tor-browser

The Tor Browser
git clone https://git.dasho.dev/tor-browser.git
Log | Files | Refs | README | LICENSE

gru.https.any.js (73299B)

      1 // META: title=test WebNN API gru operation
      2 // META: global=window
      3 // META: variant=?cpu
      4 // META: variant=?gpu
      5 // META: variant=?npu
      6 // META: script=../resources/utils.js
      7 // META: timeout=long
      8 
      9 'use strict';
     10 
     11 // https://www.w3.org/TR/webnn/#api-mlgraphbuilder-gru
     12 // Gated Recurrent Unit recurrent network uses an update, reset, and new gate
     13 // to compute the output state that rolls into the output across the temporal
     14 // sequence of the network.
     15 //
     16 // enum MLGruWeightLayout {
     17 //   "zrn",  // update-reset-new gate ordering
     18 //   "rzn"   // reset-update-new gate ordering
     19 // };
     20 //
     21 // enum MLRecurrentNetworkActivation {
     22 //   "relu",
     23 //   "sigmoid",
     24 //   "tanh"
     25 // };
     26 //
     27 // enum MLRecurrentNetworkDirection {
     28 //   "forward",
     29 //   "backward",
     30 //   "both"
     31 // };
     32 //
     33 // dictionary MLGruOptions {
     34 //   MLOperand bias;
     35 //   MLOperand recurrentBias;
     36 //   MLOperand initialHiddenState;
     37 //   boolean resetAfter = true;
     38 //   boolean returnSequence = false;
     39 //   MLRecurrentNetworkDirection direction = "forward";
     40 //   MLGruWeightLayout layout = "zrn";
     41 //   sequence<MLRecurrentNetworkActivation> activations;
     42 // };
     43 //
     44 // sequence<MLOperand> gru(MLOperand input,
     45 //                         MLOperand weight,
     46 //                         MLOperand recurrentWeight,
     47 //                         [EnforceRange] unsigned long steps,
     48 //                         [EnforceRange] unsigned long hiddenSize,
     49 //                         optional MLGruOptions options = {});
     50 
     51 
     52 const getGruPrecisionTolerance = (graphResources) => {
     53  const toleranceValueDict = {float32: 6, float16: 6};
     54  const expectedDataType =
     55      graphResources
     56          .expectedOutputs[Object.keys(graphResources.expectedOutputs)[0]]
     57          .descriptor.dataType;
     58  return {metricType: 'ULP', value: toleranceValueDict[expectedDataType]};
     59 };
     60 
     61 const gruTests = [
     62  // float32 tests
     63  {
     64    'name':
     65        "gru float32 tensors steps=1 with options.bias, options.recurrentBias and options.activations=['relu', 'relu']",
     66    'graph': {
     67      'inputs': {
     68        'gruInput': {
     69          'data': [1, 2, 2, 1, 1, 1],
     70          'descriptor': {shape: [1, 3, 2], dataType: 'float32'}
     71        },
     72        'gruWeight': {
     73          'data': [
     74            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
     75            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
     76          ],
     77          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
     78        },
     79        'gruRecurrentWeight': {
     80          'data': [
     81            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
     82            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
     83            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
     84            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
     85          ],
     86          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
     87        },
     88        'gruBias': {
     89          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
     90          'descriptor': {shape: [1, 12], dataType: 'float32'}
     91        },
     92        'gruRecurrentBias': {
     93          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
     94          'descriptor': {shape: [1, 12], dataType: 'float32'}
     95        },
     96      },
     97      'operators': [{
     98        'name': 'gru',
     99        'arguments': [
    100          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    101          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
    102          {'hiddenSize': 4}, {
    103            'options': {
    104              'bias': 'gruBias',
    105              'recurrentBias': 'gruRecurrentBias',
    106              'resetAfter': false,
    107              'activations': ['relu', 'relu']
    108            }
    109          }
    110        ],
    111        'outputs': ['gruOutput']
    112      }],
    113      'expectedOutputs': {
    114        'gruOutput': {
    115          'data':
    116              [0, 0, -0.25, -3.84, -4, -15, -2.25, -3.41, -1, -3, -1, -3.41],
    117          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    118        }
    119      }
    120    }
    121  },
    122  {
    123    'name':
    124        "gru float32 tensors steps=1 with options.bias, options.recurrentBias and options.activations=['relu', 'relu'] and reset_after=true",
    125    'graph': {
    126      'inputs': {
    127        'gruInput': {
    128          'data': [1, 2, 2, 1, 1, 1],
    129          'descriptor': {shape: [1, 3, 2], dataType: 'float32'}
    130        },
    131        'gruWeight': {
    132          'data': [
    133            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    134            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    135          ],
    136          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    137        },
    138        'gruRecurrentWeight': {
    139          'data': [
    140            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    141            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    142            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    143            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    144          ],
    145          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    146        },
    147        'gruBias': {
    148          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    149          'descriptor': {shape: [1, 12], dataType: 'float32'}
    150        },
    151        'gruRecurrentBias': {
    152          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    153          'descriptor': {shape: [1, 12], dataType: 'float32'}
    154        },
    155      },
    156      'operators': [{
    157        'name': 'gru',
    158        'arguments': [
    159          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    160          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
    161          {'hiddenSize': 4}, {
    162            'options': {
    163              'bias': 'gruBias',
    164              'recurrentBias': 'gruRecurrentBias',
    165              'resetAfter': true,
    166              'activations': ['relu', 'relu']
    167            }
    168          }
    169        ],
    170        'outputs': ['gruOutput']
    171      }],
    172      'expectedOutputs': {
    173        'gruOutput': {
    174          'data': [
    175            0, 0, -0.375, -5.7599992752075195, -6, -22.5, -3.375,
    176            -5.114999771118164, -1.5, -4.5, -1.5, -5.114999771118164
    177          ],
    178          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    179        }
    180      }
    181    }
    182  },
    183  {
    184    'name':
    185        "gru float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and explicit options.direction='forward'",
    186    'graph': {
    187      'inputs': {
    188        'gruInput': {
    189          'data': [1, 2, 2, 1, 1, 1],
    190          'descriptor': {shape: [1, 3, 2], dataType: 'float32'}
    191        },
    192        'gruWeight': {
    193          'data': [
    194            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    195            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    196          ],
    197          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    198        },
    199        'gruRecurrentWeight': {
    200          'data': [
    201            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    202            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    203            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    204            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    205          ],
    206          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    207        },
    208        'gruBias': {
    209          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    210          'descriptor': {shape: [1, 12], dataType: 'float32'}
    211        },
    212        'gruRecurrentBias': {
    213          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    214          'descriptor': {shape: [1, 12], dataType: 'float32'}
    215        },
    216      },
    217      'operators': [{
    218        'name': 'gru',
    219        'arguments': [
    220          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    221          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
    222          {'hiddenSize': 4}, {
    223            'options': {
    224              'bias': 'gruBias',
    225              'recurrentBias': 'gruRecurrentBias',
    226              'resetAfter': false,
    227              'direction': 'forward',
    228              'activations': ['relu', 'relu']
    229            }
    230          }
    231        ],
    232        'outputs': ['gruOutput']
    233      }],
    234      'expectedOutputs': {
    235        'gruOutput': {
    236          'data':
    237              [0, 0, -0.25, -3.84, -4, -15, -2.25, -3.41, -1, -3, -1, -3.41],
    238          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    239        }
    240      }
    241    }
    242  },
    243  {
    244    'name':
    245        "gru float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and explicit options.layout='zrn'",
    246    'graph': {
    247      'inputs': {
    248        'gruInput': {
    249          'data': [1, 2, 2, 1, 1, 1],
    250          'descriptor': {shape: [1, 3, 2], dataType: 'float32'}
    251        },
    252        'gruWeight': {
    253          'data': [
    254            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    255            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    256          ],
    257          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    258        },
    259        'gruRecurrentWeight': {
    260          'data': [
    261            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    262            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    263            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    264            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    265          ],
    266          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    267        },
    268        'gruBias': {
    269          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    270          'descriptor': {shape: [1, 12], dataType: 'float32'}
    271        },
    272        'gruRecurrentBias': {
    273          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    274          'descriptor': {shape: [1, 12], dataType: 'float32'}
    275        },
    276      },
    277      'operators': [{
    278        'name': 'gru',
    279        'arguments': [
    280          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    281          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
    282          {'hiddenSize': 4}, {
    283            'options': {
    284              'bias': 'gruBias',
    285              'recurrentBias': 'gruRecurrentBias',
    286              'resetAfter': false,
    287              'layout': 'zrn',
    288              'activations': ['relu', 'relu']
    289            }
    290          }
    291        ],
    292        'outputs': ['gruOutput']
    293      }],
    294      'expectedOutputs': {
    295        'gruOutput': {
    296          'data':
    297              [0, 0, -0.25, -3.84, -4, -15, -2.25, -3.41, -1, -3, -1, -3.41],
    298          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    299        }
    300      }
    301    }
    302  },
    303  {
    304    'name':
    305        "gru float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and options.layout='rzn'",
    306    'graph': {
    307      'inputs': {
    308        'gruInput': {
    309          'data': [1, 2, 2, 1, 1, 1],
    310          'descriptor': {shape: [1, 3, 2], dataType: 'float32'}
    311        },
    312        'gruWeight': {
    313          'data': [
    314            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    315            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    316          ],
    317          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    318        },
    319        'gruRecurrentWeight': {
    320          'data': [
    321            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    322            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    323            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    324            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    325          ],
    326          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    327        },
    328        'gruBias': {
    329          'data': [1, 1, 1, 1, 1, 2, 1, 2, 0.5, 0.5, 0.5, 0.5],
    330          'descriptor': {shape: [1, 12], dataType: 'float32'}
    331        },
    332        'gruRecurrentBias': {
    333          'data': [1, 1, 1, 1, 1, 2, 1, 2, 0.5, 0.5, 0.5, 0.5],
    334          'descriptor': {shape: [1, 12], dataType: 'float32'}
    335        },
    336      },
    337      'operators': [{
    338        'name': 'gru',
    339        'arguments': [
    340          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    341          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
    342          {'hiddenSize': 4}, {
    343            'options': {
    344              'bias': 'gruBias',
    345              'recurrentBias': 'gruRecurrentBias',
    346              'resetAfter': false,
    347              'layout': 'rzn',
    348              'activations': ['relu', 'relu']
    349            }
    350          }
    351        ],
    352        'outputs': ['gruOutput']
    353      }],
    354      'expectedOutputs': {
    355        'gruOutput': {
    356          'data':
    357              [0, 0, -0.25, -3.84, -4, -15, -2.25, -3.41, -1, -3, -1, -3.41],
    358          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    359        }
    360      }
    361    }
    362  },
    363  {
    364    'name':
    365        "gru float32 tensors steps=1 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and options.initialHiddenState",
    366    'graph': {
    367      'inputs': {
    368        'gruInput': {
    369          'data': [1, 2, 2, 1, 1, 1],
    370          'descriptor': {shape: [1, 3, 2], dataType: 'float32'}
    371        },
    372        'gruWeight': {
    373          'data': [
    374            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    375            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    376          ],
    377          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    378        },
    379        'gruRecurrentWeight': {
    380          'data': [
    381            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    382            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    383            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    384            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    385          ],
    386          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    387        },
    388        'gruBias': {
    389          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    390          'descriptor': {shape: [1, 12], dataType: 'float32'}
    391        },
    392        'gruRecurrentBias': {
    393          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    394          'descriptor': {shape: [1, 12], dataType: 'float32'}
    395        },
    396        'gruInitialHiddenState': {
    397          'data': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
    398          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    399        }
    400      },
    401      'operators': [{
    402        'name': 'gru',
    403        'arguments': [
    404          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    405          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
    406          {'hiddenSize': 4}, {
    407            'options': {
    408              'bias': 'gruBias',
    409              'recurrentBias': 'gruRecurrentBias',
    410              'initialHiddenState': 'gruInitialHiddenState',
    411              'resetAfter': false,
    412              'activations': ['relu', 'relu']
    413            }
    414          }
    415        ],
    416        'outputs': ['gruOutput']
    417      }],
    418      'expectedOutputs': {
    419        'gruOutput': {
    420          'data':
    421              [0, 0, -0.25, -3.84, -4, -15, -2.25, -3.41, -1, -3, -1, -3.41],
    422          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    423        }
    424      }
    425    }
    426  },
    427  {
    428    'name': 'gru float32 tensors steps=1 all options',
    429    'graph': {
    430      'inputs': {
    431        'gruInput': {
    432          'data': [1, 2, 2, 1, 1, 1],
    433          'descriptor': {shape: [1, 3, 2], dataType: 'float32'}
    434        },
    435        'gruWeight': {
    436          'data': [
    437            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    438            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    439          ],
    440          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    441        },
    442        'gruRecurrentWeight': {
    443          'data': [
    444            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    445            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    446            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    447            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    448          ],
    449          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    450        },
    451        'gruBias': {
    452          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    453          'descriptor': {shape: [1, 12], dataType: 'float32'}
    454        },
    455        'gruRecurrentBias': {
    456          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    457          'descriptor': {shape: [1, 12], dataType: 'float32'}
    458        },
    459        'gruInitialHiddenState': {
    460          'data': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
    461          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    462        }
    463      },
    464      'operators': [{
    465        'name': 'gru',
    466        'arguments': [
    467          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    468          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
    469          {'hiddenSize': 4}, {
    470            'options': {
    471              'bias': 'gruBias',
    472              'recurrentBias': 'gruRecurrentBias',
    473              'initialHiddenState': 'gruInitialHiddenState',
    474              'resetAfter': false,
    475              'returnSequence': true,
    476              'direction': 'forward',
    477              'layout': 'zrn',
    478              'activations': ['relu', 'relu']
    479            }
    480          }
    481        ],
    482        'outputs': ['gruOutput1', 'gruOutput2']
    483      }],
    484      'expectedOutputs': {
    485        'gruOutput1': {
    486          'data':
    487              [0, 0, -0.25, -3.84, -4, -15, -2.25, -3.41, -1, -3, -1, -3.41],
    488          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    489        },
    490        'gruOutput2': {
    491          'data':
    492              [0, 0, -0.25, -3.84, -4, -15, -2.25, -3.41, -1, -3, -1, -3.41],
    493          'descriptor': {shape: [1, 1, 3, 4], dataType: 'float32'}
    494        }
    495      }
    496    }
    497  },
    498  {
    499    'name':
    500        "gru float32 tensors steps=2 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and options.direction='backward'",
    501    'graph': {
    502      'inputs': {
    503        'gruInput': {
    504          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
    505          'descriptor': {shape: [2, 3, 2], dataType: 'float32'}
    506        },
    507        'gruWeight': {
    508          'data': [
    509            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    510            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    511          ],
    512          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    513        },
    514        'gruRecurrentWeight': {
    515          'data': [
    516            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    517            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    518            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    519            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    520          ],
    521          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    522        },
    523        'gruBias': {
    524          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    525          'descriptor': {shape: [1, 12], dataType: 'float32'}
    526        },
    527        'gruRecurrentBias': {
    528          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    529          'descriptor': {shape: [1, 12], dataType: 'float32'}
    530        },
    531      },
    532      'operators': [{
    533        'name': 'gru',
    534        'arguments': [
    535          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    536          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
    537          {'hiddenSize': 4}, {
    538            'options': {
    539              'bias': 'gruBias',
    540              'recurrentBias': 'gruRecurrentBias',
    541              'resetAfter': false,
    542              'direction': 'backward',
    543              'activations': ['relu', 'relu']
    544            }
    545          }
    546        ],
    547        'outputs': ['gruOutput']
    548      }],
    549      'expectedOutputs': {
    550        'gruOutput': {
    551          'data': [
    552            0, 0, -0.24974998831748963, -18.59588623046875, -2.0657243728637697,
    553            -10.551867485046387, -1.3937838077545167, -15.2454833984375,
    554            -1.1589999198913575, -9.476999282836914, -1.1589999198913575,
    555            -11.319169044494629
    556          ],
    557          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    558        }
    559      }
    560    }
    561  },
    562  {
    563    'name':
    564        "gru float32 tensors steps=2 with options.bias, options.recurrentBias, options.direction='backward', options.activations=['relu', 'relu'] and explicit options.returnSequence=false",
    565    'graph': {
    566      'inputs': {
    567        'gruInput': {
    568          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
    569          'descriptor': {shape: [2, 3, 2], dataType: 'float32'}
    570        },
    571        'gruWeight': {
    572          'data': [
    573            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    574            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    575          ],
    576          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    577        },
    578        'gruRecurrentWeight': {
    579          'data': [
    580            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    581            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    582            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    583            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    584          ],
    585          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    586        },
    587        'gruBias': {
    588          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    589          'descriptor': {shape: [1, 12], dataType: 'float32'}
    590        },
    591        'gruRecurrentBias': {
    592          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    593          'descriptor': {shape: [1, 12], dataType: 'float32'}
    594        },
    595      },
    596      'operators': [{
    597        'name': 'gru',
    598        'arguments': [
    599          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    600          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
    601          {'hiddenSize': 4}, {
    602            'options': {
    603              'bias': 'gruBias',
    604              'recurrentBias': 'gruRecurrentBias',
    605              'resetAfter': false,
    606              'returnSequence': false,
    607              'direction': 'backward',
    608              'activations': ['relu', 'relu']
    609            }
    610          }
    611        ],
    612        'outputs': ['gruOutput']
    613      }],
    614      'expectedOutputs': {
    615        'gruOutput': {
    616          'data': [
    617            0, 0, -0.24974998831748963, -18.59588623046875, -2.0657243728637697,
    618            -10.551867485046387, -1.3937838077545167, -15.2454833984375,
    619            -1.1589999198913575, -9.476999282836914, -1.1589999198913575,
    620            -11.319169044494629
    621          ],
    622          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    623        }
    624      }
    625    }
    626  },
    627  {
    628    'name':
    629        "gru float32 tensors steps=2 with options.bias, options.recurrentBias, options.direction='backward', options.activations=['relu', 'relu'] and options.returnSequence=true",
    630    'graph': {
    631      'inputs': {
    632        'gruInput': {
    633          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
    634          'descriptor': {shape: [2, 3, 2], dataType: 'float32'}
    635        },
    636        'gruWeight': {
    637          'data': [
    638            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    639            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    640          ],
    641          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    642        },
    643        'gruRecurrentWeight': {
    644          'data': [
    645            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    646            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    647            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    648            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    649          ],
    650          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    651        },
    652        'gruBias': {
    653          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    654          'descriptor': {shape: [1, 12], dataType: 'float32'}
    655        },
    656        'gruRecurrentBias': {
    657          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    658          'descriptor': {shape: [1, 12], dataType: 'float32'}
    659        },
    660      },
    661      'operators': [{
    662        'name': 'gru',
    663        'arguments': [
    664          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    665          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
    666          {'hiddenSize': 4}, {
    667            'options': {
    668              'bias': 'gruBias',
    669              'recurrentBias': 'gruRecurrentBias',
    670              'resetAfter': false,
    671              'returnSequence': true,
    672              'direction': 'backward',
    673              'activations': ['relu', 'relu']
    674            }
    675          }
    676        ],
    677        'outputs': ['gruOutput1', 'gruOutput2']
    678      }],
    679      'expectedOutputs': {
    680        'gruOutput1': {
    681          'data': [
    682            0, 0, -0.24974998831748963, -18.59588623046875, -2.0657243728637697,
    683            -10.551867485046387, -1.3937838077545167, -15.2454833984375,
    684            -1.1589999198913575, -9.476999282836914, -1.1589999198913575,
    685            -11.319169044494629
    686          ],
    687          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    688        },
    689        'gruOutput2': {
    690          'data': [
    691            0,
    692            0,
    693            -0.24974998831748963,
    694            -18.59588623046875,
    695            -2.0657243728637697,
    696            -10.551867485046387,
    697            -1.3937838077545167,
    698            -15.2454833984375,
    699            -1.1589999198913575,
    700            -9.476999282836914,
    701            -1.1589999198913575,
    702            -11.319169044494629,
    703            0,
    704            0,
    705            -0.25,
    706            -4.760000228881836,
    707            0,
    708            0,
    709            -0.25,
    710            -3.8399999141693117,
    711            -1,
    712            -3,
    713            -1,
    714            -3.4100000858306886
    715          ],
    716          'descriptor': {shape: [2, 1, 3, 4], dataType: 'float32'}
    717        }
    718      }
    719    }
    720  },
    721  {
    722    'name':
    723        "gru float32 tensors steps=2 with options.bias, options.recurrentBias, options.direction='both' and options.returnSequence=true",
    724    'graph': {
    725      'inputs': {
    726        'gruInput': {
    727          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
    728          'descriptor': {shape: [2, 3, 2], dataType: 'float32'}
    729        },
    730        'gruWeight': {
    731          'data': [
    732            1, -1, 2, -2, 0.5, -0.5, 0, 0.1, 1, -1, 2, -2, 0.5, -0.5, 0, 0.1,
    733            1, -1, 2, -2, 0.5, -0.5, 0, 0.1, 1, -1, 2, -2, 0.5, -0.5, 0, 0.1,
    734            1, -1, 2, -2, 0.5, -0.5, 0, 0.1, 1, -1, 2, -2, 0.5, -0.5, 0, 0.1
    735          ],
    736          'descriptor': {shape: [2, 12, 2], dataType: 'float32'}
    737        },
    738        'gruRecurrentWeight': {
    739          'data': [
    740            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    741            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    742            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    743            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    744            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    745            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    746            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    747            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    748          ],
    749          'descriptor': {shape: [2, 12, 4], dataType: 'float32'}
    750        },
    751        'gruBias': {
    752          'data': [
    753            1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5,
    754            1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5
    755          ],
    756          'descriptor': {shape: [2, 12], dataType: 'float32'}
    757        },
    758        'gruRecurrentBias': {
    759          'data': [
    760            1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5,
    761            1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5
    762          ],
    763          'descriptor': {shape: [2, 12], dataType: 'float32'}
    764        },
    765      },
    766      'operators': [{
    767        'name': 'gru',
    768        'arguments': [
    769          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    770          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
    771          {'hiddenSize': 4}, {
    772            'options': {
    773              'bias': 'gruBias',
    774              'recurrentBias': 'gruRecurrentBias',
    775              'resetAfter': true,
    776              'returnSequence': true,
    777              'activations': ['relu', 'relu'],
    778              'direction': 'both'
    779            }
    780          }
    781        ],
    782        'outputs': ['gruOutput1', 'gruOutput2']
    783      }],
    784      'expectedOutputs': {
    785        'gruOutput1': {
    786          'data': [
    787            0,
    788            0,
    789            -0.33243751525878906,
    790            -23.753076553344727,
    791            0,
    792            0,
    793            0,
    794            -2.213315725326538,
    795            -1.1077498197555542,
    796            -12.323249816894531,
    797            -1.1077498197555542,
    798            -14.518925666809082,
    799            0,
    800            0,
    801            -0.28068751096725464,
    802            -25.444257736206055,
    803            -1.7041922807693481,
    804            -9.28022289276123,
    805            -1.4041223526000977,
    806            -21.154621124267578,
    807            -1.1077498197555542,
    808            -12.323249816894531,
    809            -1.1077498197555542,
    810            -14.518925666809082
    811          ],
    812          'descriptor': {shape: [2, 3, 4], dataType: 'float32'}
    813        },
    814        'gruOutput2': {
    815          'data': [
    816            0,
    817            0,
    818            -0.375,
    819            -5.7599992752075195,
    820            -6,
    821            -22.5,
    822            -3.375,
    823            -5.114999771118164,
    824            -1.5,
    825            -4.5,
    826            -1.5,
    827            -5.114999771118164,
    828            0,
    829            0,
    830            -0.28068751096725464,
    831            -25.444257736206055,
    832            -1.7041922807693481,
    833            -9.28022289276123,
    834            -1.4041223526000977,
    835            -21.154621124267578,
    836            -1.1077498197555542,
    837            -12.323249816894531,
    838            -1.1077498197555542,
    839            -14.518925666809082,
    840            0,
    841            0,
    842            -0.33243751525878906,
    843            -23.753076553344727,
    844            0,
    845            0,
    846            0,
    847            -2.213315725326538,
    848            -1.1077498197555542,
    849            -12.323249816894531,
    850            -1.1077498197555542,
    851            -14.518925666809082,
    852            0,
    853            0,
    854            -0.375,
    855            -7.139999866485596,
    856            0,
    857            0,
    858            -0.375,
    859            -5.7599992752075195,
    860            -1.5,
    861            -4.5,
    862            -1.5,
    863            -5.114999771118164
    864          ],
    865          'descriptor': {shape: [2, 2, 3, 4], dataType: 'float32'}
    866        }
    867      }
    868    }
    869  },
    870  {
    871    'name': 'gru float32 tensors steps=2 with all options',
    872    'graph': {
    873      'inputs': {
    874        'gruInput': {
    875          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
    876          'descriptor': {shape: [2, 3, 2], dataType: 'float32'}
    877        },
    878        'gruWeight': {
    879          'data': [
    880            1,   -1,   2, -2,  0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,
    881            0.5, -0.5, 0, 0.1, 1,   -1,   2, -2,  0.5, -0.5, 0, 0.1
    882          ],
    883          'descriptor': {shape: [1, 12, 2], dataType: 'float32'}
    884        },
    885        'gruRecurrentWeight': {
    886          'data': [
    887            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    888            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    889            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,
    890            0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1
    891          ],
    892          'descriptor': {shape: [1, 12, 4], dataType: 'float32'}
    893        },
    894        'gruBias': {
    895          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    896          'descriptor': {shape: [1, 12], dataType: 'float32'}
    897        },
    898        'gruRecurrentBias': {
    899          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
    900          'descriptor': {shape: [1, 12], dataType: 'float32'}
    901        },
    902        'gruInitialHiddenState': {
    903          'data': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
    904          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    905        }
    906      },
    907      'operators': [{
    908        'name': 'gru',
    909        'arguments': [
    910          {'input': 'gruInput'}, {'weight': 'gruWeight'},
    911          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
    912          {'hiddenSize': 4}, {
    913            'options': {
    914              'bias': 'gruBias',
    915              'recurrentBias': 'gruRecurrentBias',
    916              'initialHiddenState': 'gruInitialHiddenState',
    917              'resetAfter': false,
    918              'returnSequence': true,
    919              'direction': 'backward',
    920              'layout': 'zrn',
    921              'activations': ['relu', 'relu']
    922            }
    923          }
    924        ],
    925        'outputs': ['gruOutput1', 'gruOutput2']
    926      }],
    927      'expectedOutputs': {
    928        'gruOutput1': {
    929          'data': [
    930            0, 0, -0.24974998831748963, -18.59588623046875, -2.0657243728637697,
    931            -10.551867485046387, -1.3937838077545167, -15.2454833984375,
    932            -1.1589999198913575, -9.476999282836914, -1.1589999198913575,
    933            -11.319169044494629
    934          ],
    935          'descriptor': {shape: [1, 3, 4], dataType: 'float32'}
    936        },
    937        'gruOutput2': {
    938          'data': [
    939            0,
    940            0,
    941            -0.24974998831748963,
    942            -18.59588623046875,
    943            -2.0657243728637697,
    944            -10.551867485046387,
    945            -1.3937838077545167,
    946            -15.2454833984375,
    947            -1.1589999198913575,
    948            -9.476999282836914,
    949            -1.1589999198913575,
    950            -11.319169044494629,
    951            0,
    952            0,
    953            -0.25,
    954            -4.760000228881836,
    955            0,
    956            0,
    957            -0.25,
    958            -3.8399999141693117,
    959            -1,
    960            -3,
    961            -1,
    962            -3.4100000858306886
    963          ],
    964          'descriptor': {shape: [2, 1, 3, 4], dataType: 'float32'}
    965        }
    966      }
    967    }
    968  },
    969 
    970  // float16 tests
    971  {
    972    'name':
    973        "gru float16 tensors steps=1 with options.bias, options.recurrentBias and options.activations=['relu', 'relu']",
    974    'graph': {
    975      'inputs': {
    976        'gruInput': {
    977          'data': [1, 2, 2, 1, 1, 1],
    978          'descriptor': {shape: [1, 3, 2], dataType: 'float16'}
    979        },
    980        'gruWeight': {
    981          'data': [
    982            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
    983            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
    984            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
    985          ],
    986          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
    987        },
    988        'gruRecurrentWeight': {
    989          'data': [
    990            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
    991            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
    992            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
    993            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
    994            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
    995            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
    996            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
    997            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
    998            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
    999            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1000            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1001            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1002          ],
   1003          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1004        },
   1005        'gruBias': {
   1006          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1007          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1008        },
   1009        'gruRecurrentBias': {
   1010          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1011          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1012        }
   1013      },
   1014      'operators': [{
   1015        'name': 'gru',
   1016        'arguments': [
   1017          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1018          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
   1019          {'hiddenSize': 4}, {
   1020            'options': {
   1021              'bias': 'gruBias',
   1022              'recurrentBias': 'gruRecurrentBias',
   1023              'resetAfter': false,
   1024              'activations': ['relu', 'relu']
   1025            }
   1026          }
   1027        ],
   1028        'outputs': ['gruOutput']
   1029      }],
   1030      'expectedOutputs': {
   1031        'gruOutput': {
   1032          'data': [
   1033            0, 0, -0.25, -3.83984375, -4, -15, -2.25, -3.41015625, -1, -3, -1,
   1034            -3.41015625
   1035          ],
   1036          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1037        }
   1038      }
   1039    }
   1040  },
   1041  {
   1042    'name':
   1043        "gru float16 tensors steps=1 with options.bias, options.recurrentBias and options.activations=['relu', 'relu'] and resetAfter=true",
   1044    'graph': {
   1045      'inputs': {
   1046        'gruInput': {
   1047          'data': [1, 2, 2, 1, 1, 1],
   1048          'descriptor': {shape: [1, 3, 2], dataType: 'float16'}
   1049        },
   1050        'gruWeight': {
   1051          'data': [
   1052            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1053            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1054            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1055          ],
   1056          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1057        },
   1058        'gruRecurrentWeight': {
   1059          'data': [
   1060            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1061            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1062            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1063            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1064            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1065            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1066            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1067            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1068            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1069            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1070            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1071            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1072          ],
   1073          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1074        },
   1075        'gruBias': {
   1076          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1077          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1078        },
   1079        'gruRecurrentBias': {
   1080          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1081          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1082        }
   1083      },
   1084      'operators': [{
   1085        'name': 'gru',
   1086        'arguments': [
   1087          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1088          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
   1089          {'hiddenSize': 4}, {
   1090            'options': {
   1091              'bias': 'gruBias',
   1092              'recurrentBias': 'gruRecurrentBias',
   1093              'resetAfter': true,
   1094              'activations': ['relu', 'relu']
   1095            }
   1096          }
   1097        ],
   1098        'outputs': ['gruOutput']
   1099      }],
   1100      'expectedOutputs': {
   1101        'gruOutput': {
   1102          'data': [
   1103            0, 0, -0.375, -5.7578125, -6, -22.5, -3.375, -5.11328125, -1.5,
   1104            -4.5, -1.5, -5.11328125
   1105          ],
   1106          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1107        }
   1108      }
   1109    }
   1110  },
   1111  {
   1112    'name':
   1113        "gru float16 tensors steps=1 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and explicit options.direction='forward'",
   1114    'graph': {
   1115      'inputs': {
   1116        'gruInput': {
   1117          'data': [1, 2, 2, 1, 1, 1],
   1118          'descriptor': {shape: [1, 3, 2], dataType: 'float16'}
   1119        },
   1120        'gruWeight': {
   1121          'data': [
   1122            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1123            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1124            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1125          ],
   1126          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1127        },
   1128        'gruRecurrentWeight': {
   1129          'data': [
   1130            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1131            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1132            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1133            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1134            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1135            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1136            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1137            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1138            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1139            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1140            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1141            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1142          ],
   1143          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1144        },
   1145        'gruBias': {
   1146          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1147          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1148        },
   1149        'gruRecurrentBias': {
   1150          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1151          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1152        }
   1153      },
   1154      'operators': [{
   1155        'name': 'gru',
   1156        'arguments': [
   1157          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1158          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
   1159          {'hiddenSize': 4}, {
   1160            'options': {
   1161              'bias': 'gruBias',
   1162              'recurrentBias': 'gruRecurrentBias',
   1163              'resetAfter': false,
   1164              'direction': 'forward',
   1165              'activations': ['relu', 'relu']
   1166            }
   1167          }
   1168        ],
   1169        'outputs': ['gruOutput']
   1170      }],
   1171      'expectedOutputs': {
   1172        'gruOutput': {
   1173          'data': [
   1174            0, 0, -0.25, -3.83984375, -4, -15, -2.25, -3.41015625, -1, -3, -1,
   1175            -3.41015625
   1176          ],
   1177          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1178        }
   1179      }
   1180    }
   1181  },
   1182  {
   1183    'name':
   1184        "gru float16 tensors steps=1 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and explicit options.layout='zrn'",
   1185    'graph': {
   1186      'inputs': {
   1187        'gruInput': {
   1188          'data': [1, 2, 2, 1, 1, 1],
   1189          'descriptor': {shape: [1, 3, 2], dataType: 'float16'}
   1190        },
   1191        'gruWeight': {
   1192          'data': [
   1193            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1194            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1195            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1196          ],
   1197          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1198        },
   1199        'gruRecurrentWeight': {
   1200          'data': [
   1201            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1202            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1203            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1204            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1205            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1206            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1207            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1208            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1209            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1210            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1211            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1212            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1213          ],
   1214          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1215        },
   1216        'gruBias': {
   1217          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1218          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1219        },
   1220        'gruRecurrentBias': {
   1221          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1222          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1223        }
   1224      },
   1225      'operators': [{
   1226        'name': 'gru',
   1227        'arguments': [
   1228          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1229          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
   1230          {'hiddenSize': 4}, {
   1231            'options': {
   1232              'bias': 'gruBias',
   1233              'recurrentBias': 'gruRecurrentBias',
   1234              'resetAfter': false,
   1235              'layout': 'zrn',
   1236              'activations': ['relu', 'relu']
   1237            }
   1238          }
   1239        ],
   1240        'outputs': ['gruOutput']
   1241      }],
   1242      'expectedOutputs': {
   1243        'gruOutput': {
   1244          'data': [
   1245            0, 0, -0.25, -3.83984375, -4, -15, -2.25, -3.41015625, -1, -3, -1,
   1246            -3.41015625
   1247          ],
   1248          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1249        }
   1250      }
   1251    }
   1252  },
   1253  {
   1254    'name':
   1255        "gru float16 tensors steps=1 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and options.layout='rzn'",
   1256    'graph': {
   1257      'inputs': {
   1258        'gruInput': {
   1259          'data': [1, 2, 2, 1, 1, 1],
   1260          'descriptor': {shape: [1, 3, 2], dataType: 'float16'}
   1261        },
   1262        'gruWeight': {
   1263          'data': [
   1264            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1265            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1266            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1267          ],
   1268          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1269        },
   1270        'gruRecurrentWeight': {
   1271          'data': [
   1272            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1273            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1274            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1275            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1276            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1277            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1278            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1279            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1280            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1281            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1282            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1283            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1284          ],
   1285          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1286        },
   1287        'gruBias': {
   1288          'data': [1, 1, 1, 1, 1, 2, 1, 2, 0.5, 0.5, 0.5, 0.5],
   1289          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1290        },
   1291        'gruRecurrentBias': {
   1292          'data': [1, 1, 1, 1, 1, 2, 1, 2, 0.5, 0.5, 0.5, 0.5],
   1293          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1294        }
   1295      },
   1296      'operators': [{
   1297        'name': 'gru',
   1298        'arguments': [
   1299          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1300          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
   1301          {'hiddenSize': 4}, {
   1302            'options': {
   1303              'bias': 'gruBias',
   1304              'recurrentBias': 'gruRecurrentBias',
   1305              'resetAfter': false,
   1306              'layout': 'rzn',
   1307              'activations': ['relu', 'relu']
   1308            }
   1309          }
   1310        ],
   1311        'outputs': ['gruOutput']
   1312      }],
   1313      'expectedOutputs': {
   1314        'gruOutput': {
   1315          'data': [
   1316            0, 0, -0.25, -3.83984375, -4, -15, -2.25, -3.41015625, -1, -3, -1,
   1317            -3.41015625
   1318          ],
   1319          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1320        }
   1321      }
   1322    }
   1323  },
   1324  {
   1325    'name':
   1326        "gru float16 tensors steps=1 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and options.initialHiddenState",
   1327    'graph': {
   1328      'inputs': {
   1329        'gruInput': {
   1330          'data': [1, 2, 2, 1, 1, 1],
   1331          'descriptor': {shape: [1, 3, 2], dataType: 'float16'}
   1332        },
   1333        'gruWeight': {
   1334          'data': [
   1335            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1336            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1337            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1338          ],
   1339          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1340        },
   1341        'gruRecurrentWeight': {
   1342          'data': [
   1343            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1344            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1345            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1346            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1347            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1348            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1349            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1350            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1351            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1352            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1353            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1354            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1355          ],
   1356          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1357        },
   1358        'gruBias': {
   1359          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1360          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1361        },
   1362        'gruRecurrentBias': {
   1363          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1364          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1365        },
   1366        'gruInitialHiddenState': {
   1367          'data': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
   1368          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1369        }
   1370      },
   1371      'operators': [{
   1372        'name': 'gru',
   1373        'arguments': [
   1374          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1375          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
   1376          {'hiddenSize': 4}, {
   1377            'options': {
   1378              'bias': 'gruBias',
   1379              'recurrentBias': 'gruRecurrentBias',
   1380              'initialHiddenState': 'gruInitialHiddenState',
   1381              'resetAfter': false,
   1382              'activations': ['relu', 'relu']
   1383            }
   1384          }
   1385        ],
   1386        'outputs': ['gruOutput']
   1387      }],
   1388      'expectedOutputs': {
   1389        'gruOutput': {
   1390          'data': [
   1391            0, 0, -0.25, -3.83984375, -4, -15, -2.25, -3.41015625, -1, -3, -1,
   1392            -3.41015625
   1393          ],
   1394          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1395        }
   1396      }
   1397    }
   1398  },
   1399  {
   1400    'name': 'gru float16 tensors steps=1 all options',
   1401    'graph': {
   1402      'inputs': {
   1403        'gruInput': {
   1404          'data': [1, 2, 2, 1, 1, 1],
   1405          'descriptor': {shape: [1, 3, 2], dataType: 'float16'}
   1406        },
   1407        'gruWeight': {
   1408          'data': [
   1409            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1410            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1411            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1412          ],
   1413          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1414        },
   1415        'gruRecurrentWeight': {
   1416          'data': [
   1417            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1418            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1419            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1420            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1421            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1422            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1423            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1424            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1425            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1426            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1427            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1428            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1429          ],
   1430          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1431        },
   1432        'gruBias': {
   1433          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1434          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1435        },
   1436        'gruRecurrentBias': {
   1437          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1438          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1439        },
   1440        'gruInitialHiddenState': {
   1441          'data': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
   1442          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1443        }
   1444      },
   1445      'operators': [{
   1446        'name': 'gru',
   1447        'arguments': [
   1448          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1449          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 1},
   1450          {'hiddenSize': 4}, {
   1451            'options': {
   1452              'bias': 'gruBias',
   1453              'recurrentBias': 'gruRecurrentBias',
   1454              'initialHiddenState': 'gruInitialHiddenState',
   1455              'resetAfter': false,
   1456              'returnSequence': true,
   1457              'direction': 'forward',
   1458              'layout': 'zrn',
   1459              'activations': ['relu', 'relu']
   1460            }
   1461          }
   1462        ],
   1463        'outputs': ['gruOutput1', 'gruOutput2']
   1464      }],
   1465      'expectedOutputs': {
   1466        'gruOutput1': {
   1467          'data': [
   1468            0, 0, -0.25, -3.83984375, -4, -15, -2.25, -3.41015625, -1, -3, -1,
   1469            -3.41015625
   1470          ],
   1471          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1472        },
   1473        'gruOutput2': {
   1474          'data': [
   1475            0, 0, -0.25, -3.83984375, -4, -15, -2.25, -3.41015625, -1, -3, -1,
   1476            -3.41015625
   1477          ],
   1478          'descriptor': {shape: [1, 1, 3, 4], dataType: 'float16'}
   1479        }
   1480      }
   1481    }
   1482  },
   1483  {
   1484    'name':
   1485        "gru float16 tensors steps=2 with options.bias, options.recurrentBias, options.activations=['relu', 'relu'] and options.direction='backward'",
   1486    'graph': {
   1487      'inputs': {
   1488        'gruInput': {
   1489          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
   1490          'descriptor': {shape: [2, 3, 2], dataType: 'float16'}
   1491        },
   1492        'gruWeight': {
   1493          'data': [
   1494            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1495            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1496            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1497          ],
   1498          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1499        },
   1500        'gruRecurrentWeight': {
   1501          'data': [
   1502            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1503            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1504            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1505            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1506            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1507            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1508            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1509            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1510            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1511            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1512            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1513            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1514          ],
   1515          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1516        },
   1517        'gruBias': {
   1518          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1519          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1520        },
   1521        'gruRecurrentBias': {
   1522          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1523          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1524        }
   1525      },
   1526      'operators': [{
   1527        'name': 'gru',
   1528        'arguments': [
   1529          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1530          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
   1531          {'hiddenSize': 4}, {
   1532            'options': {
   1533              'bias': 'gruBias',
   1534              'recurrentBias': 'gruRecurrentBias',
   1535              'resetAfter': false,
   1536              'direction': 'backward',
   1537              'activations': ['relu', 'relu']
   1538            }
   1539          }
   1540        ],
   1541        'outputs': ['gruOutput']
   1542      }],
   1543      'expectedOutputs': {
   1544        'gruOutput': {
   1545          'data': [
   1546            0, 0, -0.249755859375, -18.59375, -2.06640625, -10.5546875,
   1547            -1.39453125, -15.2421875, -1.1591796875, -9.4765625, -1.1591796875,
   1548            -11.3203125
   1549          ],
   1550          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1551        }
   1552      }
   1553    }
   1554  },
   1555  {
   1556    'name':
   1557        "gru float16 tensors steps=2 with options.bias, options.recurrentBias, options.direction='backward', options.activations=['relu', 'relu'] and explicit options.returnSequence=false",
   1558    'graph': {
   1559      'inputs': {
   1560        'gruInput': {
   1561          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
   1562          'descriptor': {shape: [2, 3, 2], dataType: 'float16'}
   1563        },
   1564        'gruWeight': {
   1565          'data': [
   1566            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1567            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1568            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1569          ],
   1570          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1571        },
   1572        'gruRecurrentWeight': {
   1573          'data': [
   1574            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1575            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1576            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1577            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1578            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1579            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1580            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1581            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1582            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1583            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1584            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1585            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1586          ],
   1587          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1588        },
   1589        'gruBias': {
   1590          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1591          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1592        },
   1593        'gruRecurrentBias': {
   1594          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1595          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1596        }
   1597      },
   1598      'operators': [{
   1599        'name': 'gru',
   1600        'arguments': [
   1601          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1602          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
   1603          {'hiddenSize': 4}, {
   1604            'options': {
   1605              'bias': 'gruBias',
   1606              'recurrentBias': 'gruRecurrentBias',
   1607              'resetAfter': false,
   1608              'returnSequence': false,
   1609              'direction': 'backward',
   1610              'activations': ['relu', 'relu']
   1611            }
   1612          }
   1613        ],
   1614        'outputs': ['gruOutput']
   1615      }],
   1616      'expectedOutputs': {
   1617        'gruOutput': {
   1618          'data': [
   1619            0, 0, -0.249755859375, -18.59375, -2.06640625, -10.5546875,
   1620            -1.39453125, -15.2421875, -1.1591796875, -9.4765625, -1.1591796875,
   1621            -11.3203125
   1622          ],
   1623          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1624        }
   1625      }
   1626    }
   1627  },
   1628  {
   1629    'name':
   1630        "gru float16 tensors steps=2 with options.bias, options.recurrentBias, options.direction='backward', options.activations=['relu', 'relu'] and options.returnSequence=true",
   1631    'graph': {
   1632      'inputs': {
   1633        'gruInput': {
   1634          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
   1635          'descriptor': {shape: [2, 3, 2], dataType: 'float16'}
   1636        },
   1637        'gruWeight': {
   1638          'data': [
   1639            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1640            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1641            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1642          ],
   1643          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1644        },
   1645        'gruRecurrentWeight': {
   1646          'data': [
   1647            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1648            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1649            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1650            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1651            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1652            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1653            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1654            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1655            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1656            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1657            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1658            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1659          ],
   1660          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1661        },
   1662        'gruBias': {
   1663          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1664          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1665        },
   1666        'gruRecurrentBias': {
   1667          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1668          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1669        }
   1670      },
   1671      'operators': [{
   1672        'name': 'gru',
   1673        'arguments': [
   1674          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1675          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
   1676          {'hiddenSize': 4}, {
   1677            'options': {
   1678              'bias': 'gruBias',
   1679              'recurrentBias': 'gruRecurrentBias',
   1680              'resetAfter': false,
   1681              'returnSequence': true,
   1682              'direction': 'backward',
   1683              'activations': ['relu', 'relu']
   1684            }
   1685          }
   1686        ],
   1687        'outputs': ['gruOutput1', 'gruOutput2']
   1688      }],
   1689      'expectedOutputs': {
   1690        'gruOutput1': {
   1691          'data': [
   1692            0, 0, -0.249755859375, -18.59375, -2.06640625, -10.5546875,
   1693            -1.39453125, -15.2421875, -1.1591796875, -9.4765625, -1.1591796875,
   1694            -11.3203125
   1695          ],
   1696          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1697        },
   1698        'gruOutput2': {
   1699          'data': [
   1700            0,
   1701            0,
   1702            -0.249755859375,
   1703            -18.59375,
   1704            -2.06640625,
   1705            -10.5546875,
   1706            -1.39453125,
   1707            -15.2421875,
   1708            -1.1591796875,
   1709            -9.4765625,
   1710            -1.1591796875,
   1711            -11.3203125,
   1712            0,
   1713            0,
   1714            -0.25,
   1715            -4.7578125,
   1716            0,
   1717            0,
   1718            -0.25,
   1719            -3.83984375,
   1720            -1,
   1721            -3,
   1722            -1,
   1723            -3.41015625
   1724          ],
   1725          'descriptor': {shape: [2, 1, 3, 4], dataType: 'float16'}
   1726        }
   1727      }
   1728    }
   1729  },
   1730  {
   1731    'name':
   1732        "gru float16 tensors steps=2 with options.bias, options.recurrentBias, options.direction='both' and options.returnSequence=true",
   1733    'graph': {
   1734      'inputs': {
   1735        'gruInput': {
   1736          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
   1737          'descriptor': {shape: [2, 3, 2], dataType: 'float16'}
   1738        },
   1739        'gruWeight': {
   1740          'data': [
   1741            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1742            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1743            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1744            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1745            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1746            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1747          ],
   1748          'descriptor': {shape: [2, 12, 2], dataType: 'float16'}
   1749        },
   1750        'gruRecurrentWeight': {
   1751          'data': [
   1752            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1753            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1754            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1755            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1756            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1757            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1758            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1759            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1760            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1761            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1762            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1763            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1764            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1765            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1766            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1767            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1768            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1769            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1770            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1771            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1772            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1773            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1774            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1775            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1776          ],
   1777          'descriptor': {shape: [2, 12, 4], dataType: 'float16'}
   1778        },
   1779        'gruBias': {
   1780          'data': [
   1781            1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5,
   1782            1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5
   1783          ],
   1784          'descriptor': {shape: [2, 12], dataType: 'float16'}
   1785        },
   1786        'gruRecurrentBias': {
   1787          'data': [
   1788            1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5,
   1789            1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5
   1790          ],
   1791          'descriptor': {shape: [2, 12], dataType: 'float16'}
   1792        }
   1793      },
   1794      'operators': [{
   1795        'name': 'gru',
   1796        'arguments': [
   1797          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1798          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
   1799          {'hiddenSize': 4}, {
   1800            'options': {
   1801              'bias': 'gruBias',
   1802              'recurrentBias': 'gruRecurrentBias',
   1803              'resetAfter': true,
   1804              'returnSequence': true,
   1805              'activations': ['relu', 'relu'],
   1806              'direction': 'both'
   1807            }
   1808          }
   1809        ],
   1810        'outputs': ['gruOutput1', 'gruOutput2']
   1811      }],
   1812      'expectedOutputs': {
   1813        'gruOutput1': {
   1814          'data': [
   1815            0,
   1816            0,
   1817            -0.33251953125,
   1818            -23.75,
   1819            0,
   1820            0,
   1821            0,
   1822            -2.21875,
   1823            -1.1083984375,
   1824            -12.328125,
   1825            -1.1083984375,
   1826            -14.5234375,
   1827            0,
   1828            0,
   1829            -0.28076171875,
   1830            -25.4375,
   1831            -1.705078125,
   1832            -9.28125,
   1833            -1.404296875,
   1834            -21.15625,
   1835            -1.1083984375,
   1836            -12.328125,
   1837            -1.1083984375,
   1838            -14.5234375
   1839          ],
   1840          'descriptor': {shape: [2, 3, 4], dataType: 'float16'}
   1841        },
   1842        'gruOutput2': {
   1843          'data': [
   1844            0,
   1845            0,
   1846            -0.375,
   1847            -5.7578125,
   1848            -6,
   1849            -22.5,
   1850            -3.375,
   1851            -5.11328125,
   1852            -1.5,
   1853            -4.5,
   1854            -1.5,
   1855            -5.11328125,
   1856            0,
   1857            0,
   1858            -0.28076171875,
   1859            -25.4375,
   1860            -1.705078125,
   1861            -9.28125,
   1862            -1.404296875,
   1863            -21.15625,
   1864            -1.1083984375,
   1865            -12.328125,
   1866            -1.1083984375,
   1867            -14.5234375,
   1868            0,
   1869            0,
   1870            -0.33251953125,
   1871            -23.75,
   1872            0,
   1873            0,
   1874            0,
   1875            -2.21875,
   1876            -1.1083984375,
   1877            -12.328125,
   1878            -1.1083984375,
   1879            -14.5234375,
   1880            0,
   1881            0,
   1882            -0.375,
   1883            -7.140625,
   1884            0,
   1885            0,
   1886            -0.375,
   1887            -5.7578125,
   1888            -1.5,
   1889            -4.5,
   1890            -1.5,
   1891            -5.11328125
   1892          ],
   1893          'descriptor': {shape: [2, 2, 3, 4], dataType: 'float16'}
   1894        }
   1895      }
   1896    }
   1897  },
   1898  {
   1899    'name': 'gru float16 tensors steps=2 with all options',
   1900    'graph': {
   1901      'inputs': {
   1902        'gruInput': {
   1903          'data': [1, 2, 2, 1, 1, 1, 3, 4, 1, 2, 1, 1],
   1904          'descriptor': {shape: [2, 3, 2], dataType: 'float16'}
   1905        },
   1906        'gruWeight': {
   1907          'data': [
   1908            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1909            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375,
   1910            1, -1, 2, -2, 0.5, -0.5, 0, 0.0999755859375
   1911          ],
   1912          'descriptor': {shape: [1, 12, 2], dataType: 'float16'}
   1913        },
   1914        'gruRecurrentWeight': {
   1915          'data': [
   1916            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1917            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1918            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1919            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1920            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1921            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1922            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1923            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1924            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1925            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1926            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375,
   1927            0.0999755859375, 0.0999755859375, 0.0999755859375, 0.0999755859375
   1928          ],
   1929          'descriptor': {shape: [1, 12, 4], dataType: 'float16'}
   1930        },
   1931        'gruBias': {
   1932          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1933          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1934        },
   1935        'gruRecurrentBias': {
   1936          'data': [1, 2, 1, 2, 1, 1, 1, 1, 0.5, 0.5, 0.5, 0.5],
   1937          'descriptor': {shape: [1, 12], dataType: 'float16'}
   1938        },
   1939        'gruInitialHiddenState': {
   1940          'data': [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
   1941          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1942        }
   1943      },
   1944      'operators': [{
   1945        'name': 'gru',
   1946        'arguments': [
   1947          {'input': 'gruInput'}, {'weight': 'gruWeight'},
   1948          {'recurrentWeight': 'gruRecurrentWeight'}, {'steps': 2},
   1949          {'hiddenSize': 4}, {
   1950            'options': {
   1951              'bias': 'gruBias',
   1952              'recurrentBias': 'gruRecurrentBias',
   1953              'initialHiddenState': 'gruInitialHiddenState',
   1954              'resetAfter': false,
   1955              'returnSequence': true,
   1956              'direction': 'backward',
   1957              'layout': 'zrn',
   1958              'activations': ['relu', 'relu']
   1959            }
   1960          }
   1961        ],
   1962        'outputs': ['gruOutput1', 'gruOutput2']
   1963      }],
   1964      'expectedOutputs': {
   1965        'gruOutput1': {
   1966          'data': [
   1967            0, 0, -0.249755859375, -18.59375, -2.06640625, -10.5546875,
   1968            -1.39453125, -15.2421875, -1.1591796875, -9.4765625, -1.1591796875,
   1969            -11.3203125
   1970          ],
   1971          'descriptor': {shape: [1, 3, 4], dataType: 'float16'}
   1972        },
   1973        'gruOutput2': {
   1974          'data': [
   1975            0,
   1976            0,
   1977            -0.249755859375,
   1978            -18.59375,
   1979            -2.06640625,
   1980            -10.5546875,
   1981            -1.39453125,
   1982            -15.2421875,
   1983            -1.1591796875,
   1984            -9.4765625,
   1985            -1.1591796875,
   1986            -11.3203125,
   1987            0,
   1988            0,
   1989            -0.25,
   1990            -4.7578125,
   1991            0,
   1992            0,
   1993            -0.25,
   1994            -3.83984375,
   1995            -1,
   1996            -3,
   1997            -1,
   1998            -3.41015625
   1999          ],
   2000          'descriptor': {shape: [2, 1, 3, 4], dataType: 'float16'}
   2001        }
   2002      }
   2003    }
   2004  }
   2005 ];
   2006 
   2007 webnn_conformance_test(
   2008    gruTests, buildAndExecuteGraph, getGruPrecisionTolerance);