to_forget_weights = Input("inputToForgetWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_input), weights_scale, weights_zero_point)) input_to_cell_weights = Input("inputToCellWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_input), weights_scale, weights_zero_point)) input_to_output_weights = Input("inputToOutputWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_input), weights_scale, weights_zero_point)) recurrent_to_input_weights = Input("recurrentToInputWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_output), weights_scale, weights_zero_point)) recurrent_to_forget_weights = Input("recurrentToForgetWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_output), weights_scale, weights_zero_point)) recurrent_to_cell_weights = Input("recurrentToCellWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_output), weights_scale, weights_zero_point)) recurrent_to_output_weights = Input("recurrentToOutputWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_output), weights_scale, weights_zero_point)) input_gate_bias = Input("inputGateBias", ("TENSOR_INT32", (n_output,), weights_scale / 128., 0)) forget_gate_bias = Input("forgetGateBias", ("TENSOR_INT32", (n_output,), weights_scale / 128., 0)) cell_gate_bias = Input("cellGateBias", ("TENSOR_INT32", (n_output,), weights_scale / 128., 0)) output_gate_bias = Input("outputGateBias", ("TENSOR_INT32", (n_output,), weights_scale / 128., 0)) prev_cell_state = Input("prevCellState", ("TENSOR_QUANT16_SYMM", (n_batch, n_cell), 1 / 2048, 0)) prev_output = Input("prevOutput", ("TENSOR_QUANT8_ASYMM", (n_batch, n_output), 1 / 128, 128)) cell_state_out = Output("cellStateOut", ("TENSOR_QUANT16_SYMM", (n_batch, n_cell), 1 / 2048, 0)) output = Output("output", ("TENSOR_QUANT8_ASYMM", (n_batch, n_output), 1 / 128, 128)) model = model.Operation("QUANTIZED_16BIT_LSTM", input_, input_to_input_weights, input_to_forget_weights, input_to_cell_weights, input_to_output_weights, recurrent_to_input_weights, recurrent_to_forget_weights, recurrent_to_cell_weights, recurrent_to_output_weights, input_gate_bias, forget_gate_bias, cell_gate_bias, output_gate_bias, prev_cell_state, prev_output ).To([cell_state_out, output]) input_dict = { input_: [166, 179, 50, 150], input_to_input_weights: [146, 250, 235, 171, 10, 218, 171, 108], input_to_forget_weights: [24, 50, 132, 179, 158, 110, 3, 169], input_to_cell_weights: [133, 34, 29, 49, 206, 109, 54, 183], input_to_output_weights: [195, 187, 11, 99, 109, 10, 218, 48], recurrent_to_input_weights: [254, 206, 77, 168, 71, 20, 215, 6, 223, 7, 118, 225, 59, 130, 174, 26], recurrent_to_forget_weights: [137, 240, 103, 52, 68, 51, 237, 112, 0, 220, 89, 23, 69, 4, 207, 253], recurrent_to_cell_weights: [172, 60, 205, 65, 14, 0, 140, 168, 240, 223, 133, 56, 142, 64, 246, 216], recurrent_to_output_weights: [106, 214, 67, 23, 59, 158, 45, 3, 119, 132, 49, 205, 129, 218, 11, 98], input_gate_bias: [-7876, 13488, -726, 32839], forget_gate_bias: [9206, -46884, -11693, -38724], cell_gate_bias: [39481, 48624, 48976, -21419], output_gate_bias: [-58999, -17050, -41852, -40538], prev_cell_state: [876, 1034, 955, -909, 761, 1029, 796, -1036], prev_output: [136, 150, 140, 115, 135, 152, 138, 112], } output_dict = { cell_state_out: [1485, 1177, 1373, -1023, 1019, 1355, 1097, -1235], output: [140, 151, 146, 112, 136, 156, 142, 112] } Example((input_dict, output_dict), model=model).AddVariations("relaxed") # TEST 2: same as the first one but only the first batch is tested and weights # are compile time constants model = Model() n_batch = 1 n_input = 2 n_cell = 4 n_output = n_cell input_ = Input("input", ("TENSOR_QUANT8_ASYMM", (n_batch, n_input), 1 / 128, 128)) weights_scale = 0.00408021 weights_zero_point = 100 input_to_input_weights = Parameter( "inputToInputWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_input), weights_scale, weights_zero_point), [146, 250, 235, 171, 10, 218, 171, 108]) input_to_forget_weights = Parameter( "inputToForgetWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_input), weights_scale, weights_zero_point), [24, 50, 132, 179, 158, 110, 3, 169]) input_to_cell_weights = Parameter( "inputToCellWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_input), weights_scale, weights_zero_point), [133, 34, 29, 49, 206, 109, 54, 183]) input_to_output_weights = Parameter( "inputToOutputWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_input), weights_scale, weights_zero_point), [195, 187, 11, 99, 109, 10, 218, 48]) recurrent_to_input_weights = Parameter( "recurrentToInputWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_output), weights_scale, weights_zero_point), [254, 206, 77, 168, 71, 20, 215, 6, 223, 7, 118, 225, 59, 130, 174, 26]) recurrent_to_forget_weights = Parameter( "recurrentToForgetWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_output), weights_scale, weights_zero_point), [137, 240, 103, 52, 68, 51, 237, 112, 0, 220, 89, 23, 69, 4, 207, 253]) recurrent_to_cell_weights = Parameter( "recurrentToCellWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_output), weights_scale, weights_zero_point), [172, 60, 205, 65, 14, 0, 140, 168, 240, 223, 133, 56, 142, 64, 246, 216]) recurrent_to_output_weights = Parameter( "recurrentToOutputWeights", ("TENSOR_QUANT8_ASYMM", (n_output, n_output), weights_scale, weights_zero_point), [106, 214, 67, 23, 59, 158, 45, 3, 119, 132, 49, 205, 129, 218, 11, 98]) input_gate_bias = Parameter("inputGateBias", ("TENSOR_INT32", (n_output,), weights_scale / 128., 0), [-7876, 13488, -726, 32839]) forget_gate_bias = Parameter("forgetGateBias", ("TENSOR_INT32", (n_output,), weights_scale / 128., 0), [9206, -46884, -11693, -38724]) cell_gate_bias = Parameter("cellGateBias", ("TENSOR_INT32", (n_output,), weights_scale / 128., 0), [39481, 48624, 48976, -21419]) output_gate_bias = Parameter("outputGateBias", ("TENSOR_INT32", (n_output,), weights_scale / 128., 0), [-58999, -17050, -41852, -40538]) prev_cell_state = Input("prevCellState", ("TENSOR_QUANT16_SYMM", (n_batch, n_cell), 1 / 2048, 0)) prev_output = Input("prevOutput", ("TENSOR_QUANT8_ASYMM", (n_batch, n_output), 1 / 128, 128)) cell_state_out = Output("cellStateOut", ("TENSOR_QUANT16_SYMM", (n_batch, n_cell), 1 / 2048, 0)) output = Output("output", ("TENSOR_QUANT8_ASYMM", (n_batch, n_output), 1 / 128, 128)) model = model.Operation("QUANTIZED_16BIT_LSTM", input_, input_to_input_weights, input_to_forget_weights, input_to_cell_weights, input_to_output_weights, recurrent_to_input_weights, recurrent_to_forget_weights, recurrent_to_cell_weights, recurrent_to_output_weights, input_gate_bias, forget_gate_bias, cell_gate_bias, output_gate_bias, prev_cell_state, prev_output).To([cell_state_out, output]) input_dict = { input_: [166, 179], prev_cell_state: [876, 1034, 955, -909], prev_output: [136, 150, 140, 115], } output_dict = { cell_state_out: [1485, 1177, 1373, -1023], output: [140, 151, 146, 112] } Example((input_dict, output_dict), model=model, name="constant_weights").AddVariations("relaxed")