Source code

Revision control

Copy as Markdown

Other Tools

Test Info:

// META: title=test WebNN API lstm operation
// META: global=window,dedicatedworker
// META: variant=?cpu
// META: variant=?gpu
// META: variant=?npu
// META: script=../resources/utils.js
// META: timeout=long
'use strict';
// Long Short-Term Memory [LSTM] recurrent network uses an input, output,
// forget, and cell gate to compute the output state that rolls into the output
// across the temporal sequence of the network.
// enum MLRecurrentNetworkDirection {
// "forward",
// "backward",
// "both"
// };
//
// enum MLRecurrentNetworkActivation {
// "relu",
// "sigmoid",
// "tanh"
// };
//
// enum MLLstmWeightLayout {
// "iofg", // input-output-forget-cell gate ordering
// "ifgo" // input-forget-cell-output gate ordering
// };
//
// dictionary MLLstmOptions {
// MLOperand bias;
// MLOperand recurrentBias;
// MLOperand peepholeWeight;
// MLOperand initialHiddenState;
// MLOperand initialCellState;
// boolean returnSequence = false;
// MLRecurrentNetworkDirection direction = "forward";
// MLLstmWeightLayout layout = "iofg";
// sequence<MLRecurrentNetworkActivation> activations;
// };
//
// sequence<MLOperand> lstm(MLOperand input,
// MLOperand weight,
// MLOperand recurrentWeight,
// [EnforceRange] unsigned long steps,
// [EnforceRange] unsigned long hiddenSize,
// optional MLLstmOptions options = {});
const getLstmPrecisionTolerance = (graphResources) => {
const toleranceValueDict = {float32: 1};
const expectedDataType =
graphResources
.expectedOutputs[Object.keys(graphResources.expectedOutputs)[0]]
.descriptor.dataType;
return {metricType: 'ULP', value: toleranceValueDict[expectedDataType]};
};
const lstmTests = [
{
'name':
'lstm float32 tensors steps=1 with options.bias, options.recurrentBias and options.activations=[\'relu\', \'relu\', \'relu\']',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'lstm float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=[\'relu\', \'relu\', \'relu\'] and options.peepholeWeight',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmPeepholeWeight': {
'data': [0, 0, 0, 0, 0, 0],
'descriptor': {shape: [1, 6], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'peepholeWeight': 'lstmPeepholeWeight',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'lstm float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=[\'relu\', \'relu\', \'relu\'] and options.initialHiddenState',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmInitialHiddenState': {
'data': [0, 0, 0, 0],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'initialHiddenState': 'lstmInitialHiddenState',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'lstm float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=[\'relu\', \'relu\', \'relu\'] and options.initialCellState',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmInitialCellState': {
'data': [0, 0, 0, 0],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'initialCellState': 'lstmInitialCellState',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'lstm float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=[\'relu\', \'relu\', \'relu\'] and explicit options.returnSequence=false',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'returnSequence': false,
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'lstm float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=[\'relu\', \'relu\', \'relu\'] and options.returnSequence=true',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'returnSequence': true,
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2', 'lstmOutput3']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput3': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'lstm float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=[\'relu\', \'relu\', \'relu\'] and explicit options.direction=\'forward\'',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'direction': 'forward',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'lstm float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=[\'relu\', \'relu\', \'relu\'] and explicit options.layout=\'iofg\'',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'layout': 'iofg',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'lstm float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=[\'relu\', \'relu\', \'relu\'] and options.layout=\'ifgo\'',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'layout': 'ifgo',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name': 'lstm float32 tensors steps=1 with all options',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmPeepholeWeight': {
'data': [0, 0, 0, 0, 0, 0],
'descriptor': {shape: [1, 6], dataType: 'float32'}
},
'lstmInitialHiddenState': {
'data': [0, 0, 0, 0],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmInitialCellState': {
'data': [0, 0, 0, 0],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 1},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'peepholeWeight': 'lstmPeepholeWeight',
'initialHiddenState': 'lstmInitialHiddenState',
'initialCellState': 'lstmInitialCellState',
'returnSequence': true,
'direction': 'forward',
'layout': 'iofg',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2', 'lstmOutput3']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [1, 4, 9, 36],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput3': {
'data': [1, 8, 27, 216],
'descriptor': {shape: [1, 1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name':
'lstm float32 tensors steps=2 with options.bias, options.recurrentBias, options.activations=[\'relu\', \'relu\', \'relu\'] and options.direction=\'backward\'',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1, 3, 4, 1, 2],
'descriptor': {shape: [2, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 2},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'direction': 'backward',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [
10.469000816345215, 58.02900695800781, 74.52900695800781,
518.948974609375
],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [
5.510000228881836, 20.01000213623047, 19.110000610351564,
75.20999908447266
],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
}
}
},
{
'name': 'lstm float32 tensors steps=2 with all options',
'graph': {
'inputs': {
'lstmInput': {
'data': [1, 2, 2, 1, 3, 4, 1, 2],
'descriptor': {shape: [2, 2, 2], dataType: 'float32'}
},
'lstmWeight': {
'data': [1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2, 1, -1, 2, -2],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmRecurrentWeight': {
'data': [
0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
0.1, 0.1, 0.1
],
'descriptor': {shape: [1, 8, 2], dataType: 'float32'}
},
'lstmBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmRecurrentBias': {
'data': [1, 2, 1, 2, 1, 2, 1, 2],
'descriptor': {shape: [1, 8], dataType: 'float32'}
},
'lstmPeepholeWeight': {
'data': [0, 0, 0, 0, 0, 0],
'descriptor': {shape: [1, 6], dataType: 'float32'}
},
'lstmInitialHiddenState': {
'data': [0, 0, 0, 0],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmInitialCellState': {
'data': [0, 0, 0, 0],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
}
},
'operators': [{
'name': 'lstm',
'arguments': [
{'input': 'lstmInput'}, {'weight': 'lstmWeight'},
{'recurrentWeight': 'lstmRecurrentWeight'}, {'steps': 2},
{'hiddenSize': 2}, {
'options': {
'bias': 'lstmBias',
'recurrentBias': 'lstmRecurrentBias',
'peepholeWeight': 'lstmPeepholeWeight',
'initialHiddenState': 'lstmInitialHiddenState',
'initialCellState': 'lstmInitialCellState',
'returnSequence': true,
'direction': 'backward',
'layout': 'iofg',
'activations': ['relu', 'relu', 'relu']
}
}
],
'outputs': ['lstmOutput1', 'lstmOutput2', 'lstmOutput3']
}],
'expectedOutputs': {
'lstmOutput1': {
'data': [
10.469000816345215, 58.02900695800781, 74.52900695800781,
518.948974609375
],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput2': {
'data': [
5.510000228881836, 20.01000213623047, 19.110000610351564,
75.20999908447266
],
'descriptor': {shape: [1, 2, 2], dataType: 'float32'}
},
'lstmOutput3': {
'data': [
10.469000816345215, 58.02900695800781, 74.52900695800781,
518.948974609375, 1, 8, 1, 8
],
'descriptor': {shape: [2, 1, 2, 2], dataType: 'float32'}
}
}
}
}
];
if (navigator.ml) {
lstmTests.forEach((test) => {
webnn_conformance_test(
buildAndExecuteGraph, getLstmPrecisionTolerance, test);
});
} else {
test(() => assert_implements(navigator.ml, 'missing navigator.ml'));
}