netrans/bin/tf_ruler_db.json

[
 {
  "ruler_name": "const",
  "src_ops_alias": [
   "C"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [],
  "src_out_tensor": [
   "C:out0"
  ],
  "acu_lys_alias": [
   "variable"
  ],
  "src_acu_in_tensor_map": [],
  "src_acu_out_tensor_map": [
   [
    "C:out0",
    "variable:out0"
   ]
  ],
  "param_map": {
   "variable": {
    "shape": [
     "ORIGIN",
     "CODE",
     "self.shape_pick(tensor['C:out0'])"
    ],
    "is_scalar": [
     "BOOL",
     "CODE",
     "True if len(self.tensor_to_numpy(tensor['C:out0']).shape) == 0 else False "
    ],
    "type": [
     "STRING",
     "CODE",
     "self.dtype_pick(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "variable": {
    "data": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "fullconnect",
  "src_ops_alias": [
   "BiasAdd",
   "MatMul",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C_1:out0",
    "MatMul:in1"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "MatMul:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "fullconnect"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "fullconnect:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "fullconnect:out0"
   ]
  ],
  "param_map": {
   "fullconnect": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[-1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "fullconnect": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'], trans=[1, 0] if self.attr_pick(node['MatMul'], 'transpose_b', False) else [0, 1])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "len(self.shape_pick(tensor['C:out0'])) < 2 and self.attr_pick(node['MatMul'], 'transpose_a', False) == False"
 },
 {
  "ruler_name": "matmul_fc",
  "src_ops_alias": [
   "MatMul",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "MatMul:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "MatMul:in0"
   ]
  ],
  "src_out_tensor": [
   "MatMul:out0"
  ],
  "acu_lys_alias": [
   "fullconnect"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "fullconnect:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "MatMul:out0",
    "fullconnect:out0"
   ]
  ],
  "param_map": {
   "fullconnect": {
    "weights": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'], trans=[1, 0] if self.attr_pick(node['MatMul'], 'transpose_b', False) else [0, 1]).shape[1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ]
   }
  },
  "blob_map": {
   "fullconnect": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'], trans=[1, 0] if self.attr_pick(node['MatMul'], 'transpose_b', False) else [0, 1])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "self.attr_pick(node['MatMul'], 'transpose_a', False) == False"
 },
 {
  "ruler_name": "matmul_fc_1",
  "src_ops_alias": [
   "MatMul"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "MatMul:in0"
   ],
   [
    "I_1:out0",
    "MatMul:in1"
   ]
  ],
  "src_out_tensor": [
   "MatMul:out0"
  ],
  "acu_lys_alias": [
   "matmul"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "matmul:in0"
   ],
   [
    "I_1:out0",
    "matmul:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "MatMul:out0",
    "matmul:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "matmul": {
    "transpose_a": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['MatMul'], 'transpose_a', False)"
    ],
    "transpose_b": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['MatMul'], 'transpose_b', False)"
    ]
   }
  },
  "blob_map": {
   "matmul": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "fc_bias",
  "src_ops_alias": [
   "Add",
   "MatMul",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "MatMul:out0",
    "Add:in0"
   ],
   [
    "C_1:out0",
    "MatMul:in1"
   ],
   [
    "C:out0",
    "Add:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "MatMul:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "fullconnect"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "fullconnect:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "fullconnect:out0"
   ]
  ],
  "param_map": {
   "fullconnect": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[-1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "fullconnect": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'], trans=[1, 0] if self.attr_pick(node['MatMul'], 'transpose_b', False) else [0, 1])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "len(self.shape_pick(tensor['C:out0'])) < 2 and self.attr_pick(node['MatMul'], 'transpose_a', False) == False"
 },
 {
  "ruler_name": "matmul",
  "src_ops_alias": [
   "BatchMatMul"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "BatchMatMul:in0"
   ],
   [
    "I_1:out0",
    "BatchMatMul:in1"
   ]
  ],
  "src_out_tensor": [
   "BatchMatMul:out0"
  ],
  "acu_lys_alias": [
   "matmul"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "matmul:in0"
   ],
   [
    "I_1:out0",
    "matmul:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BatchMatMul:out0",
    "matmul:out0"
   ]
  ],
  "param_map": {
   "matmul": {
    "transpose_a": [
     "INT",
     "CODE",
     "self.attr_pick(node['BatchMatMul'], 'adj_x', False)"
    ],
    "transpose_b": [
     "INT",
     "CODE",
     "self.attr_pick(node['BatchMatMul'], 'adj_y', False)"
    ]
   }
  },
  "blob_map": {
   "matmul": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "batchmatmulv2_to_matmul",
  "src_ops_alias": [
   "BatchMatMulV2"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "BatchMatMulV2:in0"
   ],
   [
    "I_1:out0",
    "BatchMatMulV2:in1"
   ]
  ],
  "src_out_tensor": [
   "BatchMatMulV2:out0"
  ],
  "acu_lys_alias": [
   "matmul"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "matmul:in0"
   ],
   [
    "I_1:out0",
    "matmul:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BatchMatMulV2:out0",
    "matmul:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "matmul": {
    "transpose_a": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['BatchMatMulV2'], 'adj_x', False)"
    ],
    "transpose_b": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['BatchMatMulV2'], 'adj_y', False)"
    ]
   }
  },
  "blob_map": {
   "matmul": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "relu",
  "src_ops_alias": [
   "Relu"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Relu:in0"
   ]
  ],
  "src_out_tensor": [
   "Relu:out0"
  ],
  "acu_lys_alias": [
   "relu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "relu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Relu:out0",
    "relu:out0"
   ]
  ],
  "param_map": {
   "relu": {}
  },
  "blob_map": {
   "relu": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "elu",
  "src_ops_alias": [
   "Elu"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Elu:in0"
   ]
  ],
  "src_out_tensor": [
   "Elu:out0"
  ],
  "acu_lys_alias": [
   "elu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "elu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Elu:out0",
    "elu:out0"
   ]
  ],
  "param_map": {
   "elu": {}
  },
  "blob_map": {
   "elu": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "quant_relu",
  "src_ops_alias": [
   "FakeQuantWithMinMaxVars",
   "Relu",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "Relu:out0",
    "FakeQuantWithMinMaxVars:in0"
   ],
   [
    "C_1:out0",
    "FakeQuantWithMinMaxVars:in2"
   ],
   [
    "C:out0",
    "FakeQuantWithMinMaxVars:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Relu:in0"
   ]
  ],
  "src_out_tensor": [
   "FakeQuantWithMinMaxVars:out0"
  ],
  "acu_lys_alias": [
   "relu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "relu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FakeQuantWithMinMaxVars:out0",
    "relu:out0"
   ]
  ],
  "param_map": {
   "relu": {}
  },
  "blob_map": {
   "relu": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "sigmoid",
  "src_ops_alias": [
   "Sigmoid"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Sigmoid:in0"
   ]
  ],
  "src_out_tensor": [
   "Sigmoid:out0"
  ],
  "acu_lys_alias": [
   "sigmoid"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "sigmoid:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Sigmoid:out0",
    "sigmoid:out0"
   ]
  ],
  "param_map": {
   "sigmoid": {}
  },
  "blob_map": {
   "sigmoid": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "relu6",
  "src_ops_alias": [
   "Relu6"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Relu6:in0"
   ]
  ],
  "src_out_tensor": [
   "Relu6:out0"
  ],
  "acu_lys_alias": [
   "relun"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "relun:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Relu6:out0",
    "relun:out0"
   ]
  ],
  "param_map": {
   "relun": {
    "relu_clamp_top": [
     "INT",
     "VALUE",
     6
    ]
   }
  },
  "blob_map": {
   "relun": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "quant_relu6",
  "src_ops_alias": [
   "FakeQuantWithMinMaxVars",
   "Relu6",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "Relu6:out0",
    "FakeQuantWithMinMaxVars:in0"
   ],
   [
    "C_1:out0",
    "FakeQuantWithMinMaxVars:in2"
   ],
   [
    "C:out0",
    "FakeQuantWithMinMaxVars:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Relu6:in0"
   ]
  ],
  "src_out_tensor": [
   "FakeQuantWithMinMaxVars:out0"
  ],
  "acu_lys_alias": [
   "relun"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "relun:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FakeQuantWithMinMaxVars:out0",
    "relun:out0"
   ]
  ],
  "param_map": {
   "relun": {
    "relu_clamp_top": [
     "INT",
     "VALUE",
     6
    ]
   }
  },
  "blob_map": {
   "relun": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "relun",
  "src_ops_alias": [
   "Minimum",
   "Relu",
   "C"
  ],
  "src_inter_flow": [
   [
    "Relu:out0",
    "Minimum:in0"
   ],
   [
    "C:out0",
    "Minimum:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Relu:in0"
   ]
  ],
  "src_out_tensor": [
   "Minimum:out0"
  ],
  "acu_lys_alias": [
   "relun"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "relun:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Minimum:out0",
    "relun:out0"
   ]
  ],
  "param_map": {
   "relun": {
    "relu_clamp_top": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']).tolist()[0]"
    ]
   }
  },
  "blob_map": {
   "relun": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "leakyrelu",
  "src_ops_alias": [
   "Maximum",
   "Mul",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Mul:in0"
   ],
   [
    "Mul:out0",
    "Maximum:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Maximum:in1"
   ],
   [
    "I:out0",
    "Mul:in1"
   ]
  ],
  "src_out_tensor": [
   "Maximum:out0"
  ],
  "acu_lys_alias": [
   "leakyrelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "leakyrelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Maximum:out0",
    "leakyrelu:out0"
   ]
  ],
  "param_map": {
   "leakyrelu": {
    "leaky_ratio": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "leakyrelu": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "self.tensor_to_numpy(tensor['C:out0']).ndim == 0 and self.tensor_to_numpy(tensor['C:out0']).dtype.name == 'float32'"
 },
 {
  "ruler_name": "gelu_no_approximate",
  "src_ops_alias": [
   "Mul",
   "Mul_1",
   "C",
   "Add",
   "C_1",
   "Erf",
   "RealDiv",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Mul_1:in0"
   ],
   [
    "Add:out0",
    "Mul_1:in1"
   ],
   [
    "C_1:out0",
    "Add:in0"
   ],
   [
    "Erf:out0",
    "Add:in1"
   ],
   [
    "RealDiv:out0",
    "Erf:in0"
   ],
   [
    "C_2:out0",
    "RealDiv:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "RealDiv:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "gelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "gelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gelu": {
    "approximate": [
     "BOOL",
     "VALUE",
     false
    ]
   }
  },
  "blob_map": {
   "gelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "gelu_no_approximate_tf_2_50",
  "src_ops_alias": [
   "Mul",
   "Mul_1",
   "AddV2",
   "C",
   "C_1",
   "Erf",
   "RealDiv",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "AddV2:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "AddV2:in0"
   ],
   [
    "Erf:out0",
    "AddV2:in1"
   ],
   [
    "RealDiv:out0",
    "Erf:in0"
   ],
   [
    "C_2:out0",
    "RealDiv:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul_1:in1"
   ],
   [
    "I:out0",
    "RealDiv:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "gelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "gelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gelu": {
    "approximate": [
     "BOOL",
     "VALUE",
     false
    ]
   }
  },
  "blob_map": {
   "gelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_gelu_with_approximate",
  "src_ops_alias": [
   "Mul",
   "Mul_1",
   "Add",
   "C",
   "C_1",
   "Tanh",
   "Mul_2",
   "C_2",
   "Add_1",
   "Mul_3",
   "C_3",
   "Pow",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "Add:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "Add:in0"
   ],
   [
    "Tanh:out0",
    "Add:in1"
   ],
   [
    "Mul_2:out0",
    "Tanh:in0"
   ],
   [
    "C_2:out0",
    "Mul_2:in0"
   ],
   [
    "Add_1:out0",
    "Mul_2:in1"
   ],
   [
    "Mul_3:out0",
    "Add_1:in1"
   ],
   [
    "C_3:out0",
    "Mul_3:in0"
   ],
   [
    "Pow:out0",
    "Mul_3:in1"
   ],
   [
    "C_4:out0",
    "Pow:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul_1:in1"
   ],
   [
    "I:out0",
    "Pow:in0"
   ],
   [
    "I:out0",
    "Add_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "gelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "gelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gelu": {
    "approximate": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "gelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_gelu_with_approximate_2",
  "src_ops_alias": [
   "Mul",
   "Mul_1",
   "C",
   "AddV2",
   "C_1",
   "Tanh",
   "Mul_2",
   "C_2",
   "AddV2_1",
   "Mul_3",
   "C_3",
   "Pow",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Mul_1:in0"
   ],
   [
    "AddV2:out0",
    "Mul_1:in1"
   ],
   [
    "C_1:out0",
    "AddV2:in0"
   ],
   [
    "Tanh:out0",
    "AddV2:in1"
   ],
   [
    "Mul_2:out0",
    "Tanh:in0"
   ],
   [
    "C_2:out0",
    "Mul_2:in0"
   ],
   [
    "AddV2_1:out0",
    "Mul_2:in1"
   ],
   [
    "Mul_3:out0",
    "AddV2_1:in1"
   ],
   [
    "C_3:out0",
    "Mul_3:in0"
   ],
   [
    "Pow:out0",
    "Mul_3:in1"
   ],
   [
    "C_4:out0",
    "Pow:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "Pow:in0"
   ],
   [
    "I:out0",
    "AddV2_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "gelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "gelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gelu": {
    "approximate": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "gelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_gelu_with_approximate_3",
  "src_ops_alias": [
   "Mul",
   "Mul_1",
   "C",
   "Add",
   "C_1",
   "Tanh",
   "Mul_2",
   "C_2",
   "Add_1",
   "Mul_3",
   "C_3",
   "Pow",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Mul_1:in0"
   ],
   [
    "Add:out0",
    "Mul_1:in1"
   ],
   [
    "C_1:out0",
    "Add:in0"
   ],
   [
    "Tanh:out0",
    "Add:in1"
   ],
   [
    "Mul_2:out0",
    "Tanh:in0"
   ],
   [
    "C_2:out0",
    "Mul_2:in0"
   ],
   [
    "Add_1:out0",
    "Mul_2:in1"
   ],
   [
    "Mul_3:out0",
    "Add_1:in1"
   ],
   [
    "C_3:out0",
    "Mul_3:in0"
   ],
   [
    "Pow:out0",
    "Mul_3:in1"
   ],
   [
    "C_4:out0",
    "Pow:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "Pow:in0"
   ],
   [
    "I:out0",
    "Add_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "gelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "gelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gelu": {
    "approximate": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "gelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_gelu_with_approximate_tf_2_50",
  "src_ops_alias": [
   "Mul",
   "Mul_1",
   "AddV2",
   "C",
   "C_1",
   "Tanh",
   "Mul_2",
   "C_2",
   "AddV2_1",
   "Mul_3",
   "C_3",
   "Pow",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "AddV2:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "AddV2:in0"
   ],
   [
    "Tanh:out0",
    "AddV2:in1"
   ],
   [
    "Mul_2:out0",
    "Tanh:in0"
   ],
   [
    "C_2:out0",
    "Mul_2:in0"
   ],
   [
    "AddV2_1:out0",
    "Mul_2:in1"
   ],
   [
    "Mul_3:out0",
    "AddV2_1:in1"
   ],
   [
    "C_3:out0",
    "Mul_3:in0"
   ],
   [
    "Pow:out0",
    "Mul_3:in1"
   ],
   [
    "C_4:out0",
    "Pow:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul_1:in1"
   ],
   [
    "I:out0",
    "Pow:in0"
   ],
   [
    "I:out0",
    "AddV2_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "gelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "gelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gelu": {
    "approximate": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "gelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "minimum",
  "src_ops_alias": [
   "Minimum"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Minimum:in0"
   ],
   [
    "I_1:out0",
    "Minimum:in1"
   ]
  ],
  "src_out_tensor": [
   "Minimum:out0"
  ],
  "acu_lys_alias": [
   "minimum"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "minimum:in0"
   ],
   [
    "I_1:out0",
    "minimum:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Minimum:out0",
    "minimum:out0"
   ]
  ],
  "param_map": {
   "minimum": {}
  },
  "blob_map": {
   "minimum": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "leakyrelu_2",
  "src_ops_alias": [
   "Maximum",
   "Mul",
   "C"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Maximum:in1"
   ],
   [
    "C:out0",
    "Mul:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Maximum:in0"
   ],
   [
    "I:out0",
    "Mul:in1"
   ]
  ],
  "src_out_tensor": [
   "Maximum:out0"
  ],
  "acu_lys_alias": [
   "leakyrelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "leakyrelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Maximum:out0",
    "leakyrelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "leakyrelu": {
    "leaky_ratio": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "leakyrelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "leakyrelu_3",
  "src_ops_alias": [
   "LeakyRelu"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "LeakyRelu:in0"
   ]
  ],
  "src_out_tensor": [
   "LeakyRelu:out0"
  ],
  "acu_lys_alias": [
   "leakyrelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "leakyrelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "LeakyRelu:out0",
    "leakyrelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "leakyrelu": {
    "leaky_ratio": [
     "FLOAT",
     "CODE",
     "self.attr_pick(node['LeakyRelu'], 'alpha')"
    ]
   }
  },
  "blob_map": {
   "leakyrelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "mulmax2leakyrelu",
  "src_ops_alias": [
   "Maximum",
   "Mul",
   "C"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Maximum:in0"
   ],
   [
    "C:out0",
    "Mul:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "Maximum:in1"
   ]
  ],
  "src_out_tensor": [
   "Maximum:out0"
  ],
  "acu_lys_alias": [
   "leakyrelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "leakyrelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Maximum:out0",
    "leakyrelu:out0"
   ]
  ],
  "param_map": {
   "leakyrelu": {
    "leaky_ratio": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "leakyrelu": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "abs_mul_2_leakeyrelu",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Mul_1",
   "C",
   "C_1",
   "Abs"
  ],
  "src_inter_flow": [
   [
    "Abs:out0",
    "Mul_1:in1"
   ],
   [
    "Mul_1:out0",
    "Add:in1"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C_1:out0",
    "Mul_1:in0"
   ],
   [
    "C:out0",
    "Mul:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in1"
   ],
   [
    "I:out0",
    "Abs:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "leakyrelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "leakyrelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "leakyrelu:out0"
   ]
  ],
  "param_map": {
   "leakyrelu": {
    "leaky_ratio": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']) - self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "blob_map": {
   "leakyrelu": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "math.isclose(self.tensor_to_numpy(tensor['C:out0']) + self.tensor_to_numpy(tensor['C_1:out0']), 1.0)"
 },
 {
  "ruler_name": "leaky_relu_4",
  "src_ops_alias": [
   "Add",
   "Relu",
   "Mul",
   "Mul_1",
   "C",
   "Sub",
   "C_1",
   "Abs"
  ],
  "src_inter_flow": [
   [
    "Relu:out0",
    "Add:in0"
   ],
   [
    "Mul:out0",
    "Add:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "C:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "Abs:out0",
    "Sub:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Relu:in0"
   ],
   [
    "I:out0",
    "Abs:in0"
   ],
   [
    "I:out0",
    "Sub:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "leakyrelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "leakyrelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "leakyrelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "leakyrelu": {
    "leaky_ratio": [
     "FLOAT",
     "CODE",
     "2 * self.tensor_to_numpy(tensor['C:out0'])[0] * self.tensor_to_numpy(tensor['C_1:out0'])[0] + 1"
    ]
   }
  },
  "blob_map": {
   "leakyrelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "quant_leaky_relu_4",
  "src_ops_alias": [
   "FakeQuantWithMinMaxVars",
   "Add",
   "C",
   "C_1",
   "Relu",
   "Mul",
   "Mul_1",
   "C_2",
   "Sub",
   "C_3",
   "Abs"
  ],
  "src_inter_flow": [
   [
    "Add:out0",
    "FakeQuantWithMinMaxVars:in0"
   ],
   [
    "C:out0",
    "FakeQuantWithMinMaxVars:in1"
   ],
   [
    "C_1:out0",
    "FakeQuantWithMinMaxVars:in2"
   ],
   [
    "Relu:out0",
    "Add:in0"
   ],
   [
    "Mul:out0",
    "Add:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "C_2:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "C_3:out0",
    "Mul_1:in1"
   ],
   [
    "Abs:out0",
    "Sub:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Relu:in0"
   ],
   [
    "I:out0",
    "Abs:in0"
   ],
   [
    "I:out0",
    "Sub:in0"
   ]
  ],
  "src_out_tensor": [
   "FakeQuantWithMinMaxVars:out0"
  ],
  "acu_lys_alias": [
   "leakyrelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "leakyrelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FakeQuantWithMinMaxVars:out0",
    "leakyrelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "leakyrelu": {
    "leaky_ratio": [
     "FLOAT",
     "CODE",
     "2 * self.tensor_to_numpy(tensor['C_2:out0'])[0] * self.tensor_to_numpy(tensor['C_3:out0'])[0] + 1"
    ]
   }
  },
  "blob_map": {
   "leakyrelu": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "tanh",
  "src_ops_alias": [
   "Tanh"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Tanh:in0"
   ]
  ],
  "src_out_tensor": [
   "Tanh:out0"
  ],
  "acu_lys_alias": [
   "tanh"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "tanh:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Tanh:out0",
    "tanh:out0"
   ]
  ],
  "blob_map": {
   "tanh": {}
  },
  "param_map": {
   "tanh": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "prelu",
  "src_ops_alias": [
   "Add",
   "Relu",
   "Mul",
   "Mul_1",
   "C",
   "C_1",
   "Sub",
   "Abs"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in1"
   ],
   [
    "C_1:out0",
    "Mul_1:in0"
   ],
   [
    "Abs:out0",
    "Sub:in1"
   ],
   [
    "Relu:out0",
    "Add:in0"
   ],
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "C:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Sub:in0"
   ],
   [
    "I:out0",
    "Abs:in0"
   ],
   [
    "I:out0",
    "Relu:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "prelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "prelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "prelu:out0"
   ]
  ],
  "param_map": {
   "prelu": {}
  },
  "blob_map": {
   "prelu": {
    "a": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_merge_to_prelu",
  "src_ops_alias": [
   "Sub",
   "Relu",
   "Mul",
   "C",
   "Relu_1",
   "Neg"
  ],
  "src_inter_flow": [
   [
    "Relu:out0",
    "Sub:in0"
   ],
   [
    "Mul:out0",
    "Sub:in1"
   ],
   [
    "C:out0",
    "Mul:in0"
   ],
   [
    "Relu_1:out0",
    "Mul:in1"
   ],
   [
    "Neg:out0",
    "Relu_1:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Relu:in0"
   ],
   [
    "I:out0",
    "Neg:in0"
   ]
  ],
  "src_out_tensor": [
   "Sub:out0"
  ],
  "acu_lys_alias": [
   "prelu"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "prelu:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Sub:out0",
    "prelu:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "prelu": {}
  },
  "blob_map": {
   "prelu": {
    "a": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "winograd_conv1d",
  "src_ops_alias": [
   "Conv",
   "ExpandDims",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "ExpandDims:out0",
    "Conv:in0"
   ],
   [
    "C:out0",
    "Conv:in1"
   ],
   [
    "C_1:out0",
    "ExpandDims:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ExpandDims:in0"
   ]
  ],
  "src_out_tensor": [
   "Conv:out0"
  ],
  "acu_lys_alias": [
   "conv1d_winograd"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "conv1d_winograd:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Conv:out0",
    "conv1d_winograd:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "conv1d_winograd": {
    "ksize": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[1]"
    ],
    "stride": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "conv1d_winograd": {
    "weight": [
     "CODE",
     "np.squeeze(self.tensor_to_numpy(tensor['C:out0']), 0)"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": "node_name == 'NETRANS_ATF_wino1d'"
 },
 {
  "ruler_name": "conv1d",
  "src_ops_alias": [
   "Squeeze",
   "Conv",
   "ExpandDims",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "Conv:out0",
    "Squeeze:in0"
   ],
   [
    "ExpandDims:out0",
    "Conv:in0"
   ],
   [
    "C:out0",
    "Conv:in1"
   ],
   [
    "C_1:out0",
    "ExpandDims:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ExpandDims:in0"
   ]
  ],
  "src_out_tensor": [
   "Squeeze:out0"
  ],
  "acu_lys_alias": [
   "conv1d"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "conv1d:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Squeeze:out0",
    "conv1d:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "conv1d": {
    "ksize": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[1]"
    ],
    "stride": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "conv1d": {
    "weight": [
     "CODE",
     "np.squeeze(self.tensor_to_numpy(tensor['C:out0']), 0)"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "conv1d_bias",
  "src_ops_alias": [
   "BiasAdd",
   "Squeeze",
   "C",
   "Conv",
   "ExpandDims",
   "C_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "Squeeze:out0",
    "BiasAdd:in0"
   ],
   [
    "ExpandDims:out0",
    "Conv:in0"
   ],
   [
    "C_2:out0",
    "ExpandDims:in1"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "Conv:out0",
    "Squeeze:in0"
   ],
   [
    "C_1:out0",
    "Conv:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ExpandDims:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "conv1d"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "conv1d:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "conv1d:out0"
   ]
  ],
  "param_map": {
   "conv1d": {
    "ksize": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"conv2d_weight\": \"C_1:out0\"}",
      "def r_conv2d_2_conv1d_ksize(self, node, tensor, conv2d_weight):",
      "    return max(self.shape_pick(tensor[conv2d_weight])[0], self.shape_pick(tensor[conv2d_weight])[1])",
      ""
     ]
    ],
    "stride": [
     "INT",
     "PYFUNC",
     [
      "def r_conv2d_2_conv1d_stride(self, node, tensor):",
      "    strides = self.attr_pick(node['Conv'], 'strides', [1,1,1,1])",
      "    if isinstance(strides, int):",
      "        strides = [1, strides, strides, 1]",
      "    else:",
      "        if len(strides) == 1:",
      "            strides = [1, strides[0], strides[0], 1]",
      "        elif len(strides) ==2:",
      "            strides = [1, strides[0], strides[1], 1]",
      "        elif len(strides) == 4:",
      "            pass",
      "        else:",
      "            print(\"Error: get unexpected strides '{}' from conv2d\".format(strides))",
      "            exit(-1)",
      "    return max(strides[1], strides[2])",
      ""
     ]
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "dilation": [
     "INTS",
     "PYFUNC",
     [
      "def r_conv2d_2_conv1d_dilation(self, node, tensor):",
      "    dilations = self.attr_pick(node['Conv'], 'dilations', [1,1,1,1])",
      "    if isinstance(dilations, int):",
      "        dilations = [1, dilations, dilations, 1]",
      "    else:",
      "        if len(dilations) == 1:",
      "            dilations = [1, dilations[0], dilations[0], 1]",
      "        elif len(dilations) ==2:",
      "            dilations = [1, dilations[0], dilations[1], 1]",
      "        elif len(dilations) == 4:",
      "            pass",
      "        else:",
      "            print(\"Error: get unexpected dilations '{}' from conv2d\".format(dilations))",
      "            exit(-1)",
      "    return [dilations[0], max(dilations[1], dilations[2]), dilations[3]]",
      ""
     ]
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[0]"
    ]
   }
  },
  "blob_map": {
   "conv1d": {
    "weight": [
     "CODE",
     "np.squeeze(self.tensor_to_numpy(tensor['C_1:out0']), 0)"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"conv2d_weight\": \"C_1:out0\"}",
   "def r_conv2d_2_conv1d_pre_condition(self, node, tensor, conv2d_weight):",
   "    #FIX ME: maybe we need more strict check here, such as check input is 1xN or Nx1",
   "    return self.shape_pick(tensor[conv2d_weight])[0] == 1 \\",
   "        or self.shape_pick(tensor[conv2d_weight])[1] == 1",
   ""
  ]
 },
 {
  "ruler_name": "winograd_conv2d",
  "src_ops_alias": [
   "Conv",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Conv:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv:in0"
   ]
  ],
  "src_out_tensor": [
   "Conv:out0"
  ],
  "acu_lys_alias": [
   "conv2d_winograd"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "conv2d_winograd:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Conv:out0",
    "conv2d_winograd:out0"
   ]
  ],
  "param_map": {
   "conv2d_winograd": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "pad": [
     "INTS",
     "CODE",
     "self.attr_pick(node['Conv'], 'explicit_paddings', [0,0,0,0])"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[3]"
    ],
    "wino_n": [
     "INT",
     "VALUE",
     2
    ]
   }
  },
  "blob_map": {
   "conv2d_winograd": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "node_name == 'NETRANS_ATF_wino2d'"
 },
 {
  "ruler_name": "single_convolution",
  "src_ops_alias": [
   "Conv",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Conv:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv:in0"
   ]
  ],
  "src_out_tensor": [
   "Conv:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Conv:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "pad": [
     "INTS",
     "CODE",
     "self.attr_pick(node['Conv'], 'explicit_paddings', [0,0,0,0])"
    ],
    "dilation": [
     "INTS",
     "CODE",
     "self.attr_pick(node['Conv'], 'dilations', [1,1,1,1])"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "pad_convolution",
  "src_ops_alias": [
   "Conv",
   "Pad",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Conv:in1"
   ],
   [
    "C_1:out0",
    "Pad:in1"
   ],
   [
    "Pad:out0",
    "Conv:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Pad:in0"
   ]
  ],
  "src_out_tensor": [
   "Conv:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Conv:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "pad_h": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])[1][0]"
    ],
    "pad_w": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])[2][0]"
    ],
    "pad_method": [
     "STRING",
     "VALUE",
     "padding_const"
    ],
    "pad": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])[1:3].flatten()"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "self.attr_pick(node['Conv'], 'padding') == 'VALID'"
 },
 {
  "ruler_name": "pad_convolution_badd",
  "src_ops_alias": [
   "BiasAdd",
   "Conv",
   "C",
   "Pad",
   "C_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "C_2:out0",
    "Pad:in1"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "Conv:out0",
    "BiasAdd:in0"
   ],
   [
    "C_1:out0",
    "Conv:in1"
   ],
   [
    "Pad:out0",
    "Conv:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Pad:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "pad_h": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])[1][0]"
    ],
    "pad_w": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])[2][0]"
    ],
    "pad_method": [
     "STRING",
     "VALUE",
     "padding_const"
    ],
    "pad": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])[1:3].flatten()"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "self.attr_pick(node['Conv'], 'padding') == 'VALID'"
 },
 {
  "ruler_name": "convolution_biasadd",
  "src_ops_alias": [
   "BiasAdd",
   "Conv",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "Conv:out0",
    "BiasAdd:in0"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "C_1:out0",
    "Conv:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "conv_badd_quant_w_quant_o",
  "src_ops_alias": [
   "FakeQuantWithMinMaxVars",
   "BiasAdd",
   "C",
   "C_1",
   "Conv",
   "C_2",
   "FakeQuantWithMinMaxVars_1",
   "C_3",
   "C_4",
   "C_5"
  ],
  "src_inter_flow": [
   [
    "C_5:out0",
    "FakeQuantWithMinMaxVars_1:in2"
   ],
   [
    "C_1:out0",
    "FakeQuantWithMinMaxVars:in2"
   ],
   [
    "FakeQuantWithMinMaxVars_1:out0",
    "Conv:in1"
   ],
   [
    "C_2:out0",
    "BiasAdd:in1"
   ],
   [
    "C_4:out0",
    "FakeQuantWithMinMaxVars_1:in1"
   ],
   [
    "C:out0",
    "FakeQuantWithMinMaxVars:in1"
   ],
   [
    "Conv:out0",
    "BiasAdd:in0"
   ],
   [
    "C_3:out0",
    "FakeQuantWithMinMaxVars_1:in0"
   ],
   [
    "BiasAdd:out0",
    "FakeQuantWithMinMaxVars:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv:in0"
   ]
  ],
  "src_out_tensor": [
   "FakeQuantWithMinMaxVars:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FakeQuantWithMinMaxVars:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "conv_badd_quant_w",
  "src_ops_alias": [
   "BiasAdd",
   "Conv",
   "C_2",
   "FakeQuantWithMinMaxVars_1",
   "C_3",
   "C_4",
   "C_5"
  ],
  "src_inter_flow": [
   [
    "C_5:out0",
    "FakeQuantWithMinMaxVars_1:in2"
   ],
   [
    "FakeQuantWithMinMaxVars_1:out0",
    "Conv:in1"
   ],
   [
    "C_2:out0",
    "BiasAdd:in1"
   ],
   [
    "C_4:out0",
    "FakeQuantWithMinMaxVars_1:in1"
   ],
   [
    "Conv:out0",
    "BiasAdd:in0"
   ],
   [
    "C_3:out0",
    "FakeQuantWithMinMaxVars_1:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "conv_add",
  "src_ops_alias": [
   "Add",
   "Conv",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C_1:out0",
    "Conv:in1"
   ],
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "Conv:out0",
    "Add:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "bias": [
     "CODE",
     "np.reshape(self.tensor_to_numpy(tensor['C:out0']), [-1])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "conv_bias",
  "src_ops_alias": [
   "Add",
   "Conv",
   "C",
   "FakeQuantWithMinMaxVars",
   "C_1",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "Conv:out0",
    "Add:in0"
   ],
   [
    "FakeQuantWithMinMaxVars:out0",
    "Conv:in1"
   ],
   [
    "C_1:out0",
    "FakeQuantWithMinMaxVars:in0"
   ],
   [
    "C_2:out0",
    "FakeQuantWithMinMaxVars:in1"
   ],
   [
    "C_3:out0",
    "FakeQuantWithMinMaxVars:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[3]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "bias": [
     "CODE",
     "np.reshape(self.tensor_to_numpy(tensor['C:out0']), [-1])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dwconv",
  "src_ops_alias": [
   "DepthwiseConv2dNative",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "DepthwiseConv2dNative:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "DepthwiseConv2dNative:in0"
   ]
  ],
  "src_out_tensor": [
   "DepthwiseConv2dNative:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "DepthwiseConv2dNative:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'padding')"
    ],
    "pad": [
     "INTS",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'],                                'explicit_paddings', [0,0,0,0,0,0,0,0])"
    ],
    "dilation": [
     "INTS",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'],                                 'dilations', [1,1,1,1])"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[2] * self.shape_pick(tensor['C:out0'])[3]"
    ],
    "group_number": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[2]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "np.reshape(self.tensor_to_numpy(tensor['C:out0']), [self.shape_pick(tensor['C:out0'])[0], self.shape_pick(tensor['C:out0'])[1], 1, -1])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dwconv_quant_w_quant_output",
  "src_ops_alias": [
   "DepthwiseConv2dNative",
   "FakeQuantWithMinMaxVars",
   "C_1",
   "C_2",
   "C_3",
   "FakeQuantWithMinMaxVars_1",
   "C_4",
   "C_5"
  ],
  "src_inter_flow": [
   [
    "C_3:out0",
    "FakeQuantWithMinMaxVars:in2"
   ],
   [
    "C_1:out0",
    "FakeQuantWithMinMaxVars:in0"
   ],
   [
    "FakeQuantWithMinMaxVars:out0",
    "DepthwiseConv2dNative:in1"
   ],
   [
    "C_2:out0",
    "FakeQuantWithMinMaxVars:in1"
   ],
   [
    "DepthwiseConv2dNative:out0",
    "FakeQuantWithMinMaxVars_1:in0"
   ],
   [
    "C_4:out0",
    "FakeQuantWithMinMaxVars_1:in1"
   ],
   [
    "C_5:out0",
    "FakeQuantWithMinMaxVars_1:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "DepthwiseConv2dNative:in0"
   ]
  ],
  "src_out_tensor": [
   "FakeQuantWithMinMaxVars_1:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FakeQuantWithMinMaxVars_1:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2] * self.shape_pick(tensor['C_1:out0'])[3]"
    ],
    "group_number": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "np.reshape(self.tensor_to_numpy(tensor['C_1:out0']), [self.shape_pick(tensor['C_1:out0'])[0], self.shape_pick(tensor['C_1:out0'])[1], 1, -1])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dwconv_add_quant_w",
  "src_ops_alias": [
   "Add",
   "DepthwiseConv2dNative",
   "C",
   "FakeQuantWithMinMaxVars",
   "C_1",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "C_3:out0",
    "FakeQuantWithMinMaxVars:in2"
   ],
   [
    "C_1:out0",
    "FakeQuantWithMinMaxVars:in0"
   ],
   [
    "FakeQuantWithMinMaxVars:out0",
    "DepthwiseConv2dNative:in1"
   ],
   [
    "DepthwiseConv2dNative:out0",
    "Add:in0"
   ],
   [
    "C_2:out0",
    "FakeQuantWithMinMaxVars:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "DepthwiseConv2dNative:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2] * self.shape_pick(tensor['C_1:out0'])[3]"
    ],
    "group_number": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "np.reshape(self.tensor_to_numpy(tensor['C_1:out0']), [self.shape_pick(tensor['C_1:out0'])[0], self.shape_pick(tensor['C_1:out0'])[1], 1, -1])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dw_bias",
  "src_ops_alias": [
   "BiasAdd",
   "DepthwiseConv2dNative",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "DepthwiseConv2dNative:out0",
    "BiasAdd:in0"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "C_1:out0",
    "DepthwiseConv2dNative:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "DepthwiseConv2dNative:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2] * self.shape_pick(tensor['C_1:out0'])[3]"
    ],
    "group_number": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "np.reshape(self.tensor_to_numpy(tensor['C_1:out0']), [self.shape_pick(tensor['C_1:out0'])[0], self.shape_pick(tensor['C_1:out0'])[1], 1, -1])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dwconv_quant_transposed_w",
  "src_ops_alias": [
   "DepthwiseConv2dNative",
   "Transpose",
   "FakeQuantWithMinMaxVarsPerChannel",
   "C",
   "C_1",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "Transpose:out0",
    "DepthwiseConv2dNative:in1"
   ],
   [
    "FakeQuantWithMinMaxVarsPerChannel:out0",
    "Transpose:in0"
   ],
   [
    "C:out0",
    "Transpose:in1"
   ],
   [
    "C_1:out0",
    "FakeQuantWithMinMaxVarsPerChannel:in0"
   ],
   [
    "C_2:out0",
    "FakeQuantWithMinMaxVarsPerChannel:in1"
   ],
   [
    "C_3:out0",
    "FakeQuantWithMinMaxVarsPerChannel:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "DepthwiseConv2dNative:in0"
   ]
  ],
  "src_out_tensor": [
   "DepthwiseConv2dNative:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "DepthwiseConv2dNative:out0",
    "convolution:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'],'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'],'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'],'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2] * self.shape_pick(tensor['C_1:out0'])[3]"
    ],
    "group_number": [
     "INT",
     "CODE",
     "np.transpose(self.tensor_to_numpy(tensor['C_1:out0']), self.tensor_to_numpy(tensor['C:out0'])).shape[2]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "np.reshape(np.transpose(self.tensor_to_numpy(tensor['C_1:out0']), self.tensor_to_numpy(tensor['C:out0'])), [self.shape_pick(tensor['C_1:out0'])[0], self.shape_pick(tensor['C_1:out0'])[1], 1, -1])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "pad_dwconv_quant_transposed_w",
  "src_ops_alias": [
   "DepthwiseConv2dNative",
   "Pad",
   "Transpose",
   "C",
   "FakeQuantWithMinMaxVarsPerChannel",
   "C_1",
   "C_2",
   "C_3",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Pad:out0",
    "DepthwiseConv2dNative:in0"
   ],
   [
    "Transpose:out0",
    "DepthwiseConv2dNative:in1"
   ],
   [
    "C:out0",
    "Pad:in1"
   ],
   [
    "FakeQuantWithMinMaxVarsPerChannel:out0",
    "Transpose:in0"
   ],
   [
    "C_1:out0",
    "Transpose:in1"
   ],
   [
    "C_2:out0",
    "FakeQuantWithMinMaxVarsPerChannel:in0"
   ],
   [
    "C_3:out0",
    "FakeQuantWithMinMaxVarsPerChannel:in1"
   ],
   [
    "C_4:out0",
    "FakeQuantWithMinMaxVarsPerChannel:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Pad:in0"
   ]
  ],
  "src_out_tensor": [
   "DepthwiseConv2dNative:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "DepthwiseConv2dNative:out0",
    "convolution:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'],'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'],'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'],'padding')"
    ],
    "pad_h": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])[1][0]"
    ],
    "pad_w": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])[2][0]"
    ],
    "pad_method": [
     "STRING",
     "VALUE",
     "padding_const"
    ],
    "pad": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])[1:3].flatten()"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[2] * self.shape_pick(tensor['C_2:out0'])[3]"
    ],
    "group_number": [
     "INT",
     "CODE",
     "np.transpose(self.tensor_to_numpy(tensor['C_2:out0']), self.tensor_to_numpy(tensor['C_1:out0'])).shape[2]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "np.reshape(np.transpose(self.tensor_to_numpy(tensor['C_2:out0']), self.tensor_to_numpy(tensor['C_1:out0'])), [self.shape_pick(tensor['C_2:out0'])[0], self.shape_pick(tensor['C_2:out0'])[1], 1, -1])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dilation_conv_bias",
  "src_ops_alias": [
   "BiasAdd",
   "BatchToSpaceND",
   "C",
   "Conv",
   "C_1",
   "C_2",
   "SpaceToBatchND",
   "C_3",
   "C_4",
   "C_5"
  ],
  "src_inter_flow": [
   [
    "C_5:out0",
    "SpaceToBatchND:in2"
   ],
   [
    "Conv:out0",
    "BatchToSpaceND:in0"
   ],
   [
    "C_2:out0",
    "BatchToSpaceND:in2"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "SpaceToBatchND:out0",
    "Conv:in0"
   ],
   [
    "C_1:out0",
    "BatchToSpaceND:in1"
   ],
   [
    "C_4:out0",
    "SpaceToBatchND:in1"
   ],
   [
    "C_3:out0",
    "Conv:in1"
   ],
   [
    "BatchToSpaceND:out0",
    "BiasAdd:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SpaceToBatchND:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding', 'SAME')"
    ],
    "pad_method": [
     "STRING",
     "VALUE",
     "padding_const"
    ],
    "pad": [
     "INTS",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"pad_name\": \"C_5:out0\", \"block_shape_name\": \"C_4:out0\", \"crop_name\": \"C_2:out0\"}",
      "def r_dilation_conv_pad(self, node, tensor, pad_name, block_shape_name, crop_name):",
      "    pads = self.tensor_to_numpy(tensor[pad_name])",
      "    block_shapes = self.tensor_to_numpy(tensor[block_shape_name])",
      "    crops = self.tensor_to_numpy(tensor[crop_name])",
      "    conv_pads = [[0,0],[0,0]]",
      "    for i in range(len(block_shapes)):",
      "        conv_pads[i][0] = pads[i][0] - crops[i][0]",
      "        conv_pads[i][1] = pads[i][1] - crops[i][1]",
      "    import numpy as np",
      "    conv_pads = np.asarray(conv_pads, np.int32)",
      "    return conv_pads.flatten()",
      ""
     ]
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[3]"
    ],
    "dilation": [
     "INTS",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"conv_name\": \"Conv\", \"block_shape_name\": \"C_4:out0\"}",
      "def r_dilation_conv_dilation(self, node, tensor, conv_name, block_shape_name):",
      "    #Not support dilation in the batch and depth dim",
      "    #TODO:maybe need add pre_condition",
      "    conv_d = self.attr_pick(node[conv_name], 'dilations')[1]",
      "    block_shape = self.tensor_to_numpy(tensor[block_shape_name])",
      "    if len(block_shape) == 1:",
      "        return [1, block_shape, block_shape, 1]  # nhwc",
      "    elif len(block_shape) == 2:",
      "        return [1, block_shape[0], block_shape[1], 1]  #nhwc",
      "    else:",
      "        raise ValueError('Unsupported block shape')",
      "",
      "    return [1, 1, 1, 1]",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"kernel_name\": \"C_3:out0\", \"conv_name\": \"Conv\", \"block_shape_name\": \"C_4:out0\"}",
   "def r_dilation_conv_pre_condition(self, node, tensor, kernel_name, conv_name, block_shape_name):",
   "    kernel_h = self.shape_pick(tensor[kernel_name])[0]",
   "    kernel_w = self.shape_pick(tensor[kernel_name])[1]",
   "    stride_h = self.attr_pick(node[conv_name], 'strides')[1]",
   "    stride_w = self.attr_pick(node[conv_name], 'strides')[2]",
   "    dilation = self.attr_pick(node[conv_name], 'dilations')[1]",
   "    block_shape = self.tensor_to_numpy(tensor[block_shape_name])[0]",
   "    dilation = dilation * block_shape",
   "",
   "    ker_h = (kernel_h - 1) * (dilation - 1) + kernel_h",
   "    ker_w = (kernel_w - 1) * (dilation - 1) + kernel_w",
   "",
   "    return (ker_h / stride_h < 16) and (ker_w / stride_w < 16)",
   ""
  ]
 },
 {
  "ruler_name": "pad_dialated_conv_bias",
  "src_ops_alias": [
   "Pad",
   "C_6",
   "BiasAdd",
   "BatchToSpaceND",
   "C",
   "Conv",
   "C_1",
   "C_2",
   "SpaceToBatchND",
   "C_3",
   "C_4",
   "C_5"
  ],
  "src_inter_flow": [
   [
    "Pad:out0",
    "SpaceToBatchND:in0"
   ],
   [
    "C_6:out0",
    "Pad:in1"
   ],
   [
    "C_5:out0",
    "SpaceToBatchND:in2"
   ],
   [
    "Conv:out0",
    "BatchToSpaceND:in0"
   ],
   [
    "C_2:out0",
    "BatchToSpaceND:in2"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "SpaceToBatchND:out0",
    "Conv:in0"
   ],
   [
    "C_1:out0",
    "BatchToSpaceND:in1"
   ],
   [
    "C_4:out0",
    "SpaceToBatchND:in1"
   ],
   [
    "C_3:out0",
    "Conv:in1"
   ],
   [
    "BatchToSpaceND:out0",
    "BiasAdd:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Pad:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding', 'SAME')"
    ],
    "pad_method": [
     "STRING",
     "VALUE",
     "padding_const"
    ],
    "pad": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_6:out0'])[1:3].flatten()"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[3]"
    ],
    "dilation": [
     "INTS",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"conv_name\": \"Conv\", \"block_shape_name\": \"C_4:out0\"}",
      "def r_dilation_conv_dilation(self, node, tensor, conv_name, block_shape_name):",
      "    #Not support dilation in the batch and depth dim",
      "    #TODO:maybe need add pre_condition",
      "    conv_d = self.attr_pick(node[conv_name], 'dilations')[1]",
      "    block_shape = self.tensor_to_numpy(tensor[block_shape_name])",
      "    if len(block_shape) == 1:",
      "        return [1, block_shape, block_shape, 1]  # nhwc",
      "    elif len(block_shape) == 2:",
      "        return [1, block_shape[0], block_shape[1], 1]  #nhwc",
      "    else:",
      "        raise ValueError('Unsupported block shape')",
      "",
      "    return [1, 1, 1, 1]",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"kernel_name\": \"C_3:out0\", \"conv_name\": \"Conv\", \"block_shape_name\": \"C_4:out0\"}",
   "def r_dilation_conv_pre_condition(self, node, tensor, kernel_name, conv_name, block_shape_name):",
   "    kernel_h = self.shape_pick(tensor[kernel_name])[0]",
   "    kernel_w = self.shape_pick(tensor[kernel_name])[1]",
   "    stride_h = self.attr_pick(node[conv_name], 'strides')[1]",
   "    stride_w = self.attr_pick(node[conv_name], 'strides')[2]",
   "    dilation = self.attr_pick(node[conv_name], 'dilations')[1]",
   "    block_shape = self.tensor_to_numpy(tensor[block_shape_name])[0]",
   "    dilation = dilation * block_shape",
   "",
   "    ker_h = (kernel_h - 1) * (dilation - 1) + kernel_h",
   "    ker_w = (kernel_w - 1) * (dilation - 1) + kernel_w",
   "",
   "    return (ker_h / stride_h < 16) and (ker_w / stride_w < 16)",
   ""
  ]
 },
 {
  "ruler_name": "dialated_conv",
  "src_ops_alias": [
   "BatchToSpaceND",
   "BiasAdd",
   "C",
   "C_1",
   "Conv",
   "C_2",
   "SpaceToBatchND",
   "C_3",
   "C_4",
   "C_5"
  ],
  "src_inter_flow": [
   [
    "C_3:out0",
    "Conv:in1"
   ],
   [
    "Conv:out0",
    "BiasAdd:in0"
   ],
   [
    "C:out0",
    "BatchToSpaceND:in1"
   ],
   [
    "BiasAdd:out0",
    "BatchToSpaceND:in0"
   ],
   [
    "SpaceToBatchND:out0",
    "Conv:in0"
   ],
   [
    "C_2:out0",
    "BiasAdd:in1"
   ],
   [
    "C_4:out0",
    "SpaceToBatchND:in1"
   ],
   [
    "C_1:out0",
    "BatchToSpaceND:in2"
   ],
   [
    "C_5:out0",
    "SpaceToBatchND:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SpaceToBatchND:in0"
   ]
  ],
  "src_out_tensor": [
   "BatchToSpaceND:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BatchToSpaceND:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "VALUE",
     "SAME"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[3]"
    ],
    "dilation": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0'])[0]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"kernel_name\": \"C_3:out0\", \"conv_name\": \"Conv\", \"block_shape_name\": \"C_4:out0\"}",
   "def r_dilation_conv_pre_condition(self, node, tensor, kernel_name, conv_name, block_shape_name):",
   "    kernel_h = self.shape_pick(tensor[kernel_name])[0]",
   "    kernel_w = self.shape_pick(tensor[kernel_name])[1]",
   "    stride_h = self.attr_pick(node[conv_name], 'strides')[1]",
   "    stride_w = self.attr_pick(node[conv_name], 'strides')[2]",
   "    dilation = self.attr_pick(node[conv_name], 'dilations')[1]",
   "    block_shape = self.tensor_to_numpy(tensor[block_shape_name])[0]",
   "    dilation = dilation * block_shape",
   "",
   "    ker_h = (kernel_h - 1) * (dilation - 1) + kernel_h",
   "    ker_w = (kernel_w - 1) * (dilation - 1) + kernel_w",
   "",
   "    return (ker_h / stride_h < 16) and (ker_w / stride_w < 16)",
   ""
  ]
 },
 {
  "ruler_name": "dilated_conv_no_biasadd",
  "src_ops_alias": [
   "BatchToSpaceND",
   "Conv",
   "C",
   "C_1",
   "SpaceToBatchND",
   "C_2",
   "C_3",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Conv:out0",
    "BatchToSpaceND:in0"
   ],
   [
    "C:out0",
    "BatchToSpaceND:in1"
   ],
   [
    "C_1:out0",
    "BatchToSpaceND:in2"
   ],
   [
    "SpaceToBatchND:out0",
    "Conv:in0"
   ],
   [
    "C_2:out0",
    "Conv:in1"
   ],
   [
    "C_3:out0",
    "SpaceToBatchND:in1"
   ],
   [
    "C_4:out0",
    "SpaceToBatchND:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SpaceToBatchND:in0"
   ]
  ],
  "src_out_tensor": [
   "BatchToSpaceND:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BatchToSpaceND:out0",
    "convolution:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding', 'SAME')"
    ],
    "pad_method": [
     "STRING",
     "VALUE",
     "padding_const"
    ],
    "pad": [
     "INTS",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"pad_name\": \"C_4:out0\", \"block_shape_name\": \"C_3:out0\", \"crop_name\": \"C_1:out0\"}",
      "def r_dilation_conv_pad(self, node, tensor, pad_name, block_shape_name, crop_name):",
      "    pads = self.tensor_to_numpy(tensor[pad_name])",
      "    block_shapes = self.tensor_to_numpy(tensor[block_shape_name])",
      "    crops = self.tensor_to_numpy(tensor[crop_name])",
      "    conv_pads = [[0,0],[0,0]]",
      "    for i in range(len(block_shapes)):",
      "        conv_pads[i][0] = pads[i][0] - crops[i][0]",
      "        conv_pads[i][1] = pads[i][1] - crops[i][1]",
      "    import numpy as np",
      "    conv_pads = np.asarray(conv_pads, np.int32)",
      "    return conv_pads.flatten()",
      ""
     ]
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[3]"
    ],
    "dilation": [
     "INTS",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"conv_name\": \"Conv\", \"block_shape_name\": \"C_3:out0\"}",
      "def r_dilation_conv_dilation(self, node, tensor, conv_name, block_shape_name):",
      "    #Not support dilation in the batch and depth dim",
      "    #TODO:maybe need add pre_condition",
      "    conv_d = self.attr_pick(node[conv_name], 'dilations')[1]",
      "    block_shape = self.tensor_to_numpy(tensor[block_shape_name])",
      "    if len(block_shape) == 1:",
      "        return [1, block_shape, block_shape, 1]  # nhwc",
      "    elif len(block_shape) == 2:",
      "        return [1, block_shape[0], block_shape[1], 1]  #nhwc",
      "    else:",
      "        raise ValueError('Unsupported block shape')",
      "",
      "    return [1, 1, 1, 1]",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"kernel_name\": \"C_2:out0\", \"conv_name\": \"Conv\", \"block_shape_name\": \"C_3:out0\"}",
   "def r_dilation_conv_pre_condition(self, node, tensor, kernel_name, conv_name, block_shape_name):",
   "    kernel_h = self.shape_pick(tensor[kernel_name])[0]",
   "    kernel_w = self.shape_pick(tensor[kernel_name])[1]",
   "    stride_h = self.attr_pick(node[conv_name], 'strides')[1]",
   "    stride_w = self.attr_pick(node[conv_name], 'strides')[2]",
   "    dilation = self.attr_pick(node[conv_name], 'dilations')[1]",
   "    block_shape = self.tensor_to_numpy(tensor[block_shape_name])[0]",
   "    dilation = dilation * block_shape",
   "",
   "    ker_h = (kernel_h - 1) * (dilation - 1) + kernel_h",
   "    ker_w = (kernel_w - 1) * (dilation - 1) + kernel_w",
   "",
   "    return (ker_h / stride_h < 16) and (ker_w / stride_w < 16)",
   ""
  ]
 },
 {
  "ruler_name": "dialated_conv1d",
  "src_ops_alias": [
   "BatchToSpaceND",
   "Squeeze",
   "C",
   "C_1",
   "Conv",
   "ExpandDims",
   "C_2",
   "SpaceToBatchND",
   "C_3",
   "C_4",
   "C_5"
  ],
  "src_inter_flow": [
   [
    "Squeeze:out0",
    "BatchToSpaceND:in0"
   ],
   [
    "C:out0",
    "BatchToSpaceND:in1"
   ],
   [
    "C_1:out0",
    "BatchToSpaceND:in2"
   ],
   [
    "Conv:out0",
    "Squeeze:in0"
   ],
   [
    "ExpandDims:out0",
    "Conv:in0"
   ],
   [
    "C_2:out0",
    "Conv:in1"
   ],
   [
    "SpaceToBatchND:out0",
    "ExpandDims:in0"
   ],
   [
    "C_3:out0",
    "ExpandDims:in1"
   ],
   [
    "C_4:out0",
    "SpaceToBatchND:in1"
   ],
   [
    "C_5:out0",
    "SpaceToBatchND:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SpaceToBatchND:in0"
   ]
  ],
  "src_out_tensor": [
   "BatchToSpaceND:out0"
  ],
  "acu_lys_alias": [
   "conv1d"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "conv1d:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BatchToSpaceND:out0",
    "conv1d:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "conv1d": {
    "ksize": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[1]"
    ],
    "stride": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[3]"
    ],
    "dilation": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0'])[0]"
    ]
   }
  },
  "blob_map": {
   "conv1d": {
    "weight": [
     "CODE",
     "np.squeeze(self.tensor_to_numpy(tensor['C_2:out0']), 0)"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dialated_conv1d_bias",
  "src_ops_alias": [
   "BiasAdd",
   "BatchToSpaceND",
   "C",
   "Squeeze",
   "C_1",
   "C_2",
   "Conv",
   "ExpandDims",
   "C_3",
   "SpaceToBatchND",
   "C_4",
   "C_5",
   "C_6"
  ],
  "src_inter_flow": [
   [
    "BatchToSpaceND:out0",
    "BiasAdd:in0"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "Squeeze:out0",
    "BatchToSpaceND:in0"
   ],
   [
    "C_1:out0",
    "BatchToSpaceND:in1"
   ],
   [
    "C_2:out0",
    "BatchToSpaceND:in2"
   ],
   [
    "Conv:out0",
    "Squeeze:in0"
   ],
   [
    "ExpandDims:out0",
    "Conv:in0"
   ],
   [
    "C_3:out0",
    "Conv:in1"
   ],
   [
    "SpaceToBatchND:out0",
    "ExpandDims:in0"
   ],
   [
    "C_4:out0",
    "ExpandDims:in1"
   ],
   [
    "C_5:out0",
    "SpaceToBatchND:in1"
   ],
   [
    "C_6:out0",
    "SpaceToBatchND:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SpaceToBatchND:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "conv1d"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "conv1d:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "conv1d:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "conv1d": {
    "ksize": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[1]"
    ],
    "stride": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_3:out0'])[3]"
    ],
    "dilation": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_5:out0'])[0]"
    ]
   }
  },
  "blob_map": {
   "conv1d": {
    "weight": [
     "CODE",
     "np.squeeze(self.tensor_to_numpy(tensor['C_3:out0']), 0)"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dialated_deptwise",
  "src_ops_alias": [
   "BatchToSpaceND",
   "DepthwiseConv2dNative",
   "C",
   "C_1",
   "SpaceToBatchND",
   "C_2",
   "C_3",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "DepthwiseConv2dNative:out0",
    "BatchToSpaceND:in0"
   ],
   [
    "C_4:out0",
    "SpaceToBatchND:in2"
   ],
   [
    "C_1:out0",
    "BatchToSpaceND:in2"
   ],
   [
    "C_2:out0",
    "DepthwiseConv2dNative:in1"
   ],
   [
    "C_3:out0",
    "SpaceToBatchND:in1"
   ],
   [
    "C:out0",
    "BatchToSpaceND:in1"
   ],
   [
    "SpaceToBatchND:out0",
    "DepthwiseConv2dNative:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SpaceToBatchND:in0"
   ]
  ],
  "src_out_tensor": [
   "BatchToSpaceND:out0"
  ],
  "acu_lys_alias": [
   "convolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "convolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BatchToSpaceND:out0",
    "convolution:out0"
   ]
  ],
  "param_map": {
   "convolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthwiseConv2dNative'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "VALUE",
     "SAME"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[3] * self.shape_pick(tensor['C_2:out0'])[2]"
    ],
    "dilation": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])[0]"
    ],
    "group_number": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[2]"
    ]
   }
  },
  "blob_map": {
   "convolution": {
    "weight": [
     "CODE",
     "np.reshape( self.tensor_to_numpy(tensor['C_2:out0']), [self.shape_pick(tensor['C_2:out0'])[0],self.shape_pick(tensor['C_2:out0'])[1], 1, -1])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "single_deconvolution",
  "src_ops_alias": [
   "DConv",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C_1:out0",
    "DConv:in1"
   ],
   [
    "C:out0",
    "DConv:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "DConv:in2"
   ]
  ],
  "src_out_tensor": [
   "DConv:out0"
  ],
  "acu_lys_alias": [
   "deconvolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "deconvolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "DConv:out0",
    "deconvolution:out0"
   ]
  ],
  "param_map": {
   "deconvolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['DConv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['DConv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['DConv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2]"
    ],
    "output_shape": [
     "INTS",
     "CODE",
     "self.deconv_output_shape(self.tensor_to_numpy(tensor['C:out0']))"
    ]
   }
  },
  "blob_map": {
   "deconvolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dconvolution_biasadd",
  "src_ops_alias": [
   "BiasAdd",
   "DConv",
   "C",
   "C_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "C_2:out0",
    "DConv:in1"
   ],
   [
    "DConv:out0",
    "BiasAdd:in0"
   ],
   [
    "C_1:out0",
    "DConv:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "DConv:in2"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "deconvolution"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "deconvolution:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "deconvolution:out0"
   ]
  ],
  "param_map": {
   "deconvolution": {
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[0]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['DConv'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['DConv'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['DConv'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[2]"
    ],
    "output_shape": [
     "INTS",
     "CODE",
     "self.deconv_output_shape(self.tensor_to_numpy(tensor['C_1:out0']))"
    ]
   }
  },
  "blob_map": {
   "deconvolution": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "deconvolution3d",
  "src_ops_alias": [
   "Conv3DBackpropInputV2",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Conv3DBackpropInputV2:in0"
   ],
   [
    "C_1:out0",
    "Conv3DBackpropInputV2:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv3DBackpropInputV2:in2"
   ]
  ],
  "src_out_tensor": [
   "Conv3DBackpropInputV2:out0"
  ],
  "acu_lys_alias": [
   "deconvolution3d"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "deconvolution3d:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Conv3DBackpropInputV2:out0",
    "deconvolution3d:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "deconvolution3d": {
    "ksize_d": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2]"
    ],
    "stride_d": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv3DBackpropInputV2'], 'strides')[0]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv3DBackpropInputV2'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv3DBackpropInputV2'], 'strides')[2]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv3DBackpropInputV2'], 'padding')"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[3]"
    ],
    "output_shape": [
     "INTS",
     "CODE",
     "self.deconv_output_shape(self.tensor_to_numpy(tensor['C:out0']))"
    ]
   }
  },
  "blob_map": {
   "deconvolution3d": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "conv3d_no_bias",
  "src_ops_alias": [
   "Conv3D",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Conv3D:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv3D:in0"
   ]
  ],
  "src_out_tensor": [
   "Conv3D:out0"
  ],
  "acu_lys_alias": [
   "conv3d"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "conv3d:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Conv3D:out0",
    "conv3d:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "conv3d": {
    "ksize_d": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[0]"
    ],
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[1]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[2]"
    ],
    "stride_d": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv3D'], 'strides')[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv3D'], 'strides')[2]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv3D'], 'strides')[3]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv3D'], 'padding')"
    ],
    "pad_method": [
     "STRING",
     "VALUE",
     "auto"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[-1]"
    ]
   }
  },
  "blob_map": {
   "conv3d": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "conv3d_with_bias",
  "src_ops_alias": [
   "BiasAdd",
   "Conv3D",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "Conv3D:out0",
    "BiasAdd:in0"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "C_1:out0",
    "Conv3D:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Conv3D:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "conv3d"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "conv3d:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "conv3d:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "conv3d": {
    "ksize_d": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[0]"
    ],
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2]"
    ],
    "stride_d": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv3D'], 'strides')[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv3D'], 'strides')[2]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['Conv3D'], 'strides')[3]"
    ],
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Conv3D'], 'padding')"
    ],
    "pad_method": [
     "STRING",
     "VALUE",
     "auto"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[-1]"
    ]
   }
  },
  "blob_map": {
   "conv3d": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "lrn",
  "src_ops_alias": [
   "LRN"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "LRN:in0"
   ]
  ],
  "src_out_tensor": [
   "LRN:out0"
  ],
  "acu_lys_alias": [
   "localresponsenormalization_tf"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "localresponsenormalization_tf:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "LRN:out0",
    "localresponsenormalization_tf:out0"
   ]
  ],
  "param_map": {
   "localresponsenormalization_tf": {
    "local_size": [
     "INT",
     "CODE",
     "self.attr_pick(node['LRN'], 'depth_radius') * 2 + 1"
    ],
    "bias": [
     "FLOAT",
     "CODE",
     "self.attr_pick(node['LRN'], 'bias', 1.0)"
    ],
    "alpha": [
     "FLOAT",
     "CODE",
     "self.attr_pick(node['LRN'], 'alpha', 1e-4)"
    ],
    "beta": [
     "FLOAT",
     "CODE",
     "self.attr_pick(node['LRN'], 'beta', 0.75)"
    ],
    "type": [
     "STRING",
     "VALUE",
     "NORM_ACROSS_CHANNELS"
    ]
   }
  },
  "blob_map": {
   "localresponsenormalization_tf": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "l2normalize",
  "src_ops_alias": [
   "Mul",
   "Rsqrt",
   "Maximum",
   "Sum",
   "C",
   "Square",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "Rsqrt:out0",
    "Mul:in1"
   ],
   [
    "Maximum:out0",
    "Rsqrt:in0"
   ],
   [
    "Sum:out0",
    "Maximum:in0"
   ],
   [
    "C:out0",
    "Maximum:in1"
   ],
   [
    "Square:out0",
    "Sum:in0"
   ],
   [
    "C_1:out0",
    "Sum:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "Square:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "l2normalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "l2normalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "l2normalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "l2normalize": {
    "l2n_dim": [
     "INT_OR_INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {},
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "l2normalize_l",
  "src_ops_alias": [
   "RealDiv",
   "Sqrt",
   "Add",
   "C",
   "Sum",
   "C_1",
   "Pow",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "Sqrt:out0",
    "RealDiv:in1"
   ],
   [
    "Add:out0",
    "Sqrt:in0"
   ],
   [
    "Sum:out0",
    "Add:in0"
   ],
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "Pow:out0",
    "Sum:in0"
   ],
   [
    "C_1:out0",
    "Sum:in1"
   ],
   [
    "C_2:out0",
    "Pow:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "RealDiv:in0"
   ],
   [
    "I:out0",
    "Pow:in0"
   ]
  ],
  "src_out_tensor": [
   "RealDiv:out0"
  ],
  "acu_lys_alias": [
   "l2normalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "l2normalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "RealDiv:out0",
    "l2normalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "l2normalize": {
    "l2n_dim": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {},
  "priority_tip": 0,
  "pre_condition": "self.tensor_to_numpy(tensor['C_2:out0']) == 2"
 },
 {
  "ruler_name": "l2normscale",
  "src_ops_alias": [
   "Mul",
   "Mul_1",
   "C",
   "Rsqrt",
   "Maximum",
   "Sum",
   "C_1",
   "Square",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "C:out0",
    "Mul:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in1"
   ],
   [
    "Maximum:out0",
    "Rsqrt:in0"
   ],
   [
    "Sum:out0",
    "Maximum:in0"
   ],
   [
    "C_1:out0",
    "Maximum:in1"
   ],
   [
    "Square:out0",
    "Sum:in0"
   ],
   [
    "C_2:out0",
    "Sum:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul_1:in0"
   ],
   [
    "I:out0",
    "Square:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "l2normalizescale"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "l2normalizescale:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "l2normalizescale:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "l2normalizescale": {
    "l2n_dim": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "blob_map": {
   "l2normalizescale": {
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "fusebatchnorm",
  "src_ops_alias": [
   "FusedBatchNorm",
   "C",
   "C_1",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "C_3:out0",
    "FusedBatchNorm:in4"
   ],
   [
    "C_2:out0",
    "FusedBatchNorm:in3"
   ],
   [
    "C:out0",
    "FusedBatchNorm:in1"
   ],
   [
    "C_1:out0",
    "FusedBatchNorm:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "FusedBatchNorm:in0"
   ]
  ],
  "src_out_tensor": [
   "FusedBatchNorm:out0"
  ],
  "acu_lys_alias": [
   "batchnormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "batchnormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FusedBatchNorm:out0",
    "batchnormalize:out0"
   ]
  ],
  "param_map": {
   "batchnormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.attr_pick(node['FusedBatchNorm'], 'epsilon', 1e-4)"
    ]
   }
  },
  "blob_map": {
   "batchnormalize": {
    "gamma": [
     "CODE",
     "None if len(self.tensor_to_numpy(tensor['C:out0'])) == 0 else self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "beta": [
     "CODE",
     "None if len(self.tensor_to_numpy(tensor['C_1:out0'])) == 0 else self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "mean": [
     "CODE",
     "None if len(self.tensor_to_numpy(tensor['C_2:out0'])) == 0 else self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "variance": [
     "CODE",
     "None if len(self.tensor_to_numpy(tensor['C_3:out0'])) == 0 else self.tensor_to_numpy(tensor['C_3:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "fusebatchnormv3",
  "src_ops_alias": [
   "FusedBatchNormV3",
   "C",
   "C_1",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "C_3:out0",
    "FusedBatchNormV3:in4"
   ],
   [
    "C_2:out0",
    "FusedBatchNormV3:in3"
   ],
   [
    "C:out0",
    "FusedBatchNormV3:in1"
   ],
   [
    "C_1:out0",
    "FusedBatchNormV3:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "FusedBatchNormV3:in0"
   ]
  ],
  "src_out_tensor": [
   "FusedBatchNormV3:out0"
  ],
  "acu_lys_alias": [
   "batchnormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "batchnormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FusedBatchNormV3:out0",
    "batchnormalize:out0"
   ]
  ],
  "param_map": {
   "batchnormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.attr_pick(node['FusedBatchNormV3'],'epsilon', 1e-4)"
    ]
   }
  },
  "blob_map": {
   "batchnormalize": {
    "gamma": [
     "CODE",
     "None if len(self.tensor_to_numpy(tensor['C:out0'])) == 0 else self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "beta": [
     "CODE",
     "None if len(self.tensor_to_numpy(tensor['C_1:out0'])) == 0 else self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "mean": [
     "CODE",
     "None if len(self.tensor_to_numpy(tensor['C_2:out0'])) == 0 else self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "variance": [
     "CODE",
     "None if len(self.tensor_to_numpy(tensor['C_3:out0'])) == 0 else self.tensor_to_numpy(tensor['C_3:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "fusebatchnorm_as_instancenorm",
  "src_ops_alias": [
   "FusedBatchNorm",
   "C",
   "C_1",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "C_3:out0",
    "FusedBatchNorm:in4"
   ],
   [
    "C_2:out0",
    "FusedBatchNorm:in3"
   ],
   [
    "C:out0",
    "FusedBatchNorm:in1"
   ],
   [
    "C_1:out0",
    "FusedBatchNorm:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "FusedBatchNorm:in0"
   ]
  ],
  "src_out_tensor": [
   "FusedBatchNorm:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FusedBatchNorm:out0",
    "instancenormalize:out0"
   ]
  ],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.attr_pick(node['FusedBatchNorm'], 'epsilon', 1e-4)"
    ],
    "axis": [
     "INTS",
     "VALUE",
     [
      0,
      1,
      2
     ]
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 1,
  "pre_condition": "self.attr_pick(node['FusedBatchNorm'], 'is_training', True) == True and len(self.tensor_to_numpy(tensor['C_2:out0'])) == 0 and len(self.tensor_to_numpy(tensor['C_3:out0'])) == 0"
 },
 {
  "ruler_name": "mul_add_2_bn",
  "src_ops_alias": [
   "Add",
   "Mul",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C_1:out0",
    "Mul:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "batchnormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "batchnormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "batchnormalize:out0"
   ]
  ],
  "param_map": {
   "batchnormalize": {
    "eps": [
     "FLOAT",
     "VALUE",
     1e-05
    ]
   }
  },
  "blob_map": {
   "batchnormalize": {
    "beta": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "gamma": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "len(self.shape_pick(tensor['C:out0'])) == 1 and self.shape_pick(tensor['C:out0'])[0] == self.shape_pick(tensor['I:out0'])[-1]and len(self.shape_pick(tensor['C_1:out0'])) == 1 and self.shape_pick(tensor['C_1:out0'])[0] == self.shape_pick(tensor['C:out0'])[0]"
 },
 {
  "ruler_name": "r_instancenorm_tf_1_13_scale_true_center_true",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Sub",
   "Mul_1",
   "C",
   "Mul_2",
   "Rsqrt",
   "C_1",
   "Mean",
   "Add_1",
   "C_2",
   "Mean_1",
   "C_3",
   "SquaredDifference",
   "C_4",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ],
   [
    "Mean:out0",
    "Mul_2:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C_3:out0",
    "Add_1:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "Sub:out0",
    "Add:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "VALUE",
     1e-05
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_4:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "instance_normalize",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Sub",
   "Mul_1",
   "C",
   "Mul_2",
   "Rsqrt",
   "C_1",
   "Add_1",
   "Add_2",
   "Mul_3",
   "Reshape",
   "Sub_1",
   "C_2",
   "Sum",
   "C_3",
   "StopGradient",
   "C_4",
   "Mul_4",
   "Square",
   "Sub_2",
   "C_5",
   "Mean",
   "Sum_1",
   "C_6",
   "SquaredDifference",
   "C_7"
  ],
  "src_inter_flow": [
   [
    "Sum_1:out0",
    "Mul_4:in0"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "Sub_1:out0",
    "Add_2:in0"
   ],
   [
    "Sum:out0",
    "Mul_3:in0"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Add_2:out0",
    "Rsqrt:in0"
   ],
   [
    "Sub:out0",
    "Add:in1"
   ],
   [
    "C_4:out0",
    "Reshape:in1"
   ],
   [
    "C_6:out0",
    "Mean:in1"
   ],
   [
    "Reshape:out0",
    "Add_1:in1"
   ],
   [
    "Mul_4:out0",
    "Sub_1:in0"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C_5:out0",
    "Sum:in1"
   ],
   [
    "StopGradient:out0",
    "Sub_2:in1"
   ],
   [
    "SquaredDifference:out0",
    "Sum_1:in0"
   ],
   [
    "C_3:out0",
    "Mul_4:in1"
   ],
   [
    "Mul_3:out0",
    "Add_1:in0"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ],
   [
    "C_7:out0",
    "Sum_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "Sub_2:out0",
    "Sum:in0"
   ],
   [
    "C_2:out0",
    "Add_2:in1"
   ],
   [
    "Mul_3:out0",
    "Square:in0"
   ],
   [
    "Add_1:out0",
    "Mul_2:in0"
   ],
   [
    "Square:out0",
    "Sub_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "StopGradient:out0",
    "Reshape:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "C_3:out0",
    "Mul_3:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "Sub_2:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_6:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_6:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "instance_norm_2",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Sub",
   "Mul_1",
   "C",
   "Mul_2",
   "Rsqrt",
   "C_1",
   "Squeeze",
   "Add_1",
   "Mean",
   "Squeeze_1",
   "C_2",
   "C_3",
   "Mean_1",
   "SquaredDifference",
   "C_4",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Sub:out0",
    "Add:in1"
   ],
   [
    "Mean_1:out0",
    "Squeeze_1:in0"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "C_3:out0",
    "Mean:in1"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ],
   [
    "Squeeze_1:out0",
    "Add_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "Squeeze:out0",
    "Mul_2:in0"
   ],
   [
    "C_2:out0",
    "Add_1:in1"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "Squeeze:in0"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_3:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "instancenormalize",
  "src_ops_alias": [
   "Add",
   "Mul",
   "C",
   "C_1",
   "RealDiv",
   "Sub",
   "Pow",
   "Mean",
   "Add_1",
   "C_2",
   "C_3",
   "Mean_1",
   "C_4",
   "SquaredDifference",
   "C_5"
  ],
  "src_inter_flow": [
   [
    "C_4:out0",
    "Add_1:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_3:out0",
    "Mean:in1"
   ],
   [
    "C_1:out0",
    "Mul:in0"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "Add_1:out0",
    "Pow:in0"
   ],
   [
    "RealDiv:out0",
    "Mul:in1"
   ],
   [
    "C_2:out0",
    "Pow:in1"
   ],
   [
    "Sub:out0",
    "RealDiv:in0"
   ],
   [
    "Mean:out0",
    "SquaredDifference:in1"
   ],
   [
    "Pow:out0",
    "RealDiv:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_5:out0",
    "Mean_1:in1"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Sub:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_3:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "r_instancenorm_4",
  "src_ops_alias": [
   "Add",
   "Mul",
   "C",
   "C_1",
   "RealDiv",
   "Sub",
   "Pow",
   "Mean",
   "Add_1",
   "C_2",
   "C_3",
   "Mean_1",
   "C_4",
   "SquaredDifference",
   "C_5",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "C_1:out0",
    "Mul:in0"
   ],
   [
    "RealDiv:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "RealDiv:in0"
   ],
   [
    "Pow:out0",
    "RealDiv:in1"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ],
   [
    "Add_1:out0",
    "Pow:in0"
   ],
   [
    "C_2:out0",
    "Pow:in1"
   ],
   [
    "C_3:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_4:out0",
    "Add_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_5:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Sub:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_3:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "r_instancenorm_5",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Sub",
   "Mul_1",
   "C",
   "Mul_2",
   "Rsqrt",
   "C_1",
   "Mean",
   "Add_1",
   "C_2",
   "Mean_1",
   "C_3",
   "SquaredDifference",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Sub:out0",
    "Add:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "Mean:out0",
    "Mul_2:in0"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_3:out0",
    "Add_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "Mean:out0",
    "SquaredDifference:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_4:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "r_instancenorm_6",
  "src_ops_alias": [
   "AddV2",
   "Mul",
   "Sub",
   "Mul_1",
   "C",
   "Mul_2",
   "Rsqrt",
   "C_1",
   "Mean",
   "AddV2_1",
   "C_2",
   "Mean_1",
   "C_3",
   "SquaredDifference",
   "C_4",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "AddV2:in0"
   ],
   [
    "Sub:out0",
    "AddV2:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "Mean:out0",
    "Mul_2:in0"
   ],
   [
    "AddV2_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "AddV2_1:in0"
   ],
   [
    "C_3:out0",
    "AddV2_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ]
  ],
  "src_out_tensor": [
   "AddV2:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "AddV2:out0",
    "instancenormalize:out0"
   ]
  ],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_4:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "r_instancenorm_7",
  "src_ops_alias": [
   "AddV2",
   "Mul",
   "C",
   "C_1",
   "RealDiv",
   "Sub",
   "Sqrt",
   "Mean",
   "AddV2_1",
   "C_2",
   "Mean_1",
   "C_3",
   "SquaredDifference",
   "C_4",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "AddV2:in0"
   ],
   [
    "C:out0",
    "AddV2:in1"
   ],
   [
    "C_1:out0",
    "Mul:in0"
   ],
   [
    "RealDiv:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "RealDiv:in0"
   ],
   [
    "Sqrt:out0",
    "RealDiv:in1"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ],
   [
    "AddV2_1:out0",
    "Sqrt:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "AddV2_1:in0"
   ],
   [
    "C_3:out0",
    "AddV2_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Sub:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ]
  ],
  "src_out_tensor": [
   "AddV2:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"input_name\": \"I:out0\", \"bias_name\": \"C:out0\"}",
      "def r_instancenorm_get_bias(self, node, tensor, input_name, bias_name):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    if len(bias.shape) == 1:",
      "        bias = bias * np.ones([input_shape[-1]]).astype(bias.dtype)",
      "    return bias",
      ""
     ]
    ],
    "scale": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"input_name\": \"I:out0\", \"scale_name\": \"C_1:out0\"}",
      "def r_instancenorm_get_scale(self, node, tensor, input_name, scale_name):",
      "    import numpy as np",
      "    scale = self.tensor_to_numpy(tensor[scale_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_rank = len(input_shape)",
      "    if len(scale.shape) == 1:",
      "        scale = scale * np.ones([input_shape[-1]]).astype(scale.dtype)",
      "    return scale",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_4:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "r_quant_instancenorm_7",
  "src_ops_alias": [
   "FakeQuantWithMinMaxVars",
   "AddV2",
   "C",
   "C_1",
   "Mul",
   "C_2",
   "C_3",
   "RealDiv",
   "Sub",
   "Sqrt",
   "Mean",
   "AddV2_1",
   "C_4",
   "Mean_1",
   "C_5",
   "SquaredDifference",
   "C_6",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "AddV2:out0",
    "FakeQuantWithMinMaxVars:in0"
   ],
   [
    "C:out0",
    "FakeQuantWithMinMaxVars:in1"
   ],
   [
    "C_1:out0",
    "FakeQuantWithMinMaxVars:in2"
   ],
   [
    "Mul:out0",
    "AddV2:in0"
   ],
   [
    "C_2:out0",
    "AddV2:in1"
   ],
   [
    "C_3:out0",
    "Mul:in0"
   ],
   [
    "RealDiv:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "RealDiv:in0"
   ],
   [
    "Sqrt:out0",
    "RealDiv:in1"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ],
   [
    "AddV2_1:out0",
    "Sqrt:in0"
   ],
   [
    "C_4:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "AddV2_1:in0"
   ],
   [
    "C_5:out0",
    "AddV2_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_6:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Sub:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ]
  ],
  "src_out_tensor": [
   "FakeQuantWithMinMaxVars:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FakeQuantWithMinMaxVars:out0",
    "instancenormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_5:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_6:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"input_name\": \"I:out0\", \"bias_name\": \"C_2:out0\"}",
      "def r_instancenorm_get_bias(self, node, tensor, input_name, bias_name):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    if len(bias.shape) == 1:",
      "        bias = bias * np.ones([input_shape[-1]]).astype(bias.dtype)",
      "    return bias",
      ""
     ]
    ],
    "scale": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"input_name\": \"I:out0\", \"scale_name\": \"C_3:out0\"}",
      "def r_instancenorm_get_scale(self, node, tensor, input_name, scale_name):",
      "    import numpy as np",
      "    scale = self.tensor_to_numpy(tensor[scale_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_rank = len(input_shape)",
      "    if len(scale.shape) == 1:",
      "        scale = scale * np.ones([input_shape[-1]]).astype(scale.dtype)",
      "    return scale",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_6:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "r_instancenorm_tf_1_13_scale_false_center_false",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Mul_1",
   "Rsqrt",
   "Neg",
   "Add_1",
   "Mean",
   "Mean_1",
   "C",
   "C_1",
   "SquaredDifference",
   "C_2",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in1"
   ],
   [
    "Neg:out0",
    "Mul_1:in0"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "Mean:out0",
    "Neg:in0"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C:out0",
    "Add_1:in1"
   ],
   [
    "C_1:out0",
    "Mean:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_2:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0']).tolist()"
    ]
   }
  },
  "blob_map": {},
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_1:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "r_instancenorm_tf_1_13_scale_true_center_false",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Mul_1",
   "Mul_2",
   "Neg",
   "Rsqrt",
   "C",
   "Mean",
   "Add_1",
   "C_1",
   "Mean_1",
   "C_2",
   "SquaredDifference",
   "C_3",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in1"
   ],
   [
    "Mul_2:out0",
    "Mul:in1"
   ],
   [
    "Mul_2:out0",
    "Mul_1:in1"
   ],
   [
    "Neg:out0",
    "Mul_1:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul_2:in0"
   ],
   [
    "C:out0",
    "Mul_2:in1"
   ],
   [
    "Mean:out0",
    "Neg:in0"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_1:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_2:out0",
    "Add_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_3:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_1:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "r_instancenorm_tf_1_13_scale_false_center_true",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Sub",
   "Rsqrt",
   "C",
   "Mul_1",
   "Add_1",
   "Mean",
   "Mean_1",
   "C_1",
   "C_2",
   "SquaredDifference",
   "C_3",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Sub:out0",
    "Add:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ],
   [
    "Mul_1:out0",
    "Sub:in1"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in1"
   ],
   [
    "Mean:out0",
    "Mul_1:in0"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_1:out0",
    "Add_1:in1"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_3:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "instancenormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "instancenormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "instancenormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "instancenormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "instancenormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_names\": [\"I:out0\", \"C_2:out0\"]}",
   "def r_instancenorm_pre_condition(self, node, tensor, input_names):",
   "    input_shape = self.shape_pick(tensor[input_names[0]])",
   "    axis = self.tensor_to_numpy(tensor[input_names[1]]).tolist()",
   "    # TODO:maybe need refine the condition",
   "    #return ((len(input_shape) == 4 or len(input_shape) == 3) and axis == [1,2])\\",
   "    #       or (len(input_shape) == 2 and axis == [0,1])",
   "    return (len(input_shape) == 4 and axis == [1,2]) or \\",
   "           (len(input_shape) == 4 and axis == [0,1,2] and input_shape[0] == 1)",
   ""
  ]
 },
 {
  "ruler_name": "layer_norm",
  "src_ops_alias": [
   "Add",
   "Mul",
   "C",
   "Mul_1",
   "C_1",
   "Sub",
   "Rsqrt",
   "Mean",
   "Add_1",
   "C_2",
   "Mean_1",
   "C_3",
   "Square",
   "C_4",
   "Sub_1"
  ],
  "src_inter_flow": [
   [
    "Sub_1:out0",
    "Square:in0"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "C_3:out0",
    "Add_1:in1"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "Square:out0",
    "Mean_1:in0"
   ],
   [
    "Mean:out0",
    "Sub_1:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in1"
   ],
   [
    "C_1:out0",
    "Mul:in1"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Sub:in0"
   ],
   [
    "I:out0",
    "Sub_1:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "layernormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "layernormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "layernormalize:out0"
   ]
  ],
  "param_map": {
   "layernormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ]
   }
  },
  "blob_map": {
   "layernormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"axis_tensor_name\": \"C_2:out0\", \"input_tensor_name\": \"I:out0\"}",
   "def r_layernorm_pre_condition(self, node, tensor, axis_tensor_name, input_tensor_name):",
   "    ''' the axis list only does not contain the batch dim '''",
   "    axis = self.tensor_to_numpy(tensor[axis_tensor_name])",
   "    _shape = list(self.shape_pick(tensor[input_tensor_name]))",
   "    input_dims = len(_shape)",
   "    if axis.shape == 0:  # for scalar",
   "        axis = [axis]",
   "    need_axis = list(range(input_dims))",
   "    need_axis.pop(0)  # no batch dim",
   "    return list(axis) == need_axis",
   ""
  ]
 },
 {
  "ruler_name": "layer_norm_sqrdiff",
  "src_ops_alias": [
   "Add",
   "Mul",
   "C",
   "Mul_1",
   "C_1",
   "Sub",
   "Rsqrt",
   "Mean",
   "Add_1",
   "C_2",
   "Mean_1",
   "C_3",
   "SquaredDifference",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "C_1:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in1"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_3:out0",
    "Add_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "Mean:out0",
    "SquaredDifference:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "Sub:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "layernormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "layernormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "layernormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "layernormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ]
   }
  },
  "blob_map": {
   "layernormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"axis_tensor_name\": \"C_2:out0\", \"input_tensor_name\": \"I:out0\"}",
   "def r_layernorm_pre_condition(self, node, tensor, axis_tensor_name, input_tensor_name):",
   "    ''' the axis list only does not contain the batch dim '''",
   "    axis = self.tensor_to_numpy(tensor[axis_tensor_name])",
   "    _shape = list(self.shape_pick(tensor[input_tensor_name]))",
   "    input_dims = len(_shape)",
   "    if axis.shape == 0:  # for scalar",
   "        axis = [axis]",
   "    need_axis = list(range(input_dims))",
   "    need_axis.pop(0)  # no batch dim",
   "    return list(axis) == need_axis",
   ""
  ]
 },
 {
  "ruler_name": "layer_norm_sqrdiffv2",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Sub",
   "Mul_1",
   "C",
   "Mul_2",
   "Rsqrt",
   "C_1",
   "Mean",
   "Add_1",
   "C_2",
   "Mean_1",
   "C_3",
   "SquaredDifference",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Sub:out0",
    "Add:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "Mean:out0",
    "Mul_2:in0"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_3:out0",
    "Add_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "Mean:out0",
    "SquaredDifference:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "layernormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "layernormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "layernormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "layernormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ]
   }
  },
  "blob_map": {
   "layernormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_layernorm_no_scale_bias",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Mul_1",
   "Reshape",
   "Rsqrt",
   "Neg",
   "C",
   "Add_1",
   "Mean",
   "Mean_1",
   "C_1",
   "C_2",
   "SquaredDifference",
   "C_3",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in1"
   ],
   [
    "Reshape:out0",
    "Mul:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in1"
   ],
   [
    "Neg:out0",
    "Mul_1:in0"
   ],
   [
    "C:out0",
    "Reshape:in1"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "Mean:out0",
    "Neg:in0"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_1:out0",
    "Add_1:in1"
   ],
   [
    "Reshape:out0",
    "Mean:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_3:out0",
    "Mean_1:in1"
   ],
   [
    "Reshape:out0",
    "SquaredDifference:in0"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Reshape:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "reshape",
   "layernormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reshape:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "layernormalize:out0"
   ]
  ],
  "acu_inter_flow": [
   [
    "reshape:out0",
    "layernormalize:in0"
   ]
  ],
  "param_map": {
   "layernormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   },
   "reshape": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "layernormalize": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "layernorm_with_moment",
  "src_ops_alias": [
   "AddV2",
   "Mul",
   "Sub",
   "Mul_1",
   "C",
   "Mul_2",
   "Rsqrt",
   "C_1",
   "Mean",
   "AddV2_1",
   "C_2",
   "Mean_1",
   "C_3",
   "SquaredDifference",
   "C_4",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "AddV2:in0"
   ],
   [
    "Sub:out0",
    "AddV2:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "Mean:out0",
    "Mul_2:in0"
   ],
   [
    "AddV2_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "AddV2_1:in0"
   ],
   [
    "C_3:out0",
    "AddV2_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ]
  ],
  "src_out_tensor": [
   "AddV2:out0"
  ],
  "acu_lys_alias": [
   "layernormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "layernormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "AddV2:out0",
    "layernormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "layernormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "axis_list": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "layernormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"axis_tensor_name\": \"C_2:out0\", \"input_tensor_name\": \"I:out0\"}",
   "def r_layernorm_pre_condition(self, node, tensor, axis_tensor_name, input_tensor_name):",
   "    ''' the axis list only does not contain the batch dim '''",
   "    axis = self.tensor_to_numpy(tensor[axis_tensor_name])",
   "    _shape = list(self.shape_pick(tensor[input_tensor_name]))",
   "    input_dims = len(_shape)",
   "    if axis.shape == 0:  # for scalar",
   "        axis = [axis]",
   "    need_axis = list(range(input_dims))",
   "    need_axis.pop(0)  # no batch dim",
   "    return list(axis) == need_axis",
   ""
  ]
 },
 {
  "ruler_name": "r_layernormalize1",
  "src_ops_alias": [
   "Add",
   "Mul",
   "C",
   "RealDiv",
   "C_1",
   "Sub",
   "Add_1",
   "Mean",
   "Sqrt",
   "C_2",
   "C_3",
   "Mean_1",
   "Square",
   "C_4",
   "Sub_1",
   "Mean_2",
   "C_5"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "RealDiv:out0",
    "Mul:in0"
   ],
   [
    "C_1:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "RealDiv:in0"
   ],
   [
    "Add_1:out0",
    "RealDiv:in1"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ],
   [
    "Sqrt:out0",
    "Add_1:in0"
   ],
   [
    "C_2:out0",
    "Add_1:in1"
   ],
   [
    "C_3:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "Sqrt:in0"
   ],
   [
    "Square:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "Sub_1:out0",
    "Square:in0"
   ],
   [
    "Mean_2:out0",
    "Sub_1:in1"
   ],
   [
    "C_5:out0",
    "Mean_2:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "Mean_2:in0"
   ],
   [
    "I:out0",
    "Sub_1:in0"
   ],
   [
    "I:out0",
    "Sub:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "layernormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "layernormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "layernormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "layernormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "axis_list": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "layernormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"axis_tensor_name\": \"C_3:out0\", \"input_tensor_name\": \"I:out0\"}",
   "def r_layernorm_pre_condition(self, node, tensor, axis_tensor_name, input_tensor_name):",
   "    ''' the axis list only does not contain the batch dim '''",
   "    axis = self.tensor_to_numpy(tensor[axis_tensor_name])",
   "    _shape = list(self.shape_pick(tensor[input_tensor_name]))",
   "    input_dims = len(_shape)",
   "    if axis.shape == 0:  # for scalar",
   "        axis = [axis]",
   "    need_axis = list(range(input_dims))",
   "    need_axis.pop(0)  # no batch dim",
   "    return list(axis) == need_axis",
   ""
  ]
 },
 {
  "ruler_name": "r_layernorm2",
  "src_ops_alias": [
   "Add",
   "Mul",
   "C",
   "Mul_1",
   "C_1",
   "Sub",
   "Rsqrt",
   "Mean",
   "Add_1",
   "C_2",
   "Mean_1",
   "C_3",
   "Square",
   "C_4",
   "Sub_1"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "C_1:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in1"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_3:out0",
    "Add_1:in1"
   ],
   [
    "Square:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "Sub_1:out0",
    "Square:in0"
   ],
   [
    "Mean:out0",
    "Sub_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "Sub:in0"
   ],
   [
    "I:out0",
    "Sub_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "layernormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "layernormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "layernormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "layernormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "axis_list": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "layernormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_layernorm3",
  "src_ops_alias": [
   "Mean",
   "C",
   "StopGradient",
   "SquaredDifference",
   "Mean_1",
   "C_1",
   "AddV2",
   "C_2",
   "Rsqrt",
   "Mul",
   "C_3",
   "Mul_1",
   "Mul_2",
   "Sub",
   "C_4",
   "AddV2_1"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Mean:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_1:out0",
    "Mean_1:in1"
   ],
   [
    "Mean_1:out0",
    "AddV2:in0"
   ],
   [
    "C_2:out0",
    "AddV2:in1"
   ],
   [
    "AddV2:out0",
    "Rsqrt:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul:in0"
   ],
   [
    "C_3:out0",
    "Mul:in1"
   ],
   [
    "Mul:out0",
    "Mul_1:in1"
   ],
   [
    "Mean:out0",
    "Mul_2:in0"
   ],
   [
    "Mul:out0",
    "Mul_2:in1"
   ],
   [
    "C_4:out0",
    "Sub:in0"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "Mul_1:out0",
    "AddV2_1:in0"
   ],
   [
    "Sub:out0",
    "AddV2_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Mul_1:in0"
   ]
  ],
  "src_out_tensor": [
   "AddV2_1:out0"
  ],
  "acu_lys_alias": [
   "layernormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "layernormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "AddV2_1:out0",
    "layernormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "layernormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ]
   }
  },
  "blob_map": {
   "layernormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_layernorm4",
  "src_ops_alias": [
   "AddV2",
   "Mul",
   "C",
   "Mul_1",
   "C_1",
   "Sub",
   "Rsqrt",
   "Mean",
   "AddV2_1",
   "C_2",
   "Mean_1",
   "C_3",
   "SquaredDifference",
   "C_4",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "AddV2:in0"
   ],
   [
    "C:out0",
    "AddV2:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "C_1:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in1"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ],
   [
    "AddV2_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "AddV2_1:in0"
   ],
   [
    "C_3:out0",
    "AddV2_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Sub:in0"
   ]
  ],
  "src_out_tensor": [
   "AddV2:out0"
  ],
  "acu_lys_alias": [
   "layernormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "layernormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "AddV2:out0",
    "layernormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "layernormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "axis_list": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "layernormalize": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_moments_bnsingle",
  "src_ops_alias": [
   "Add",
   "Mul",
   "C",
   "Mul_1",
   "C_1",
   "Sub",
   "Rsqrt",
   "Mean",
   "Add_1",
   "C_2",
   "Mean_1",
   "C_3",
   "Square",
   "C_4",
   "Sub_1"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "C:out0",
    "Add:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in0"
   ],
   [
    "C_1:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in1"
   ],
   [
    "Mean:out0",
    "Sub:in1"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "Add_1:in0"
   ],
   [
    "C_3:out0",
    "Add_1:in1"
   ],
   [
    "Square:out0",
    "Mean_1:in0"
   ],
   [
    "C_4:out0",
    "Mean_1:in1"
   ],
   [
    "Sub_1:out0",
    "Square:in0"
   ],
   [
    "Mean:out0",
    "Sub_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Sub_1:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "Sub:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "moments",
   "batchnorm_single"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "moments:in0"
   ],
   [
    "I:out0",
    "batchnorm_single:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "batchnorm_single:out0"
   ]
  ],
  "acu_inter_flow": [
   [
    "moments:out0",
    "batchnorm_single:in1"
   ],
   [
    "moments:out1",
    "batchnorm_single:in2"
   ]
  ],
  "param_map": {
   "moments": {
    "axis_list": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0']).tolist()"
    ],
    "keep_dims": [
     "BOOL",
     "VALUE",
     true
    ]
   },
   "batchnorm_single": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ]
   }
  },
  "blob_map": {
   "batchnorm_single": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_tensor_name\": \"I:out0\"}",
   "def r_check_moments_bnsingle(self, node, tensor, input_tensor_name):",
   "    _shape = list(self.shape_pick(tensor[input_tensor_name]))",
   "    input_dims = len(_shape)",
   "    if input_dims == 3:",
   "        return True",
   "    return False",
   ""
  ]
 },
 {
  "ruler_name": "moments",
  "src_ops_alias": [
   "Mean",
   "Mean_1",
   "C",
   "SquaredDifference",
   "C_1",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Mean:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_1:out0",
    "Mean_1:in1"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "SquaredDifference:in0"
   ]
  ],
  "src_out_tensor": [
   "Mean:out0",
   "Mean_1:out0"
  ],
  "acu_lys_alias": [
   "moments"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "moments:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mean:out0",
    "moments:out0"
   ],
   [
    "Mean_1:out0",
    "moments:out1"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "moments": {
    "axis_list": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']).tolist()"
    ],
    "keep_dims": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "moments": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_moments2",
  "src_ops_alias": [
   "Sub",
   "Add",
   "Mul",
   "Square",
   "Mul_1",
   "Reshape",
   "Sum",
   "C",
   "Sum_1",
   "StopGradient",
   "C_1",
   "SquaredDifference",
   "C_2",
   "Sub_1",
   "C_3",
   "Mean",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Sub:in0"
   ],
   [
    "Square:out0",
    "Sub:in1"
   ],
   [
    "Mul_1:out0",
    "Add:in0"
   ],
   [
    "Reshape:out0",
    "Add:in1"
   ],
   [
    "Sum:out0",
    "Mul:in0"
   ],
   [
    "C:out0",
    "Mul:in1"
   ],
   [
    "Mul_1:out0",
    "Square:in0"
   ],
   [
    "C:out0",
    "Mul_1:in1"
   ],
   [
    "Sum_1:out0",
    "Mul_1:in0"
   ],
   [
    "StopGradient:out0",
    "Reshape:in0"
   ],
   [
    "C_1:out0",
    "Reshape:in1"
   ],
   [
    "SquaredDifference:out0",
    "Sum:in0"
   ],
   [
    "C_2:out0",
    "Sum:in1"
   ],
   [
    "Sub_1:out0",
    "Sum_1:in0"
   ],
   [
    "C_3:out0",
    "Sum_1:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "StopGradient:out0",
    "Sub_1:in1"
   ],
   [
    "C_4:out0",
    "Mean:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SquaredDifference:in0"
   ],
   [
    "I:out0",
    "Mean:in0"
   ],
   [
    "I:out0",
    "Sub_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Sub:out0",
   "Add:out0"
  ],
  "acu_lys_alias": [
   "moments"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "moments:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "moments:out0"
   ],
   [
    "Sub:out0",
    "moments:out1"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "moments": {
    "axis_list": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0']).tolist()"
    ],
    "keep_dims": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "moments": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "batchnorm_single",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Sub",
   "Mul_1",
   "C",
   "Mul_2",
   "Rsqrt",
   "C_1",
   "Add_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Sub:out0",
    "Add:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "Add_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "Add_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I_1:out0",
    "Mul_2:in0"
   ],
   [
    "I_2:out0",
    "Add_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "batchnorm_single"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "batchnorm_single:in0"
   ],
   [
    "I_1:out0",
    "batchnorm_single:in1"
   ],
   [
    "I_2:out0",
    "batchnorm_single:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "batchnorm_single:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "batchnorm_single": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ]
   }
  },
  "blob_map": {
   "batchnorm_single": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "batchnorm_single_1",
  "src_ops_alias": [
   "AddV2",
   "Mul",
   "Sub",
   "Mul_1",
   "C",
   "Mul_2",
   "Rsqrt",
   "C_1",
   "AddV2_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "AddV2:in0"
   ],
   [
    "Sub:out0",
    "AddV2:in1"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Sub:in0"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "C_1:out0",
    "Mul_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "AddV2_1:out0",
    "Rsqrt:in0"
   ],
   [
    "C_2:out0",
    "AddV2_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I_1:out0",
    "Mul_2:in0"
   ],
   [
    "I_2:out0",
    "AddV2_1:in0"
   ]
  ],
  "src_out_tensor": [
   "AddV2:out0"
  ],
  "acu_lys_alias": [
   "batchnorm_single"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "batchnorm_single:in0"
   ],
   [
    "I_1:out0",
    "batchnorm_single:in1"
   ],
   [
    "I_2:out0",
    "batchnorm_single:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "AddV2:out0",
    "batchnorm_single:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "batchnorm_single": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ]
   }
  },
  "blob_map": {
   "batchnorm_single": {
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_group_norm",
  "src_ops_alias": [
   "Reshape",
   "AddV2",
   "C",
   "Mul",
   "Sub",
   "Reshape_1",
   "Mul_1",
   "C_1",
   "Mul_2",
   "C_2",
   "Rsqrt",
   "C_3",
   "Mean",
   "AddV2_1",
   "C_4",
   "Mean_1",
   "C_5",
   "SquaredDifference",
   "C_6",
   "StopGradient"
  ],
  "src_inter_flow": [
   [
    "AddV2:out0",
    "Reshape:in0"
   ],
   [
    "C:out0",
    "Reshape:in1"
   ],
   [
    "Mul:out0",
    "AddV2:in0"
   ],
   [
    "Sub:out0",
    "AddV2:in1"
   ],
   [
    "Reshape_1:out0",
    "Mul:in0"
   ],
   [
    "Mul_1:out0",
    "Mul:in1"
   ],
   [
    "C_1:out0",
    "Sub:in0"
   ],
   [
    "Mul_2:out0",
    "Sub:in1"
   ],
   [
    "C_2:out0",
    "Reshape_1:in1"
   ],
   [
    "Rsqrt:out0",
    "Mul_1:in0"
   ],
   [
    "C_3:out0",
    "Mul_1:in1"
   ],
   [
    "Mul_1:out0",
    "Mul_2:in1"
   ],
   [
    "Mean:out0",
    "Mul_2:in0"
   ],
   [
    "AddV2_1:out0",
    "Rsqrt:in0"
   ],
   [
    "Reshape_1:out0",
    "Mean:in0"
   ],
   [
    "C_4:out0",
    "Mean:in1"
   ],
   [
    "Mean_1:out0",
    "AddV2_1:in0"
   ],
   [
    "C_5:out0",
    "AddV2_1:in1"
   ],
   [
    "SquaredDifference:out0",
    "Mean_1:in0"
   ],
   [
    "C_6:out0",
    "Mean_1:in1"
   ],
   [
    "Reshape_1:out0",
    "SquaredDifference:in0"
   ],
   [
    "StopGradient:out0",
    "SquaredDifference:in1"
   ],
   [
    "Mean:out0",
    "StopGradient:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Reshape_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Reshape:out0"
  ],
  "acu_lys_alias": [
   "groupnormalize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "groupnormalize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Reshape:out0",
    "groupnormalize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "groupnormalize": {
    "eps": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_5:out0'])"
    ],
    "num_groups": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"name\": \"Reshape_1:out0\"}",
      "def r_get_group_norm_group(self, node, tensor, name):",
      "    _shape = list(self.shape_pick(tensor[name]))",
      "    return _shape[-2]",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "groupnormalize": {
    "scale": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"name\": \"Reshape_1:out0\"}",
   "def r_group_norm_pre_condition(self, node, tensor, name):",
   "    _rank = len(self.shape_pick(tensor[name]))",
   "    # Group input/output shape is [N,H,W,G,C/G]",
   "    if _rank == 5:",
   "        return True",
   "    return False",
   ""
  ]
 },
 {
  "ruler_name": "gather",
  "src_ops_alias": [
   "Gather"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Gather:in0"
   ],
   [
    "I_1:out0",
    "Gather:in1"
   ]
  ],
  "src_out_tensor": [
   "Gather:out0"
  ],
  "acu_lys_alias": [
   "gather"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gather:in0"
   ],
   [
    "I_1:out0",
    "gather:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Gather:out0",
    "gather:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gather": {}
  },
  "blob_map": {
   "gather": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "gather_axis",
  "src_ops_alias": [
   "GatherV2",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "GatherV2:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "GatherV2:in0"
   ],
   [
    "I_1:out0",
    "GatherV2:in1"
   ]
  ],
  "src_out_tensor": [
   "GatherV2:out0"
  ],
  "acu_lys_alias": [
   "gather"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gather:in0"
   ],
   [
    "I_1:out0",
    "gather:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "GatherV2:out0",
    "gather:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gather": {
    "axis": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "batch_dims": [
     "INT",
     "CODE",
     "self.attr_pick(node['GatherV2'], 'batch_dims')"
    ]
   }
  },
  "blob_map": {
   "gather": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "gather_nd",
  "src_ops_alias": [
   "GatherNd"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "GatherNd:in0"
   ],
   [
    "I_1:out0",
    "GatherNd:in1"
   ]
  ],
  "src_out_tensor": [
   "GatherNd:out0"
  ],
  "acu_lys_alias": [
   "gathernd"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gathernd:in0"
   ],
   [
    "I_1:out0",
    "gathernd:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "GatherNd:out0",
    "gathernd:out0"
   ]
  ],
  "param_map": {
   "gathernd": {}
  },
  "blob_map": {
   "gathernd": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "scatter_nd",
  "src_ops_alias": [
   "ScatterNd",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "ScatterNd:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ScatterNd:in0"
   ],
   [
    "I_1:out0",
    "ScatterNd:in1"
   ]
  ],
  "src_out_tensor": [
   "ScatterNd:out0"
  ],
  "acu_lys_alias": [
   "scatternd"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "scatternd:in0"
   ],
   [
    "I_1:out0",
    "scatternd:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ScatterNd:out0",
    "scatternd:out0"
   ]
  ],
  "param_map": {
   "scatternd": {
    "shape": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "scatternd": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "tile",
  "src_ops_alias": [
   "Tile",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Tile:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Tile:in0"
   ]
  ],
  "src_out_tensor": [
   "Tile:out0"
  ],
  "acu_lys_alias": [
   "tile"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "tile:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Tile:out0",
    "tile:out0"
   ]
  ],
  "param_map": {
   "tile": {
    "multiples": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "tile": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "repeat_2inputs_with_axis",
  "src_ops_alias": [
   "GatherV2",
   "Reshape",
   "Squeeze",
   "C",
   "Tile",
   "C_1",
   "Where",
   "ExpandDims",
   "Pack",
   "Reshape_1",
   "C_2",
   "C_3",
   "Maximum",
   "C_4",
   "Less",
   "C_5",
   "C_6",
   "Max",
   "Range",
   "Cast",
   "BroadcastTo",
   "C_7",
   "C_8",
   "C_9",
   "ExpandDims_1",
   "Cast_1",
   "C_10",
   "C_11"
  ],
  "src_inter_flow": [
   [
    "Reshape:out0",
    "GatherV2:in0"
   ],
   [
    "Squeeze:out0",
    "GatherV2:in1"
   ],
   [
    "C:out0",
    "GatherV2:in2"
   ],
   [
    "Tile:out0",
    "Reshape:in0"
   ],
   [
    "C_1:out0",
    "Reshape:in1"
   ],
   [
    "Where:out0",
    "Squeeze:in0"
   ],
   [
    "ExpandDims:out0",
    "Tile:in0"
   ],
   [
    "Pack:out0",
    "Tile:in1"
   ],
   [
    "Reshape_1:out0",
    "Where:in0"
   ],
   [
    "C_2:out0",
    "ExpandDims:in1"
   ],
   [
    "C_3:out0",
    "Pack:in0"
   ],
   [
    "Maximum:out0",
    "Pack:in1"
   ],
   [
    "C_4:out0",
    "Pack:in2"
   ],
   [
    "Less:out0",
    "Reshape_1:in0"
   ],
   [
    "C_5:out0",
    "Reshape_1:in1"
   ],
   [
    "C_6:out0",
    "Maximum:in0"
   ],
   [
    "Max:out0",
    "Maximum:in1"
   ],
   [
    "Range:out0",
    "Less:in0"
   ],
   [
    "Cast:out0",
    "Less:in1"
   ],
   [
    "BroadcastTo:out0",
    "Max:in0"
   ],
   [
    "C_7:out0",
    "Max:in1"
   ],
   [
    "Maximum:out0",
    "Range:in1"
   ],
   [
    "C_8:out0",
    "Range:in0"
   ],
   [
    "C_9:out0",
    "Range:in2"
   ],
   [
    "ExpandDims_1:out0",
    "Cast:in0"
   ],
   [
    "Cast_1:out0",
    "BroadcastTo:in0"
   ],
   [
    "C_10:out0",
    "BroadcastTo:in1"
   ],
   [
    "BroadcastTo:out0",
    "ExpandDims_1:in0"
   ],
   [
    "C_11:out0",
    "ExpandDims_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ExpandDims:in0"
   ],
   [
    "I_1:out0",
    "Cast_1:in0"
   ]
  ],
  "src_out_tensor": [
   "GatherV2:out0"
  ],
  "acu_lys_alias": [
   "repeat"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "repeat:in0"
   ],
   [
    "I_1:out0",
    "repeat:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "GatherV2:out0",
    "repeat:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "repeat": {
    "axis": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "repeat": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "sequencemask_with_2input",
  "src_ops_alias": [
   "Less",
   "Range",
   "Cast",
   "C",
   "C_1",
   "ExpandDims",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "Range:out0",
    "Less:in0"
   ],
   [
    "Cast:out0",
    "Less:in1"
   ],
   [
    "C:out0",
    "Range:in0"
   ],
   [
    "C_1:out0",
    "Range:in2"
   ],
   [
    "ExpandDims:out0",
    "Cast:in0"
   ],
   [
    "C_2:out0",
    "ExpandDims:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ExpandDims:in0"
   ],
   [
    "I_1:out0",
    "Range:in1"
   ]
  ],
  "src_out_tensor": [
   "Less:out0"
  ],
  "acu_lys_alias": [
   "sequence_mask"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "sequence_mask:in0"
   ],
   [
    "I_1:out0",
    "sequence_mask:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Less:out0",
    "sequence_mask:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "sequence_mask": {}
  },
  "blob_map": {
   "sequence_mask": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "max_pool",
  "src_ops_alias": [
   "MaxPool"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "MaxPool:in0"
   ]
  ],
  "src_out_tensor": [
   "MaxPool:out0"
  ],
  "acu_lys_alias": [
   "pooling"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "pooling:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "MaxPool:out0",
    "pooling:out0"
   ]
  ],
  "param_map": {
   "pooling": {
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['MaxPool'], 'padding')"
    ],
    "round_type": [
     "STRING",
     "VALUE",
     "floor"
    ],
    "type": [
     "STRING",
     "VALUE",
     "MAX"
    ],
    "ksize_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool'], 'ksize')[1]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool'], 'ksize')[2]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool'], 'strides')[2]"
    ]
   }
  },
  "blob_map": {
   "pooling": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "max_pool3d",
  "src_ops_alias": [
   "MaxPool3D"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "MaxPool3D:in0"
   ]
  ],
  "src_out_tensor": [
   "MaxPool3D:out0"
  ],
  "acu_lys_alias": [
   "pool3d"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "pool3d:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "MaxPool3D:out0",
    "pool3d:out0"
   ]
  ],
  "param_map": {
   "pool3d": {
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['MaxPool3D'], 'padding')"
    ],
    "round_type": [
     "STRING",
     "VALUE",
     "floor"
    ],
    "type": [
     "STRING",
     "VALUE",
     "MAX"
    ],
    "ksize_d": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool3D'], 'ksize')[1]"
    ],
    "ksize_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool3D'], 'ksize')[2]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool3D'], 'ksize')[3]"
    ],
    "stride_d": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool3D'], 'strides')[1]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool3D'], 'strides')[2]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['MaxPool3D'], 'strides')[3]"
    ],
    "pad_method": [
     "STRING",
     "VALUE",
     "auto"
    ]
   }
  },
  "blob_map": {
   "pool3d": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "avgpool",
  "src_ops_alias": [
   "AvgPool"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "AvgPool:in0"
   ]
  ],
  "src_out_tensor": [
   "AvgPool:out0"
  ],
  "acu_lys_alias": [
   "pooling"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "pooling:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "AvgPool:out0",
    "pooling:out0"
   ]
  ],
  "param_map": {
   "pooling": {
    "padding": [
     "STRING",
     "CODE",
     "self.attr_pick(node['AvgPool'], 'padding')"
    ],
    "round_type": [
     "STRING",
     "VALUE",
     "floor"
    ],
    "type": [
     "STRING",
     "VALUE",
     "AVG"
    ],
    "ksize_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['AvgPool'], 'ksize')[1]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['AvgPool'], 'ksize')[2]"
    ],
    "stride_h": [
     "INT",
     "CODE",
     "self.attr_pick(node['AvgPool'], 'strides')[1]"
    ],
    "stride_w": [
     "INT",
     "CODE",
     "self.attr_pick(node['AvgPool'], 'strides')[2]"
    ]
   }
  },
  "blob_map": {
   "pooling": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "reduce_mean_keepshape",
  "src_ops_alias": [
   "Mean",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Mean:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ]
  ],
  "src_out_tensor": [
   "Mean:out0"
  ],
  "acu_lys_alias": [
   "pooling"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "pooling:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mean:out0",
    "pooling:out0"
   ]
  ],
  "param_map": {
   "pooling": {
    "padding": [
     "STRING",
     "VALUE",
     "VALID"
    ],
    "round_type": [
     "STRING",
     "VALUE",
     "floor"
    ],
    "type": [
     "STRING",
     "VALUE",
     "AVG"
    ],
    "stride_h": [
     "INT",
     "VALUE",
     1
    ],
    "stride_w": [
     "INT",
     "VALUE",
     1
    ],
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['I:out0'])[1]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['I:out0'])[2]"
    ]
   }
  },
  "blob_map": {
   "pooling": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 1,
  "pre_condition": "self.attr_pick(node['Mean'], 'keep_dims') == True and self.tensor_to_numpy(tensor['C:out0']).tolist() == [1, 2]"
 },
 {
  "ruler_name": "reduce_mean_nokeepshape",
  "src_ops_alias": [
   "Mean",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Mean:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ]
  ],
  "src_out_tensor": [
   "Mean:out0"
  ],
  "acu_lys_alias": [
   "pooling",
   "reshape"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "pooling:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mean:out0",
    "reshape:out0"
   ]
  ],
  "param_map": {
   "reshape": {
    "shape": [
     "INTS",
     "VALUE",
     [
      0,
      -1
     ]
    ]
   },
   "pooling": {
    "padding": [
     "STRING",
     "VALUE",
     "VALID"
    ],
    "round_type": [
     "STRING",
     "VALUE",
     "floor"
    ],
    "type": [
     "STRING",
     "VALUE",
     "AVG"
    ],
    "stride_h": [
     "INT",
     "VALUE",
     1
    ],
    "stride_w": [
     "INT",
     "VALUE",
     1
    ],
    "ksize_h": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['I:out0'])[1]"
    ],
    "ksize_w": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['I:out0'])[2]"
    ]
   }
  },
  "blob_map": {
   "reshape": {},
   "pooling": {}
  },
  "acu_inter_flow": [
   [
    "pooling:out0",
    "reshape:in0"
   ]
  ],
  "priority_tip": 1,
  "pre_condition": "self.attr_pick(node['Mean'], 'keep_dims') == False and self.tensor_to_numpy(tensor['C:out0']).tolist() == [1, 2]"
 },
 {
  "ruler_name": "reducemean",
  "src_ops_alias": [
   "Mean",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Mean:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mean:in0"
   ]
  ],
  "src_out_tensor": [
   "Mean:out0"
  ],
  "acu_lys_alias": [
   "reducemean"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reducemean:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mean:out0",
    "reducemean:out0"
   ]
  ],
  "param_map": {
   "reducemean": {
    "axis_list": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"axis_tensor_name\": \"C:out0\"}",
      "def r_reduce_x_get_axis(self, node, tensor, axis_tensor_name):",
      "    axis = self.tensor_to_numpy(tensor[axis_tensor_name]).tolist()",
      "    if isinstance(axis,list):",
      "        return axis",
      "    else:",
      "        return [axis]",
      ""
     ]
    ],
    "keep_dims": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['Mean'], 'keep_dims')"
    ]
   }
  },
  "blob_map": {
   "reducemean": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_reducemax",
  "src_ops_alias": [
   "Max",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Max:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Max:in0"
   ]
  ],
  "src_out_tensor": [
   "Max:out0"
  ],
  "acu_lys_alias": [
   "reducemax"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reducemax:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Max:out0",
    "reducemax:out0"
   ]
  ],
  "param_map": {
   "reducemax": {
    "axis_list": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"axis_tensor_name\": \"C:out0\"}",
      "def r_reduce_x_get_axis(self, node, tensor, axis_tensor_name):",
      "    axis = self.tensor_to_numpy(tensor[axis_tensor_name]).tolist()",
      "    if isinstance(axis,list):",
      "        return axis",
      "    else:",
      "        return [axis]",
      ""
     ]
    ],
    "keep_dims": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['Max'], 'keep_dims')"
    ]
   }
  },
  "blob_map": {
   "reducemax": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "batch_to_space",
  "src_ops_alias": [
   "BatchToSpaceND",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "BatchToSpaceND:in1"
   ],
   [
    "C_1:out0",
    "BatchToSpaceND:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "BatchToSpaceND:in0"
   ]
  ],
  "src_out_tensor": [
   "BatchToSpaceND:out0"
  ],
  "acu_lys_alias": [
   "batch2space"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "batch2space:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BatchToSpaceND:out0",
    "batch2space:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "batch2space": {
    "block_shape": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "block_crops": [
     "ORIGIN",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "batch2space": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "space_to_batch",
  "src_ops_alias": [
   "SpaceToBatchND",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "SpaceToBatchND:in1"
   ],
   [
    "C_1:out0",
    "SpaceToBatchND:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "SpaceToBatchND:in0"
   ]
  ],
  "src_out_tensor": [
   "SpaceToBatchND:out0"
  ],
  "acu_lys_alias": [
   "space2batch"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "space2batch:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "SpaceToBatchND:out0",
    "space2batch:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "space2batch": {
    "block_shape": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "block_paddings": [
     "ORIGIN",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0']).tolist()"
    ]
   }
  },
  "blob_map": {
   "space2batch": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "poolwithargmax",
  "src_ops_alias": [
   "MaxPoolWithArgmax"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "MaxPoolWithArgmax:in0"
   ]
  ],
  "src_out_tensor": [
   "MaxPoolWithArgmax:out0"
  ],
  "acu_lys_alias": [
   "poolwithargmax"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "poolwithargmax:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "MaxPoolWithArgmax:out0",
    "poolwithargmax:out0"
   ]
  ],
  "param_map": {
   "poolwithargmax": {}
  },
  "blob_map": {
   "poolwithargmax": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "reshape",
  "src_ops_alias": [
   "Reshape",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Reshape:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Reshape:in0"
   ]
  ],
  "src_out_tensor": [
   "Reshape:out0"
  ],
  "acu_lys_alias": [
   "reshape"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reshape:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Reshape:out0",
    "reshape:out0"
   ]
  ],
  "param_map": {
   "reshape": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "reshape": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "expand_dims",
  "src_ops_alias": [
   "ExpandDims",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "ExpandDims:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ExpandDims:in0"
   ]
  ],
  "src_out_tensor": [
   "ExpandDims:out0"
  ],
  "acu_lys_alias": [
   "reshape"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reshape:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ExpandDims:out0",
    "reshape:out0"
   ]
  ],
  "param_map": {
   "reshape": {
    "shape": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"axis_tensor_name\": \"C:out0\"}",
      "def r_get_expand_dim_out_shape(self, node, tensor, axis_tensor_name):",
      "    input_shape = list(self.shape_pick(tensor['I:out0']))",
      "    input_shape_dims = len(input_shape)",
      "    axis = self.tensor_to_numpy(tensor['C:out0'])",
      "    if axis.ndim == 1:",
      "        axis = axis[0]",
      "    # convert netagive axis to positive to calculate output shape easier",
      "    if axis < 0:",
      "        axis = (axis + (input_shape_dims + 1)) % (input_shape_dims + 1)",
      "    output_shape = input_shape[0:axis] + [1] + input_shape[axis:]",
      "    return output_shape",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "reshape": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "squeeze",
  "src_ops_alias": [
   "Squeeze"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Squeeze:in0"
   ]
  ],
  "src_out_tensor": [
   "Squeeze:out0"
  ],
  "acu_lys_alias": [
   "reshape"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reshape:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Squeeze:out0",
    "reshape:out0"
   ]
  ],
  "param_map": {
   "reshape": {
    "shape": [
     "INTS",
     "CODE",
     "self.squeeze_shapes(self.attr_pick(node['Squeeze'], 'squeeze_dims', None), self.shape_pick(tensor['I:out0']))"
    ]
   }
  },
  "blob_map": {
   "reshape": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "transpose",
  "src_ops_alias": [
   "Transpose",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Transpose:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Transpose:in0"
   ]
  ],
  "src_out_tensor": [
   "Transpose:out0"
  ],
  "acu_lys_alias": [
   "permute"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "permute:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Transpose:out0",
    "permute:out0"
   ]
  ],
  "param_map": {
   "permute": {
    "perm": [
     "STRING",
     "CODE",
     "' '.join([str(perm) for perm in self.tensor_to_numpy(tensor['C:out0'])])"
    ]
   }
  },
  "blob_map": {
   "permute": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "transpose_to_noop",
  "src_ops_alias": [
   "Transpose",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Transpose:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Transpose:in0"
   ]
  ],
  "src_out_tensor": [
   "Transpose:out0"
  ],
  "acu_lys_alias": [
   "noop"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "noop:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Transpose:out0",
    "noop:out0"
   ]
  ],
  "param_map": {
   "noop": {}
  },
  "blob_map": {
   "noop": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 1,
  "pre_condition": "self.tensor_to_numpy(tensor['C:out0']).tolist() == [ i for i in range(len(self.tensor_to_numpy(tensor['C:out0'])))]"
 },
 {
  "ruler_name": "depth2space",
  "src_ops_alias": [
   "DepthToSpace"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "DepthToSpace:in0"
   ]
  ],
  "src_out_tensor": [
   "DepthToSpace:out0"
  ],
  "acu_lys_alias": [
   "depth2space"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "depth2space:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "DepthToSpace:out0",
    "depth2space:out0"
   ]
  ],
  "param_map": {
   "depth2space": {
    "block_size": [
     "INT",
     "CODE",
     "self.attr_pick(node['DepthToSpace'], 'block_size')"
    ]
   }
  },
  "blob_map": {
   "depth2space": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "space2depth",
  "src_ops_alias": [
   "SpaceToDepth"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "SpaceToDepth:in0"
   ]
  ],
  "src_out_tensor": [
   "SpaceToDepth:out0"
  ],
  "acu_lys_alias": [
   "space2depth"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "space2depth:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "SpaceToDepth:out0",
    "space2depth:out0"
   ]
  ],
  "param_map": {
   "space2depth": {
    "block_size": [
     "INT",
     "CODE",
     "self.attr_pick(node['SpaceToDepth'], 'block_size')"
    ]
   }
  },
  "blob_map": {
   "space2depth": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "reverse",
  "src_ops_alias": [
   "ReverseV2",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "ReverseV2:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ReverseV2:in0"
   ]
  ],
  "src_out_tensor": [
   "ReverseV2:out0"
  ],
  "acu_lys_alias": [
   "reverse"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reverse:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ReverseV2:out0",
    "reverse:out0"
   ]
  ],
  "param_map": {
   "reverse": {
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "reverse": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "pad",
  "src_ops_alias": [
   "Pad",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Pad:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Pad:in0"
   ]
  ],
  "src_out_tensor": [
   "Pad:out0"
  ],
  "acu_lys_alias": [
   "pad"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "pad:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Pad:out0",
    "pad:out0"
   ]
  ],
  "param_map": {
   "pad": {
    "padding_value": [
     "ORIGIN",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']).tolist()"
    ],
    "padding_mode": [
     "ORIGIN",
     "CODE",
     "self.attr_pick(node['Pad'], 'mode', 'CONSTANT')"
    ]
   }
  },
  "blob_map": {
   "pad": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_padv2",
  "src_ops_alias": [
   "PadV2",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "PadV2:in1"
   ],
   [
    "C_1:out0",
    "PadV2:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "PadV2:in0"
   ]
  ],
  "src_out_tensor": [
   "PadV2:out0"
  ],
  "acu_lys_alias": [
   "pad"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "pad:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "PadV2:out0",
    "pad:out0"
   ]
  ],
  "param_map": {
   "pad": {
    "padding_value": [
     "ORIGIN",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']).tolist()"
    ],
    "padding_mode": [
     "ORIGIN",
     "CODE",
     "self.attr_pick(node['PadV2'], 'mode', 'CONSTANT')"
    ],
    "padding_const": [
     "ORIGIN",
     "CODE",
     "np.round(self.tensor_to_numpy(tensor['C_1:out0']).tolist(), 6)"
    ]
   }
  },
  "blob_map": {
   "pad": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "mirrorpad",
  "src_ops_alias": [
   "MirrorPad",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "MirrorPad:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "MirrorPad:in0"
   ]
  ],
  "src_out_tensor": [
   "MirrorPad:out0"
  ],
  "acu_lys_alias": [
   "pad"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "pad:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "MirrorPad:out0",
    "pad:out0"
   ]
  ],
  "param_map": {
   "pad": {
    "padding_value": [
     "ORIGIN",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']).tolist()"
    ],
    "padding_mode": [
     "ORIGIN",
     "CODE",
     "self.attr_pick(node['MirrorPad'], 'mode', 'CONSTANT')"
    ]
   }
  },
  "blob_map": {
   "pad": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "resieze_biliner",
  "src_ops_alias": [
   "ResizeBilinear",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "ResizeBilinear:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ResizeBilinear:in0"
   ]
  ],
  "src_out_tensor": [
   "ResizeBilinear:out0"
  ],
  "acu_lys_alias": [
   "image_resize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "image_resize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ResizeBilinear:out0",
    "image_resize:out0"
   ]
  ],
  "param_map": {
   "image_resize": {
    "new_size": [
     "ORIGIN",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']).tolist()"
    ],
    "align_corners": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['ResizeBilinear'], 'align_corners')"
    ],
    "half_pixel": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['ResizeBilinear'], 'half_pixel_centers', False)"
    ],
    "type": [
     "STRING",
     "VALUE",
     "bilinear"
    ]
   }
  },
  "blob_map": {
   "image_resize": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "resizenearest",
  "src_ops_alias": [
   "ResizeNearestNeighbor",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "ResizeNearestNeighbor:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ResizeNearestNeighbor:in0"
   ]
  ],
  "src_out_tensor": [
   "ResizeNearestNeighbor:out0"
  ],
  "acu_lys_alias": [
   "image_resize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "image_resize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ResizeNearestNeighbor:out0",
    "image_resize:out0"
   ]
  ],
  "param_map": {
   "image_resize": {
    "new_size": [
     "ORIGIN",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0']).tolist()"
    ],
    "align_corners": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['ResizeNearestNeighbor'], 'align_corners', False)"
    ],
    "half_pixel": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['ResizeNearestNeighbor'], 'half_pixel_centers', False)"
    ],
    "type": [
     "STRING",
     "VALUE",
     "nearest"
    ]
   }
  },
  "blob_map": {
   "image_resize": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "upsampling",
  "src_ops_alias": [
   "Reshape",
   "Mul",
   "C",
   "Reshape_1",
   "C_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "Reshape_1:out0",
    "Mul:in0"
   ],
   [
    "C_2:out0",
    "Reshape_1:in1"
   ],
   [
    "C:out0",
    "Reshape:in1"
   ],
   [
    "Mul:out0",
    "Reshape:in0"
   ],
   [
    "C_1:out0",
    "Mul:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Reshape_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Reshape:out0"
  ],
  "acu_lys_alias": [
   "upsampling"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "upsampling:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Reshape:out0",
    "upsampling:out0"
   ]
  ],
  "param_map": {
   "upsampling": {
    "factor": [
     "ORIGIN",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[2]"
    ]
   }
  },
  "blob_map": {
   "upsampling": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "(self.tensor_to_numpy(tensor['C_1:out0']) == np.ones([1, 1, 2, 1, 2, 1])).all()"
 },
 {
  "ruler_name": "concat_strip_in1",
  "src_ops_alias": [
   "ConcatV2",
   "C_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "C_1:out0",
    "ConcatV2:in1"
   ],
   [
    "C_2:out0",
    "ConcatV2:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I_0:out0",
    "ConcatV2:in0"
   ]
  ],
  "src_out_tensor": [
   "ConcatV2:out0"
  ],
  "acu_lys_alias": [
   "noop"
  ],
  "src_acu_in_tensor_map": [
   [
    "I_0:out0",
    "noop:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ConcatV2:out0",
    "noop:out0"
   ]
  ],
  "param_map": {
   "noop": {}
  },
  "blob_map": {
   "noop": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 1,
  "pre_condition": "0 in self.shape_pick(tensor['C_1:out0'])"
 },
 {
  "ruler_name": "split",
  "src_ops_alias": [
   "Split",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Split:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Split:in1"
   ]
  ],
  "src_out_tensor": [
   "Split:out0"
  ],
  "acu_lys_alias": [
   "noop"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "noop:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Split:out0",
    "noop:out0"
   ]
  ],
  "param_map": {
   "noop": {}
  },
  "blob_map": {
   "noop": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 1,
  "pre_condition": "self.attr_pick(node['Split'], 'num_split') == 1"
 },
 {
  "ruler_name": "stride_slice",
  "src_ops_alias": [
   "StridedSlice",
   "C",
   "C_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "StridedSlice:in1"
   ],
   [
    "C_2:out0",
    "StridedSlice:in3"
   ],
   [
    "C_1:out0",
    "StridedSlice:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "StridedSlice:in0"
   ]
  ],
  "src_out_tensor": [
   "StridedSlice:out0"
  ],
  "acu_lys_alias": [
   "stridedslice"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "stridedslice:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "StridedSlice:out0",
    "stridedslice:out0"
   ]
  ],
  "param_map": {
   "stridedslice": {
    "slice_begin": [
     "INTS",
     "CODE",
     "[int(p) for p in self.tensor_to_numpy(tensor['C:out0'])]"
    ],
    "slice_end": [
     "INTS",
     "CODE",
     "[int(p) for p in self.tensor_to_numpy(tensor['C_1:out0'])]"
    ],
    "slice_strides": [
     "INTS",
     "CODE",
     "[int(p) for p in self.tensor_to_numpy(tensor['C_2:out0'])]"
    ],
    "slice_begin_mask": [
     "INT",
     "CODE",
     "self.attr_pick(node['StridedSlice'], 'begin_mask', 0)"
    ],
    "slice_end_mask": [
     "INT",
     "CODE",
     "self.attr_pick(node['StridedSlice'], 'end_mask', 0)"
    ],
    "slice_ellipsis_mask": [
     "INT",
     "CODE",
     "self.attr_pick(node['StridedSlice'], 'ellipsis_mask', 0)"
    ],
    "slice_new_axis_mask": [
     "INT",
     "CODE",
     "self.attr_pick(node['StridedSlice'], 'new_axis_mask', 0)"
    ],
    "slice_shrink_axis_mask": [
     "INT",
     "CODE",
     "self.attr_pick(node['StridedSlice'], 'shrink_axis_mask', 0)"
    ]
   }
  },
  "blob_map": {
   "stridedslice": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "slice",
  "src_ops_alias": [
   "Slice",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C_1:out0",
    "Slice:in2"
   ],
   [
    "C:out0",
    "Slice:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Slice:in0"
   ]
  ],
  "src_out_tensor": [
   "Slice:out0"
  ],
  "acu_lys_alias": [
   "slice"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "slice:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Slice:out0",
    "slice:out0"
   ]
  ],
  "param_map": {
   "slice": {
    "begin": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "size": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "blob_map": {
   "slice": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_slice_two_inputs",
  "src_ops_alias": [
   "Slice",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Slice:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Slice:in0"
   ],
   [
    "I_1:out0",
    "Slice:in1"
   ]
  ],
  "src_out_tensor": [
   "Slice:out0"
  ],
  "acu_lys_alias": [
   "slice"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "slice:in0"
   ],
   [
    "I_1:out0",
    "slice:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Slice:out0",
    "slice:out0"
   ]
  ],
  "param_map": {
   "slice": {
    "begin": [
     "ORIGIN",
     "VALUE",
     []
    ],
    "size": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "slice": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "softmax",
  "src_ops_alias": [
   "Softmax"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Softmax:in0"
   ]
  ],
  "src_out_tensor": [
   "Softmax:out0"
  ],
  "acu_lys_alias": [
   "softmax"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "softmax:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Softmax:out0",
    "softmax:out0"
   ]
  ],
  "param_map": {
   "softmax": {}
  },
  "blob_map": {
   "softmax": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "softmax_calc",
  "src_ops_alias": [
   "RealDiv",
   "Exp",
   "Sum",
   "Sub",
   "C",
   "Max",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "Sum:out0",
    "RealDiv:in1"
   ],
   [
    "Exp:out0",
    "RealDiv:in0"
   ],
   [
    "Max:out0",
    "Sub:in1"
   ],
   [
    "Exp:out0",
    "Sum:in0"
   ],
   [
    "C:out0",
    "Sum:in1"
   ],
   [
    "Sub:out0",
    "Exp:in0"
   ],
   [
    "C_1:out0",
    "Max:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Max:in0"
   ],
   [
    "I:out0",
    "Sub:in0"
   ]
  ],
  "src_out_tensor": [
   "RealDiv:out0"
  ],
  "acu_lys_alias": [
   "softmax"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "softmax:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "RealDiv:out0",
    "softmax:out0"
   ]
  ],
  "param_map": {
   "softmax": {
    "sf_axis": [
     "INT",
     "VALUE",
     -1
    ]
   }
  },
  "blob_map": {
   "softmax": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "log_softmax",
  "src_ops_alias": [
   "LogSoftmax"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "LogSoftmax:in0"
   ]
  ],
  "src_out_tensor": [
   "LogSoftmax:out0"
  ],
  "acu_lys_alias": [
   "log_softmax"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "log_softmax:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "LogSoftmax:out0",
    "log_softmax:out0"
   ]
  ],
  "param_map": {
   "log_softmax": {}
  },
  "blob_map": {
   "log_softmax": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "mish",
  "src_ops_alias": [
   "Mul",
   "Tanh",
   "Softplus"
  ],
  "src_inter_flow": [
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "Softplus:out0",
    "Tanh:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Softplus:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "mish"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "mish:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "mish:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "mish": {}
  },
  "blob_map": {
   "mish": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "mish_1",
  "src_ops_alias": [
   "Mul",
   "Tanh",
   "Softplus"
  ],
  "src_inter_flow": [
   [
    "Tanh:out0",
    "Mul:in0"
   ],
   [
    "Softplus:out0",
    "Tanh:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Softplus:in0"
   ],
   [
    "I:out0",
    "Mul:in1"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "mish"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "mish:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "mish:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "mish": {}
  },
  "blob_map": {
   "mish": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "rank",
  "src_ops_alias": [
   "Rank"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Rank:in0"
   ]
  ],
  "src_out_tensor": [
   "Rank:out0"
  ],
  "acu_lys_alias": [
   "rank"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "rank:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Rank:out0",
    "rank:out0"
   ]
  ],
  "param_map": {
   "rank": {}
  },
  "blob_map": {
   "rank": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "add",
  "src_ops_alias": [
   "Add"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Add:in0"
   ],
   [
    "I_1:out0",
    "Add:in1"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "add"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "add:in0"
   ],
   [
    "I_1:out0",
    "add:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "add:out0"
   ]
  ],
  "param_map": {
   "add": {}
  },
  "blob_map": {
   "add": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_addv2",
  "src_ops_alias": [
   "AddV2"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "AddV2:in0"
   ],
   [
    "I_1:out0",
    "AddV2:in1"
   ]
  ],
  "src_out_tensor": [
   "AddV2:out0"
  ],
  "acu_lys_alias": [
   "add"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "add:in0"
   ],
   [
    "I_1:out0",
    "add:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "AddV2:out0",
    "add:out0"
   ]
  ],
  "param_map": {
   "add": {}
  },
  "blob_map": {
   "add": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "greater",
  "src_ops_alias": [
   "Greater"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Greater:in0"
   ],
   [
    "I_1:out0",
    "Greater:in1"
   ]
  ],
  "src_out_tensor": [
   "Greater:out0"
  ],
  "acu_lys_alias": [
   "greater"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "greater:in0"
   ],
   [
    "I_1:out0",
    "greater:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Greater:out0",
    "greater:out0"
   ]
  ],
  "param_map": {
   "greater": {}
  },
  "blob_map": {
   "greater": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "greater_equal",
  "src_ops_alias": [
   "GreaterEqual"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "GreaterEqual:in0"
   ],
   [
    "I_1:out0",
    "GreaterEqual:in1"
   ]
  ],
  "src_out_tensor": [
   "GreaterEqual:out0"
  ],
  "acu_lys_alias": [
   "greater_equal"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "greater_equal:in0"
   ],
   [
    "I_1:out0",
    "greater_equal:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "GreaterEqual:out0",
    "greater_equal:out0"
   ]
  ],
  "param_map": {
   "greater_equal": {}
  },
  "blob_map": {
   "greater_equal": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "less",
  "src_ops_alias": [
   "Less"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Less:in0"
   ],
   [
    "I_1:out0",
    "Less:in1"
   ]
  ],
  "src_out_tensor": [
   "Less:out0"
  ],
  "acu_lys_alias": [
   "less"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "less:in0"
   ],
   [
    "I_1:out0",
    "less:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Less:out0",
    "less:out0"
   ]
  ],
  "param_map": {
   "less": {}
  },
  "blob_map": {
   "less": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "less_equal",
  "src_ops_alias": [
   "LessEqual"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "LessEqual:in0"
   ],
   [
    "I_1:out0",
    "LessEqual:in1"
   ]
  ],
  "src_out_tensor": [
   "LessEqual:out0"
  ],
  "acu_lys_alias": [
   "less_equal"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "less_equal:in0"
   ],
   [
    "I_1:out0",
    "less_equal:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "LessEqual:out0",
    "less_equal:out0"
   ]
  ],
  "param_map": {
   "less_equal": {}
  },
  "blob_map": {
   "less_equal": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "equal",
  "src_ops_alias": [
   "Equal"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Equal:in0"
   ],
   [
    "I_1:out0",
    "Equal:in1"
   ]
  ],
  "src_out_tensor": [
   "Equal:out0"
  ],
  "acu_lys_alias": [
   "equal"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "equal:in0"
   ],
   [
    "I_1:out0",
    "equal:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Equal:out0",
    "equal:out0"
   ]
  ],
  "param_map": {
   "equal": {}
  },
  "blob_map": {
   "equal": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "not_equal",
  "src_ops_alias": [
   "NotEqual"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "NotEqual:in0"
   ],
   [
    "I_1:out0",
    "NotEqual:in1"
   ]
  ],
  "src_out_tensor": [
   "NotEqual:out0"
  ],
  "acu_lys_alias": [
   "not_equal"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "not_equal:in0"
   ],
   [
    "I_1:out0",
    "not_equal:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "NotEqual:out0",
    "not_equal:out0"
   ]
  ],
  "param_map": {
   "not_equal": {}
  },
  "blob_map": {
   "not_equal": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "floor_div",
  "src_ops_alias": [
   "FloorDiv"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "FloorDiv:in0"
   ],
   [
    "I_1:out0",
    "FloorDiv:in1"
   ]
  ],
  "src_out_tensor": [
   "FloorDiv:out0"
  ],
  "acu_lys_alias": [
   "floor_div"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "floor_div:in0"
   ],
   [
    "I_1:out0",
    "floor_div:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "FloorDiv:out0",
    "floor_div:out0"
   ]
  ],
  "param_map": {
   "floor_div": {}
  },
  "blob_map": {
   "floor_div": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "div",
  "src_ops_alias": [
   "Div"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Div:in0"
   ],
   [
    "I_1:out0",
    "Div:in1"
   ]
  ],
  "src_out_tensor": [
   "Div:out0"
  ],
  "acu_lys_alias": [
   "divide"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "divide:in0"
   ],
   [
    "I_1:out0",
    "divide:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Div:out0",
    "divide:out0"
   ]
  ],
  "param_map": {
   "divide": {}
  },
  "blob_map": {
   "divide": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "logical_or",
  "src_ops_alias": [
   "LogicalOr"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "LogicalOr:in0"
   ],
   [
    "I_1:out0",
    "LogicalOr:in1"
   ]
  ],
  "src_out_tensor": [
   "LogicalOr:out0"
  ],
  "acu_lys_alias": [
   "logical_or"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "logical_or:in0"
   ],
   [
    "I_1:out0",
    "logical_or:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "LogicalOr:out0",
    "logical_or:out0"
   ]
  ],
  "param_map": {
   "logical_or": {}
  },
  "blob_map": {
   "logical_or": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_logical_xor",
  "src_ops_alias": [
   "LogicalAnd",
   "LogicalOr",
   "LogicalNot",
   "LogicalAnd_1"
  ],
  "src_inter_flow": [
   [
    "LogicalOr:out0",
    "LogicalAnd:in0"
   ],
   [
    "LogicalNot:out0",
    "LogicalAnd:in1"
   ],
   [
    "LogicalAnd_1:out0",
    "LogicalNot:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "LogicalAnd_1:in0"
   ],
   [
    "I:out0",
    "LogicalOr:in0"
   ],
   [
    "I_1:out0",
    "LogicalAnd_1:in1"
   ],
   [
    "I_1:out0",
    "LogicalOr:in1"
   ]
  ],
  "src_out_tensor": [
   "LogicalAnd:out0"
  ],
  "acu_lys_alias": [
   "logical_xor"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "logical_xor:in0"
   ],
   [
    "I_1:out0",
    "logical_xor:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "LogicalAnd:out0",
    "logical_xor:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "logical_xor": {}
  },
  "blob_map": {
   "logical_xor": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "logical_and",
  "src_ops_alias": [
   "LogicalAnd"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "LogicalAnd:in0"
   ],
   [
    "I_1:out0",
    "LogicalAnd:in1"
   ]
  ],
  "src_out_tensor": [
   "LogicalAnd:out0"
  ],
  "acu_lys_alias": [
   "logical_and"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "logical_and:in0"
   ],
   [
    "I_1:out0",
    "logical_and:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "LogicalAnd:out0",
    "logical_and:out0"
   ]
  ],
  "param_map": {
   "logical_and": {}
  },
  "blob_map": {
   "logical_and": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "pow",
  "src_ops_alias": [
   "Pow"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Pow:in0"
   ],
   [
    "I_1:out0",
    "Pow:in1"
   ]
  ],
  "src_out_tensor": [
   "Pow:out0"
  ],
  "acu_lys_alias": [
   "pow"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "pow:in0"
   ],
   [
    "I_1:out0",
    "pow:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Pow:out0",
    "pow:out0"
   ]
  ],
  "param_map": {
   "pow": {}
  },
  "blob_map": {
   "pow": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "real_div",
  "src_ops_alias": [
   "RealDiv"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "RealDiv:in0"
   ],
   [
    "I_1:out0",
    "RealDiv:in1"
   ]
  ],
  "src_out_tensor": [
   "RealDiv:out0"
  ],
  "acu_lys_alias": [
   "real_div"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "real_div:in0"
   ],
   [
    "I_1:out0",
    "real_div:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "RealDiv:out0",
    "real_div:out0"
   ]
  ],
  "param_map": {
   "real_div": {}
  },
  "blob_map": {
   "real_div": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "select",
  "src_ops_alias": [
   "Select"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Select:in0"
   ],
   [
    "I_1:out0",
    "Select:in1"
   ],
   [
    "I_2:out0",
    "Select:in2"
   ]
  ],
  "src_out_tensor": [
   "Select:out0"
  ],
  "acu_lys_alias": [
   "where"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "where:in0"
   ],
   [
    "I_1:out0",
    "where:in1"
   ],
   [
    "I_2:out0",
    "where:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Select:out0",
    "where:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "where": {}
  },
  "blob_map": {
   "where": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "where",
  "src_ops_alias": [
   "Where"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Where:in0"
   ]
  ],
  "src_out_tensor": [
   "Where:out0"
  ],
  "acu_lys_alias": [
   "where"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "where:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Where:out0",
    "where:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "where": {}
  },
  "blob_map": {
   "where": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "square",
  "src_ops_alias": [
   "Square"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Square:in0"
   ]
  ],
  "src_out_tensor": [
   "Square:out0"
  ],
  "acu_lys_alias": [
   "square"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "square:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Square:out0",
    "square:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "square": {}
  },
  "blob_map": {
   "square": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "sqrt",
  "src_ops_alias": [
   "Sqrt"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Sqrt:in0"
   ]
  ],
  "src_out_tensor": [
   "Sqrt:out0"
  ],
  "acu_lys_alias": [
   "sqrt"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "sqrt:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Sqrt:out0",
    "sqrt:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "sqrt": {}
  },
  "blob_map": {
   "sqrt": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "floor",
  "src_ops_alias": [
   "Floor"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Floor:in0"
   ]
  ],
  "src_out_tensor": [
   "Floor:out0"
  ],
  "acu_lys_alias": [
   "floor"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "floor:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Floor:out0",
    "floor:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "floor": {}
  },
  "blob_map": {
   "floor": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "rsqrt",
  "src_ops_alias": [
   "Rsqrt"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Rsqrt:in0"
   ]
  ],
  "src_out_tensor": [
   "Rsqrt:out0"
  ],
  "acu_lys_alias": [
   "rsqrt"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "rsqrt:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Rsqrt:out0",
    "rsqrt:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "rsqrt": {}
  },
  "blob_map": {
   "rsqrt": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "ceil",
  "src_ops_alias": [
   "Ceil"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Ceil:in0"
   ]
  ],
  "src_out_tensor": [
   "Ceil:out0"
  ],
  "acu_lys_alias": [
   "ceil"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "ceil:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Ceil:out0",
    "ceil:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "ceil": {}
  },
  "blob_map": {
   "ceil": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "exp",
  "src_ops_alias": [
   "Exp"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Exp:in0"
   ]
  ],
  "src_out_tensor": [
   "Exp:out0"
  ],
  "acu_lys_alias": [
   "exp"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "exp:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Exp:out0",
    "exp:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "exp": {}
  },
  "blob_map": {
   "exp": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "subtract",
  "src_ops_alias": [
   "Sub"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Sub:in0"
   ],
   [
    "I_1:out0",
    "Sub:in1"
   ]
  ],
  "src_out_tensor": [
   "Sub:out0"
  ],
  "acu_lys_alias": [
   "subtract"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "subtract:in0"
   ],
   [
    "I_1:out0",
    "subtract:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Sub:out0",
    "subtract:out0"
   ]
  ],
  "param_map": {
   "subtract": {}
  },
  "blob_map": {
   "subtract": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "add_n_2_add",
  "src_ops_alias": [
   "AddN"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "AddN:in0"
   ],
   [
    "I_1:out0",
    "AddN:in1"
   ]
  ],
  "src_out_tensor": [
   "AddN:out0"
  ],
  "acu_lys_alias": [
   "add"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "add:in0"
   ],
   [
    "I_1:out0",
    "add:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "AddN:out0",
    "add:out0"
   ]
  ],
  "param_map": {
   "add": {}
  },
  "blob_map": {
   "add": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": "self.attr_pick(node['AddN'], 'N', 0) == 2"
 },
 {
  "ruler_name": "biasadd",
  "src_ops_alias": [
   "BiasAdd"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "BiasAdd:in0"
   ],
   [
    "I_1:out0",
    "BiasAdd:in1"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "add"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "add:in0"
   ],
   [
    "I_1:out0",
    "add:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "add:out0"
   ]
  ],
  "param_map": {
   "add": {}
  },
  "blob_map": {
   "add": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "multiply",
  "src_ops_alias": [
   "Mul"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I_1:out0",
    "Mul:in1"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "multiply"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "multiply:in0"
   ],
   [
    "I_1:out0",
    "multiply:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "multiply:out0"
   ]
  ],
  "param_map": {
   "multiply": {}
  },
  "blob_map": {
   "multiply": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_mul_shape_input",
  "src_ops_alias": [
   "Mul"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "Mul:in1"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "multiply"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "multiply:in0"
   ],
   [
    "I:out0",
    "multiply:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "multiply:out0"
   ]
  ],
  "param_map": {
   "multiply": {
    "bias": [
     "BOOL",
     "VALUE",
     false
    ]
   }
  },
  "blob_map": {
   "multiply": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "abs",
  "src_ops_alias": [
   "Abs"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Abs:in0"
   ]
  ],
  "src_out_tensor": [
   "Abs:out0"
  ],
  "acu_lys_alias": [
   "abs"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "abs:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Abs:out0",
    "abs:out0"
   ]
  ],
  "param_map": {
   "abs": {}
  },
  "blob_map": {
   "abs": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "neg",
  "src_ops_alias": [
   "Neg"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Neg:in0"
   ]
  ],
  "src_out_tensor": [
   "Neg:out0"
  ],
  "acu_lys_alias": [
   "neg"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "neg:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Neg:out0",
    "neg:out0"
   ]
  ],
  "param_map": {
   "neg": {}
  },
  "blob_map": {
   "neg": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "clipbyvalue",
  "src_ops_alias": [
   "Maximum",
   "Minimum",
   "C",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "Minimum:out0",
    "Maximum:in0"
   ],
   [
    "C:out0",
    "Maximum:in1"
   ],
   [
    "C_1:out0",
    "Minimum:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Minimum:in0"
   ]
  ],
  "src_out_tensor": [
   "Maximum:out0"
  ],
  "acu_lys_alias": [
   "clipbyvalue"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "clipbyvalue:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Maximum:out0",
    "clipbyvalue:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "clipbyvalue": {
    "clip_value_min": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "clip_value_max": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "blob_map": {
   "clipbyvalue": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "eltwise_max",
  "src_ops_alias": [
   "Maximum"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Maximum:in0"
   ],
   [
    "I_1:out0",
    "Maximum:in1"
   ]
  ],
  "src_out_tensor": [
   "Maximum:out0"
  ],
  "acu_lys_alias": [
   "eltwise"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "eltwise:in0"
   ],
   [
    "I_1:out0",
    "eltwise:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Maximum:out0",
    "eltwise:out0"
   ]
  ],
  "param_map": {
   "eltwise": {
    "operation": [
     "STRING",
     "VALUE",
     "MAX"
    ]
   }
  },
  "blob_map": {
   "eltwise": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "argmax",
  "src_ops_alias": [
   "ArgMax",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "ArgMax:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ArgMax:in0"
   ]
  ],
  "src_out_tensor": [
   "ArgMax:out0"
  ],
  "acu_lys_alias": [
   "argmax"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "argmax:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ArgMax:out0",
    "argmax:out0"
   ]
  ],
  "param_map": {
   "argmax": {}
  },
  "blob_map": {
   "argmax": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "argmin",
  "src_ops_alias": [
   "ArgMin",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "ArgMin:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ArgMin:in0"
   ]
  ],
  "src_out_tensor": [
   "ArgMin:out0"
  ],
  "acu_lys_alias": [
   "argmin"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "argmin:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ArgMin:out0",
    "argmin:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "argmin": {
    "axis": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "argmin": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "placeholder",
  "src_ops_alias": [
   "Placeholder"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Placeholder:in0"
   ]
  ],
  "src_out_tensor": [
   "Placeholder:out0"
  ],
  "acu_lys_alias": [
   "noop"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "noop:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Placeholder:out0",
    "noop:out0"
   ]
  ],
  "param_map": {
   "noop": {}
  },
  "blob_map": {
   "noop": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "identity",
  "src_ops_alias": [
   "Identity"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Identity:in0"
   ]
  ],
  "src_out_tensor": [
   "Identity:out0"
  ],
  "acu_lys_alias": [
   "noop"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "noop:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Identity:out0",
    "noop:out0"
   ]
  ],
  "param_map": {
   "noop": {}
  },
  "blob_map": {
   "noop": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "cast",
  "src_ops_alias": [
   "Cast"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Cast:in0"
   ]
  ],
  "src_out_tensor": [
   "Cast:out0"
  ],
  "acu_lys_alias": [
   "cast"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "cast:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Cast:out0",
    "cast:out0"
   ]
  ],
  "param_map": {
   "cast": {
    "in_data_type": [
     "STRING",
     "CODE",
     "self.attr_pick(node['Cast'], 'SrcT')"
    ],
    "out_data_type": [
     "STRING",
     "CODE",
     "self.tf_type_enum_to_ac_type(self.attr_pick(node['Cast'], 'DstT'))"
    ]
   }
  },
  "blob_map": {
   "cast": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "dropout_train_scale",
  "src_ops_alias": [
   "Mul",
   "RealDiv",
   "Floor",
   "C",
   "Add",
   "Add_1",
   "Mul_1",
   "C_1",
   "RandomUniform",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "Floor:out0",
    "Mul:in1"
   ],
   [
    "Add_1:out0",
    "Add:in1"
   ],
   [
    "Add:out0",
    "Floor:in0"
   ],
   [
    "C_3:out0",
    "RandomUniform:in0"
   ],
   [
    "RandomUniform:out0",
    "Mul_1:in0"
   ],
   [
    "C:out0",
    "RealDiv:in1"
   ],
   [
    "C_1:out0",
    "Add_1:in1"
   ],
   [
    "C_2:out0",
    "Mul_1:in1"
   ],
   [
    "RealDiv:out0",
    "Mul:in0"
   ],
   [
    "Mul_1:out0",
    "Add_1:in0"
   ],
   [
    "C:out0",
    "Add:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "RealDiv:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "dropout"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "dropout:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "dropout:out0"
   ]
  ],
  "param_map": {
   "dropout": {
    "scale_train": [
     "BOOL",
     "VALUE",
     true
    ],
    "ratio": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "dropout": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_deepspeech_1_lstm",
  "src_ops_alias": [
   "TensorArrayGatherV3",
   "TensorArrayV3",
   "Range",
   "Exit",
   "C",
   "C_1",
   "TensorArraySizeV3",
   "C_2",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "TensorArrayWriteV3",
   "Less",
   "Less_1",
   "Enter_1",
   "Identity",
   "Select",
   "Identity_1",
   "Merge_1",
   "Enter_2",
   "Merge_2",
   "Enter_3",
   "Switch_1",
   "GreaterEqual",
   "Enter_4",
   "Mul",
   "Enter_5",
   "NextIteration_1",
   "Enter_6",
   "NextIteration_2",
   "C_3",
   "Enter_7",
   "C_4",
   "Tanh",
   "Sigmoid",
   "C_5",
   "AddV2",
   "C_6",
   "AddV2_1",
   "C_7",
   "Add",
   "Split",
   "Identity_2",
   "C_8",
   "C_9",
   "Mul_1",
   "Mul_2",
   "C_10",
   "BiasAdd",
   "Switch_2",
   "Identity_3",
   "Sigmoid_1",
   "Sigmoid_2",
   "Tanh_1",
   "MatMul",
   "Enter_8",
   "Switch_3",
   "Add_1",
   "ConcatV2",
   "Enter_9",
   "C_11",
   "Merge_3",
   "C_12",
   "TensorArrayReadV3",
   "Identity_4",
   "C_13",
   "C_14",
   "Enter_10",
   "NextIteration_3",
   "Enter_11",
   "Enter_12",
   "Switch_4",
   "C_15",
   "Select_1",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "Merge_4",
   "C_16",
   "Enter_13",
   "NextIteration_4",
   "C_17",
   "Select_2"
  ],
  "src_inter_flow": [
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "Exit:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "C_1:out0",
    "Range:in0"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "C_2:out0",
    "Range:in2"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "Exit:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "Enter_1:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Select:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_1:in0"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Enter_2:out0",
    "Less:in1"
   ],
   [
    "Merge_2:out0",
    "Less_1:in0"
   ],
   [
    "Enter_3:out0",
    "Less_1:in1"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "GreaterEqual:out0",
    "Select:in0"
   ],
   [
    "Enter_4:out0",
    "Select:in1"
   ],
   [
    "Mul:out0",
    "Select:in2"
   ],
   [
    "C:out0",
    "Enter_2:in0"
   ],
   [
    "Enter_5:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Enter_6:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "Merge_2:out0",
    "Switch_1:in0"
   ],
   [
    "C_3:out0",
    "Enter_3:in0"
   ],
   [
    "Identity:out0",
    "GreaterEqual:in0"
   ],
   [
    "Enter_7:out0",
    "GreaterEqual:in1"
   ],
   [
    "C_4:out0",
    "Enter_4:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in0"
   ],
   [
    "Sigmoid:out0",
    "Mul:in1"
   ],
   [
    "C_5:out0",
    "Enter_5:in0"
   ],
   [
    "AddV2:out0",
    "NextIteration_1:in0"
   ],
   [
    "C_6:out0",
    "Enter_6:in0"
   ],
   [
    "AddV2_1:out0",
    "NextIteration_2:in0"
   ],
   [
    "C_7:out0",
    "Enter_7:in0"
   ],
   [
    "Add:out0",
    "Tanh:in0"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "Identity:out0",
    "AddV2_1:in0"
   ],
   [
    "Identity_2:out0",
    "AddV2:in0"
   ],
   [
    "C_8:out0",
    "AddV2:in1"
   ],
   [
    "C_9:out0",
    "AddV2_1:in1"
   ],
   [
    "Mul_1:out0",
    "Add:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "C_10:out0",
    "Split:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "Identity_2:out4096",
    "C_8:in0"
   ],
   [
    "Identity_2:out4096",
    "C_9:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Identity_3:out0",
    "Mul_1:in0"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in1"
   ],
   [
    "Merge_1:out0",
    "Switch_2:in0"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Identity_2:out4096",
    "C_10:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Enter_8:out0",
    "BiasAdd:in1"
   ],
   [
    "Switch_3:out1",
    "Identity_3:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "Add_1:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Enter_9:out0",
    "MatMul:in1"
   ],
   [
    "C_11:out0",
    "Enter_8:in0"
   ],
   [
    "Split:out2",
    "Add_1:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_3:in0"
   ],
   [
    "C_12:out0",
    "Add_1:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Identity_4:out0",
    "ConcatV2:in1"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "C_13:out0",
    "ConcatV2:in2"
   ],
   [
    "C_14:out0",
    "Enter_9:in0"
   ],
   [
    "Identity_2:out4096",
    "C_12:in0"
   ],
   [
    "Enter_10:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "Identity_2:out4096",
    "C_13:in0"
   ],
   [
    "Enter_11:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_12:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "Switch_4:out1",
    "Identity_4:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_4:in1"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "C_15:out0",
    "Enter_10:in0"
   ],
   [
    "GreaterEqual:out0",
    "Select_1:in0"
   ],
   [
    "Select_1:out0",
    "NextIteration_3:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_11:in0"
   ],
   [
    "Add:out0",
    "Select_1:in2"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_12:in0"
   ],
   [
    "Identity_3:out0",
    "Select_1:in1"
   ],
   [
    "Merge_4:out0",
    "Switch_4:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "C_16:out0",
    "TensorArrayScatterV3:in1"
   ],
   [
    "GreaterEqual:out0",
    "Select_2:in0"
   ],
   [
    "Mul:out0",
    "Select_2:in2"
   ],
   [
    "Enter_13:out0",
    "Merge_4:in0"
   ],
   [
    "NextIteration_4:out0",
    "Merge_4:in1"
   ],
   [
    "C_17:out0",
    "Enter_13:in0"
   ],
   [
    "Select_2:out0",
    "NextIteration_4:in0"
   ],
   [
    "Identity_4:out0",
    "Select_2:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayGatherV3:out0"
  ],
  "acu_lys_alias": [
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstm:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayGatherV3:out0",
    "lstm:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_12:out0'])"
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_14:out0'])[1] / 4"
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_14:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_14:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_14:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_14:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_14:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_14:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_14:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_14:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_11:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_11:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_11:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_11:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "lstm_new",
  "src_ops_alias": [
   "TensorArrayGatherV2",
   "TensorArrayV2",
   "Range",
   "Exit",
   "StridedSlice",
   "C",
   "TensorArraySizeV2",
   "C_1",
   "Switch",
   "C_2",
   "C_3",
   "C_4",
   "C_5",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "Less",
   "C_6",
   "TensorArrayWriteV2",
   "Merge_1",
   "Enter_1",
   "Enter_2",
   "Identity",
   "Mul",
   "Identity_1",
   "Enter_3",
   "NextIteration_1",
   "Switch_1",
   "Tanh",
   "Sigmoid",
   "C_7",
   "Add",
   "Add_1",
   "Split",
   "C_8",
   "Mul_1",
   "Mul_2",
   "C_9",
   "Add_2",
   "Identity_2",
   "Sigmoid_1",
   "Sigmoid_2",
   "Tanh_1",
   "MatMul",
   "Enter_4",
   "Switch_2",
   "Add_3",
   "Concat",
   "Enter_5",
   "Identity_3",
   "Merge_2",
   "C_10",
   "C_11",
   "TensorArrayReadV2",
   "Identity_4",
   "Identity_5",
   "C_12",
   "Enter_6",
   "NextIteration_2",
   "Enter_7",
   "Enter_8",
   "Switch_3",
   "C_13",
   "Fill",
   "TensorArrayV2_1",
   "TensorArrayScatterV2",
   "Merge_3",
   "Pack",
   "C_14",
   "Range_1",
   "C_15",
   "Enter_9",
   "NextIteration_3",
   "StridedSlice_1",
   "C_16",
   "C_17",
   "StridedSlice_2",
   "C_18",
   "Fill_1",
   "C_19",
   "C_20",
   "C_21",
   "C_22",
   "C_23",
   "C_24",
   "C_25",
   "C_26",
   "Pack_1",
   "C_27",
   "C_28"
  ],
  "src_inter_flow": [
   [
    "TensorArrayV2:out0",
    "TensorArrayGatherV2:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV2:in1"
   ],
   [
    "Exit:out0",
    "TensorArrayGatherV2:in2"
   ],
   [
    "StridedSlice:out0",
    "TensorArrayV2:in0"
   ],
   [
    "C:out0",
    "Range:in0"
   ],
   [
    "TensorArraySizeV2:out0",
    "Range:in1"
   ],
   [
    "C_1:out0",
    "Range:in2"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "C_2:out0",
    "StridedSlice:in0"
   ],
   [
    "C_3:out0",
    "StridedSlice:in1"
   ],
   [
    "C_4:out0",
    "StridedSlice:in2"
   ],
   [
    "C_5:out0",
    "StridedSlice:in3"
   ],
   [
    "TensorArrayV2:out0",
    "TensorArraySizeV2:in0"
   ],
   [
    "Exit:out0",
    "TensorArraySizeV2:in1"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "Less:out0",
    "LoopCond:in0"
   ],
   [
    "C_6:out0",
    "Enter:in0"
   ],
   [
    "TensorArrayWriteV2:out0",
    "NextIteration:in0"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Enter_1:out0",
    "Less:in1"
   ],
   [
    "Enter_2:out0",
    "TensorArrayWriteV2:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV2:in1"
   ],
   [
    "StridedSlice:out0",
    "Enter_1:in0"
   ],
   [
    "Mul:out0",
    "TensorArrayWriteV2:in2"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV2:in3"
   ],
   [
    "TensorArrayV2:out0",
    "Enter_2:in0"
   ],
   [
    "Enter_3:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in0"
   ],
   [
    "Sigmoid:out0",
    "Mul:in1"
   ],
   [
    "C_7:out0",
    "Enter_3:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Merge_1:out0",
    "Switch_1:in0"
   ],
   [
    "Add:out0",
    "NextIteration_1:in0"
   ],
   [
    "Add_1:out0",
    "Tanh:in0"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "Identity:out0",
    "Add:in0"
   ],
   [
    "C_8:out0",
    "Add:in1"
   ],
   [
    "Mul_1:out0",
    "Add_1:in0"
   ],
   [
    "Mul_2:out0",
    "Add_1:in1"
   ],
   [
    "Identity:out4096",
    "C_8:in0"
   ],
   [
    "C_9:out0",
    "Split:in0"
   ],
   [
    "Add_2:out0",
    "Split:in1"
   ],
   [
    "Identity_2:out0",
    "Mul_1:in0"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in1"
   ],
   [
    "Identity:out4096",
    "C_9:in0"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "MatMul:out0",
    "Add_2:in0"
   ],
   [
    "Enter_4:out0",
    "Add_2:in1"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "Add_3:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Concat:out0",
    "MatMul:in0"
   ],
   [
    "Enter_5:out0",
    "MatMul:in1"
   ],
   [
    "Identity_3:out0",
    "Enter_4:in0"
   ],
   [
    "Split:out2",
    "Add_3:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_2:in0"
   ],
   [
    "C_10:out0",
    "Add_3:in1"
   ],
   [
    "C_11:out0",
    "Concat:in0"
   ],
   [
    "TensorArrayReadV2:out0",
    "Concat:in1"
   ],
   [
    "Identity_4:out0",
    "Concat:in2"
   ],
   [
    "Identity:out4096",
    "C_10:in0"
   ],
   [
    "Identity_5:out0",
    "Enter_5:in0"
   ],
   [
    "Identity:out4096",
    "C_11:in0"
   ],
   [
    "C_12:out0",
    "Identity_3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV2:in1"
   ],
   [
    "Enter_6:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "Enter_7:out0",
    "TensorArrayReadV2:in0"
   ],
   [
    "Enter_8:out0",
    "TensorArrayReadV2:in2"
   ],
   [
    "Switch_3:out1",
    "Identity_4:in0"
   ],
   [
    "Add_1:out0",
    "NextIteration_2:in0"
   ],
   [
    "C_13:out0",
    "Identity_5:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "Fill:out0",
    "Enter_6:in0"
   ],
   [
    "StridedSlice:out0",
    "TensorArrayV2_1:in0"
   ],
   [
    "TensorArrayV2_1:out0",
    "Enter_7:in0"
   ],
   [
    "TensorArrayScatterV2:out0",
    "Enter_8:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_3:in0"
   ],
   [
    "Pack:out0",
    "Fill:in0"
   ],
   [
    "C_14:out0",
    "Fill:in1"
   ],
   [
    "TensorArrayV2_1:out0",
    "TensorArrayScatterV2:in0"
   ],
   [
    "Range_1:out0",
    "TensorArrayScatterV2:in1"
   ],
   [
    "C_15:out0",
    "TensorArrayScatterV2:in3"
   ],
   [
    "Enter_9:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "Mul:out0",
    "NextIteration_3:in0"
   ],
   [
    "StridedSlice_1:out0",
    "Pack:in0"
   ],
   [
    "C_16:out0",
    "Pack:in1"
   ],
   [
    "C_17:out0",
    "Range_1:in0"
   ],
   [
    "StridedSlice_2:out0",
    "Range_1:in1"
   ],
   [
    "C_18:out0",
    "Range_1:in2"
   ],
   [
    "Fill_1:out0",
    "Enter_9:in0"
   ],
   [
    "C_19:out0",
    "StridedSlice_1:in0"
   ],
   [
    "C_20:out0",
    "StridedSlice_1:in1"
   ],
   [
    "C_21:out0",
    "StridedSlice_1:in2"
   ],
   [
    "C_22:out0",
    "StridedSlice_1:in3"
   ],
   [
    "C_23:out0",
    "StridedSlice_2:in0"
   ],
   [
    "C_24:out0",
    "StridedSlice_2:in1"
   ],
   [
    "C_25:out0",
    "StridedSlice_2:in2"
   ],
   [
    "C_26:out0",
    "StridedSlice_2:in3"
   ],
   [
    "Pack_1:out0",
    "Fill_1:in0"
   ],
   [
    "C_27:out0",
    "Fill_1:in1"
   ],
   [
    "StridedSlice_1:out0",
    "Pack_1:in0"
   ],
   [
    "C_28:out0",
    "Pack_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV2:in2"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayGatherV2:out0"
  ],
  "acu_lys_alias": [
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstm:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayGatherV2:out0",
    "lstm:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_10:out0'])"
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_13:out0'])[1] / 4"
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_13:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_13:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_13:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_13:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_13:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_13:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_13:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_13:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_12:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_12:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_12:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_12:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "lstm_hidden_cell",
  "src_ops_alias": [
   "TensorArrayGatherV3",
   "Exit",
   "Exit_1",
   "TensorArrayV3",
   "Range",
   "Exit_2",
   "Switch",
   "Switch_1",
   "C",
   "C_1",
   "TensorArraySizeV3",
   "C_2",
   "Switch_2",
   "Merge",
   "LoopCond",
   "Merge_1",
   "Merge_2",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "Enter_1",
   "NextIteration_1",
   "Enter_2",
   "NextIteration_2",
   "Add",
   "Less",
   "Less_1",
   "Mul",
   "TensorArrayWriteV3",
   "Mul_1",
   "Mul_2",
   "Merge_3",
   "Enter_3",
   "Merge_4",
   "Enter_4",
   "Sigmoid",
   "Tanh",
   "Enter_5",
   "Identity",
   "Identity_1",
   "Sigmoid_1",
   "Identity_2",
   "Sigmoid_2",
   "Tanh_1",
   "Enter_6",
   "NextIteration_3",
   "Enter_7",
   "NextIteration_4",
   "C_3",
   "Split",
   "Switch_3",
   "Add_1",
   "C_4",
   "Add_2",
   "C_5",
   "Add_3",
   "C_6",
   "BiasAdd",
   "C_7",
   "Identity_3",
   "C_8",
   "C_9",
   "MatMul",
   "Enter_8",
   "Switch_4",
   "ConcatV2",
   "Enter_9",
   "C_10",
   "TensorArrayReadV3",
   "Identity_4",
   "C_11",
   "C_12",
   "Enter_10",
   "Enter_11",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "C_13"
  ],
  "src_inter_flow": [
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "Exit_2:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "Switch_1:out0",
    "Exit_1:in0"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "C_1:out0",
    "Range:in0"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "C_2:out0",
    "Range:in2"
   ],
   [
    "Switch_2:out0",
    "Exit_2:in0"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Merge_1:out0",
    "Switch_1:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "Exit_2:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Merge_2:out0",
    "Switch_2:in0"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "Enter_1:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "Enter_2:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "Add:out0",
    "NextIteration:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "TensorArrayV3:out1",
    "Enter_2:in0"
   ],
   [
    "Mul:out0",
    "NextIteration_1:in0"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration_2:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "Merge_3:out0",
    "Less:in0"
   ],
   [
    "Enter_3:out0",
    "Less:in1"
   ],
   [
    "Merge_4:out0",
    "Less_1:in0"
   ],
   [
    "Enter_4:out0",
    "Less_1:in1"
   ],
   [
    "Sigmoid:out0",
    "Mul:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "Mul:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Enter_5:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "C:out0",
    "Enter_3:in0"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in0"
   ],
   [
    "Identity_2:out0",
    "Mul_1:in1"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "Enter_6:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_5:in0"
   ],
   [
    "Enter_7:out0",
    "Merge_4:in0"
   ],
   [
    "NextIteration_4:out0",
    "Merge_4:in1"
   ],
   [
    "Add:out0",
    "Tanh:in0"
   ],
   [
    "C_3:out0",
    "Enter_4:in0"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "Switch_2:out1",
    "Identity_1:in0"
   ],
   [
    "Switch:out1",
    "Identity_2:in0"
   ],
   [
    "Switch_3:out1",
    "Identity:in0"
   ],
   [
    "Add_1:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "C_4:out0",
    "Enter_6:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_3:in0"
   ],
   [
    "C_5:out0",
    "Enter_7:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "Add_3:out0",
    "NextIteration_4:in0"
   ],
   [
    "Merge_4:out0",
    "Switch_3:in0"
   ],
   [
    "C_6:out0",
    "Split:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Split:out2",
    "Add_1:in0"
   ],
   [
    "C_7:out0",
    "Add_1:in1"
   ],
   [
    "Identity:out0",
    "Add_3:in0"
   ],
   [
    "Identity_3:out0",
    "Add_2:in0"
   ],
   [
    "C_8:out0",
    "Add_2:in1"
   ],
   [
    "C_9:out0",
    "Add_3:in1"
   ],
   [
    "Identity_3:out4096",
    "C_6:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Enter_8:out0",
    "BiasAdd:in1"
   ],
   [
    "Identity_3:out4096",
    "C_7:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_4:in1"
   ],
   [
    "Switch_4:out1",
    "Identity_3:in0"
   ],
   [
    "Identity_3:out4096",
    "C_8:in0"
   ],
   [
    "Identity_3:out4096",
    "C_9:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_4:in0"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Enter_9:out0",
    "MatMul:in1"
   ],
   [
    "Switch_1:out1",
    "Identity_4:in0"
   ],
   [
    "C_10:out0",
    "Enter_8:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Identity_4:out0",
    "ConcatV2:in1"
   ],
   [
    "C_11:out0",
    "ConcatV2:in2"
   ],
   [
    "C_12:out0",
    "Enter_9:in0"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "Enter_10:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_11:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "Identity_3:out4096",
    "C_11:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_10:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_11:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "C_13:out0",
    "TensorArrayScatterV3:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ],
   [
    "I_1:out0",
    "Enter_1:in0"
   ],
   [
    "I_2:out0",
    "Enter:in0"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayGatherV3:out0",
   "Exit:out0",
   "Exit_1:out0"
  ],
  "acu_lys_alias": [
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstm:in0"
   ],
   [
    "I_1:out0",
    "lstm:in1"
   ],
   [
    "I_2:out0",
    "lstm:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayGatherV3:out0",
    "lstm:out0"
   ],
   [
    "Exit:out0",
    "lstm:out2"
   ],
   [
    "Exit_1:out0",
    "lstm:out1"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_7:out0'])"
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter:out0'])[1]"
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_10:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_10:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_10:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_10:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "lstm",
  "src_ops_alias": [
   "TensorArrayGatherV3",
   "TensorArrayV3",
   "Range",
   "Exit",
   "C",
   "C_1",
   "TensorArraySizeV3",
   "C_2",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "Less",
   "TensorArrayWriteV3",
   "Merge_1",
   "Enter_1",
   "Enter_2",
   "Identity",
   "Mul",
   "Identity_1",
   "Enter_3",
   "NextIteration_1",
   "Switch_1",
   "Sigmoid",
   "Tanh",
   "C_3",
   "Add",
   "Split",
   "Add_1",
   "C_4",
   "C_5",
   "BiasAdd",
   "Mul_1",
   "Mul_2",
   "MatMul",
   "Enter_4",
   "Sigmoid_1",
   "Identity_2",
   "Sigmoid_2",
   "Tanh_1",
   "ConcatV2",
   "Enter_5",
   "C_6",
   "Add_2",
   "Switch_2",
   "TensorArrayReadV3",
   "Identity_3",
   "C_7",
   "C_8",
   "C_9",
   "Merge_2",
   "Enter_6",
   "Enter_7",
   "Switch_3",
   "Enter_8",
   "NextIteration_2",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "Merge_3",
   "C_10",
   "C_11",
   "Enter_9",
   "NextIteration_3",
   "C_12"
  ],
  "src_inter_flow": [
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "Split:out2",
    "Add_2:in0"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "Mul_2:out0",
    "Add_1:in1"
   ],
   [
    "Mul:out0",
    "NextIteration_3:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_2:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "C_10:out0",
    "Enter_8:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_7:in0"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "Identity_3:out0",
    "ConcatV2:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "Exit:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_6:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "C:out0",
    "Enter_1:in0"
   ],
   [
    "Identity:out4096",
    "C_5:in0"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "C_4:out0",
    "Add:in1"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "Identity:out0",
    "Add:in0"
   ],
   [
    "Enter_4:out0",
    "BiasAdd:in1"
   ],
   [
    "C_1:out0",
    "Range:in0"
   ],
   [
    "Switch_3:out1",
    "Identity_3:in0"
   ],
   [
    "C_9:out0",
    "Add_2:in1"
   ],
   [
    "C_5:out0",
    "Split:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Add_1:out0",
    "NextIteration_2:in0"
   ],
   [
    "Enter_6:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "C_6:out0",
    "Enter_4:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "Merge_2:out0",
    "Switch_2:in0"
   ],
   [
    "Identity_2:out0",
    "Mul_1:in1"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "C_11:out0",
    "TensorArrayScatterV3:in1"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "Enter_2:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "Identity:out4096",
    "C_7:in0"
   ],
   [
    "Enter_5:out0",
    "MatMul:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "C_7:out0",
    "ConcatV2:in2"
   ],
   [
    "C_8:out0",
    "Enter_5:in0"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "Mul_1:out0",
    "Add_1:in0"
   ],
   [
    "Enter_8:out0",
    "Merge_2:in0"
   ],
   [
    "Enter_7:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "Sigmoid:out0",
    "Mul:in0"
   ],
   [
    "Mul:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "Enter_1:out0",
    "Less:in1"
   ],
   [
    "Less:out0",
    "LoopCond:in0"
   ],
   [
    "Exit:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "Identity:out4096",
    "C_4:in0"
   ],
   [
    "C_2:out0",
    "Range:in2"
   ],
   [
    "Add_2:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Add:out0",
    "NextIteration_1:in0"
   ],
   [
    "Enter_9:out0",
    "Merge_3:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "Identity:out4096",
    "C_9:in0"
   ],
   [
    "C_3:out0",
    "Enter_3:in0"
   ],
   [
    "Merge_1:out0",
    "Switch_1:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Enter_3:out0",
    "Merge_1:in0"
   ],
   [
    "Add_1:out0",
    "Tanh:in0"
   ],
   [
    "C_12:out0",
    "Enter_9:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayGatherV3:out0"
  ],
  "acu_lys_alias": [
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstm:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayGatherV3:out0",
    "lstm:out0"
   ]
  ],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0'])"
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter_5:out0'])[1] / 4"
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_6:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_6:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_6:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_6:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "tensor_dot_bias_add",
  "src_ops_alias": [
   "BiasAdd",
   "Reshape",
   "C",
   "MatMul",
   "C_1",
   "Reshape_1",
   "C_2",
   "Transpose",
   "C_3",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "C_4:out0",
    "Transpose:in1"
   ],
   [
    "C_2:out0",
    "MatMul:in1"
   ],
   [
    "Transpose:out0",
    "Reshape_1:in0"
   ],
   [
    "MatMul:out0",
    "Reshape:in0"
   ],
   [
    "Reshape:out0",
    "BiasAdd:in0"
   ],
   [
    "Reshape_1:out0",
    "MatMul:in0"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "C_1:out0",
    "Reshape:in1"
   ],
   [
    "C_3:out0",
    "Reshape_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Transpose:in0"
   ]
  ],
  "src_out_tensor": [
   "BiasAdd:out0"
  ],
  "acu_lys_alias": [
   "fullconnect"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "fullconnect:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "BiasAdd:out0",
    "fullconnect:out0"
   ]
  ],
  "param_map": {
   "fullconnect": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[0]"
    ],
    "bias": [
     "BOOL",
     "PYFUNC",
     [
      "def r_fc_tensor_dot_rule_get_param_bias(self, node, tensor):",
      "    return True",
      ""
     ]
    ],
    "axis": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"reshape_param\": \"C_3:out0\", \"transpose_out\": \"Transpose:out0\"}",
      "def r_fc_tensor_dot_rule_get_param_axis(self, node, tensor, reshape_param, transpose_out):",
      "    import functools",
      "    shape_before_matmul = self.tensor_to_numpy(tensor[reshape_param])",
      "    input_shape = self.shape_pick(tensor[transpose_out])",
      "",
      "    multiply = lambda x, y: x * y",
      "    positive_pos = -1",
      "    positive_pos_actural_value = -1",
      "    for i in range(len(shape_before_matmul)):",
      "        if shape_before_matmul[i] == -1:",
      "            positive_pos = i",
      "            break",
      "    if positive_pos > -1:",
      "        positive_pos_actural_value = functools.reduce(multiply, input_shape)",
      "        for i in range(len(shape_before_matmul)):",
      "            if shape_before_matmul[i] != -1:",
      "                positive_pos_actural_value = positive_pos_actural_value / shape_before_matmul[i]",
      "    if positive_pos_actural_value > 0:",
      "        shape_before_matmul[positive_pos] = positive_pos_actural_value",
      "",
      "    axes = []",
      "    for i in range(1, len(input_shape)):",
      "        first = input_shape[:i]",
      "        second = input_shape[i:]",
      "        first = functools.reduce(multiply, first)",
      "        second = functools.reduce(multiply, second)",
      "        if first == shape_before_matmul[0] and second == shape_before_matmul[1]:",
      "            axes.append(i)",
      "",
      "    if len(axes) == 0:",
      "        raise ValueError(\"Cannot find axis from {} to {}\".format(input_shape, shape_before_matmul))",
      "    elif len(axes) > 1:",
      "        print(\"Warning: Find multiple axis from {} to {}\".format(input_shape, shape_before_matmul))",
      "",
      "    if len(axes) == 0 or len(axes) > 1:",
      "        return -1",
      "    else:",
      "        return axes[0]",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "fullconnect": {
    "weight": [
     "PYFUNC",
     [
      "def r_fc_tensor_dot_rule_get_weight(self, node, tensor):",
      "    return self.tensor_to_numpy(tensor['C_2:out0'])",
      ""
     ]
    ],
    "bias": [
     "PYFUNC",
     [
      "def r_fc_tensor_dot_rule_get_bias(self, node, tensor):",
      "    return self.tensor_to_numpy(tensor['C:out0'])",
      ""
     ]
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": [
   "def r_fc_tensor_dot_rule_pre_condition(self, node, tensor):",
   "    return True",
   ""
  ]
 },
 {
  "ruler_name": "tensor_dot",
  "src_ops_alias": [
   "Reshape",
   "MatMul",
   "C",
   "Reshape_1",
   "C_1",
   "Transpose",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "MatMul:out0",
    "Reshape:in0"
   ],
   [
    "C:out0",
    "Reshape:in1"
   ],
   [
    "Reshape_1:out0",
    "MatMul:in0"
   ],
   [
    "C_1:out0",
    "MatMul:in1"
   ],
   [
    "Transpose:out0",
    "Reshape_1:in0"
   ],
   [
    "C_2:out0",
    "Reshape_1:in1"
   ],
   [
    "C_3:out0",
    "Transpose:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Transpose:in0"
   ]
  ],
  "src_out_tensor": [
   "Reshape:out0"
  ],
  "acu_lys_alias": [
   "fullconnect"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "fullconnect:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Reshape:out0",
    "fullconnect:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "fullconnect": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ],
    "axis": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"reshape_param\": \"C_2:out0\", \"transpose_out\": \"Transpose:out0\"}",
      "def r_fc_tensor_dot_rule_get_param_axis(self, node, tensor, reshape_param, transpose_out):",
      "    import functools",
      "    shape_before_matmul = self.tensor_to_numpy(tensor[reshape_param])",
      "    input_shape = self.shape_pick(tensor[transpose_out])",
      "",
      "    multiply = lambda x, y: x * y",
      "    positive_pos = -1",
      "    positive_pos_actural_value = -1",
      "    for i in range(len(shape_before_matmul)):",
      "        if shape_before_matmul[i] == -1:",
      "            positive_pos = i",
      "            break",
      "    if positive_pos > -1:",
      "        positive_pos_actural_value = functools.reduce(multiply, input_shape)",
      "        for i in range(len(shape_before_matmul)):",
      "            if shape_before_matmul[i] != -1:",
      "                positive_pos_actural_value = positive_pos_actural_value / shape_before_matmul[i]",
      "    if positive_pos_actural_value > 0:",
      "        shape_before_matmul[positive_pos] = positive_pos_actural_value",
      "",
      "    axes = []",
      "    for i in range(1, len(input_shape)):",
      "        first = input_shape[:i]",
      "        second = input_shape[i:]",
      "        first = functools.reduce(multiply, first)",
      "        second = functools.reduce(multiply, second)",
      "        if first == shape_before_matmul[0] and second == shape_before_matmul[1]:",
      "            axes.append(i)",
      "",
      "    if len(axes) == 0:",
      "        raise ValueError(\"Cannot find axis from {} to {}\".format(input_shape, shape_before_matmul))",
      "    elif len(axes) > 1:",
      "        print(\"Warning: Find multiple axis from {} to {}\".format(input_shape, shape_before_matmul))",
      "",
      "    if len(axes) == 0 or len(axes) > 1:",
      "        return -1",
      "    else:",
      "        return axes[0]",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "fullconnect": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "lstmunit",
  "src_ops_alias": [
   "Mul",
   "Tanh",
   "Sigmoid",
   "Add",
   "Split",
   "Mul_1",
   "Mul_2",
   "C",
   "BiasAdd",
   "Sigmoid_1",
   "Sigmoid_2",
   "Tanh_1",
   "MatMul",
   "C_1",
   "Add_1",
   "ConcatV2",
   "C_2",
   "C_3",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in0"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in1"
   ],
   [
    "Split:out2",
    "Add_1:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "C:out0",
    "Split:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "C_4:out0",
    "ConcatV2:in2"
   ],
   [
    "Add_1:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Sigmoid:out0",
    "Mul:in1"
   ],
   [
    "C_2:out0",
    "MatMul:in1"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "C_3:out0",
    "Add_1:in1"
   ],
   [
    "C_1:out0",
    "BiasAdd:in1"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Add:out0",
    "Tanh:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ConcatV2:in0"
   ],
   [
    "I_1:out0",
    "ConcatV2:in1"
   ],
   [
    "I_2:out0",
    "Mul_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0",
   "Add:out0"
  ],
  "acu_lys_alias": [
   "lstmunit"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstmunit:in0"
   ],
   [
    "I_1:out0",
    "lstmunit:in1"
   ],
   [
    "I_2:out0",
    "lstmunit:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "lstmunit:out0"
   ],
   [
    "Mul:out0",
    "lstmunit:out1"
   ],
   [
    "Add:out0",
    "lstmunit:out2"
   ]
  ],
  "param_map": {
   "lstmunit": {
    "weights": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_2:out0\"}",
      "def r_lstmunit_rule_get_param_weights(self, node, tensor, tensor_name): # 'C_2:out0'",
      "    return self.shape_pick(tensor[tensor_name])[1] / 4",
      ""
     ]
    ],
    "num_proj": [
     "ORIGIN",
     "VALUE",
     null
    ],
    "forget_bias": [
     "FLOAT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_3:out0\"}",
      "def r_lstmunit_rule_get_param_forget_bias(self, node, tensor, tensor_name): #",
      "    forget_bias = self.tensor_to_numpy(tensor[tensor_name])",
      "    return forget_bias",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "lstmunit": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_2:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_2:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_2:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_2:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_2:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_2:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_2:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_2:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_1:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_1:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_1:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_1:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "lstmunit_x",
  "src_ops_alias": [
   "Slice",
   "Slice_1",
   "Pack",
   "C",
   "C_1",
   "Pack_1",
   "C_2",
   "C_3",
   "AddV2",
   "Mul",
   "Mul_1",
   "Mul_2",
   "Sigmoid",
   "Tanh",
   "Sigmoid_1",
   "Sigmoid_2",
   "Tanh_1",
   "Split",
   "AddV2_1",
   "C_4",
   "BiasAdd",
   "C_5",
   "MatMul",
   "C_6",
   "ConcatV2",
   "C_7",
   "C_8"
  ],
  "src_inter_flow": [
   [
    "Pack:out0",
    "Slice:in0"
   ],
   [
    "C:out0",
    "Slice:in1"
   ],
   [
    "C_1:out0",
    "Slice:in2"
   ],
   [
    "Pack_1:out0",
    "Slice_1:in0"
   ],
   [
    "C_2:out0",
    "Slice_1:in1"
   ],
   [
    "C_3:out0",
    "Slice_1:in2"
   ],
   [
    "AddV2:out0",
    "Pack:in0"
   ],
   [
    "Mul:out0",
    "Pack:in1"
   ],
   [
    "AddV2:out0",
    "Pack_1:in0"
   ],
   [
    "Mul:out0",
    "Pack_1:in1"
   ],
   [
    "Mul_1:out0",
    "AddV2:in0"
   ],
   [
    "Mul_2:out0",
    "AddV2:in1"
   ],
   [
    "Sigmoid:out0",
    "Mul:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in0"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "AddV2:out0",
    "Tanh:in0"
   ],
   [
    "AddV2_1:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "C_4:out0",
    "Split:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Split:out2",
    "AddV2_1:in0"
   ],
   [
    "C_5:out0",
    "AddV2_1:in1"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "C_6:out0",
    "BiasAdd:in1"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "C_7:out0",
    "MatMul:in1"
   ],
   [
    "C_8:out0",
    "ConcatV2:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ConcatV2:in0"
   ],
   [
    "I_1:out0",
    "ConcatV2:in1"
   ],
   [
    "I_2:out0",
    "Mul_1:in1"
   ]
  ],
  "src_out_tensor": [
   "Slice:out0",
   "Slice_1:out0"
  ],
  "acu_lys_alias": [
   "lstmunit"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstmunit:in0"
   ],
   [
    "I_1:out0",
    "lstmunit:in1"
   ],
   [
    "I_2:out0",
    "lstmunit:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Slice_1:out0",
    "lstmunit:out0"
   ],
   [
    "Slice_1:out0",
    "lstmunit:out1"
   ],
   [
    "Slice:out0",
    "lstmunit:out2"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstmunit": {
    "weights": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_7:out0\"}",
      "def r_lstmunit_rule_get_param_weights(self, node, tensor, tensor_name): # 'C_2:out0'",
      "    return self.shape_pick(tensor[tensor_name])[1] / 4",
      ""
     ]
    ],
    "num_proj": [
     "ORIGIN",
     "VALUE",
     null
    ],
    "forget_bias": [
     "FLOAT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_5:out0\"}",
      "def r_lstmunit_rule_get_param_forget_bias(self, node, tensor, tensor_name): #",
      "    forget_bias = self.tensor_to_numpy(tensor[tensor_name])",
      "    return forget_bias",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "lstmunit": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_7:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_7:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_7:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_7:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_7:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_7:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_7:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_7:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_6:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_6:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_6:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_6:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "static_lstmunit",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Mul_1",
   "Mul_2",
   "Sigmoid",
   "Tanh",
   "Sigmoid_1",
   "Sigmoid_2",
   "Tanh_1",
   "Split",
   "Add_1",
   "C",
   "BiasAdd",
   "C_1",
   "MatMul",
   "C_2",
   "ConcatV2",
   "C_3",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Mul_1:out0",
    "Add:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "Sigmoid:out0",
    "Mul:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in0"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "Add:out0",
    "Tanh:in0"
   ],
   [
    "Add_1:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "C:out0",
    "Split:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Split:out2",
    "Add_1:in0"
   ],
   [
    "C_1:out0",
    "Add_1:in1"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "C_2:out0",
    "BiasAdd:in1"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "C_3:out0",
    "MatMul:in1"
   ],
   [
    "C_4:out0",
    "ConcatV2:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ConcatV2:in0"
   ],
   [
    "I_1:out0",
    "ConcatV2:in1"
   ],
   [
    "I_2:out0",
    "Mul_1:in1"
   ]
  ],
  "src_out_tensor": [
   "Add:out0",
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "lstmunit"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstmunit:in0"
   ],
   [
    "I_1:out0",
    "lstmunit:in1"
   ],
   [
    "I_2:out0",
    "lstmunit:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "lstmunit:out0"
   ],
   [
    "Mul:out0",
    "lstmunit:out1"
   ],
   [
    "Add:out0",
    "lstmunit:out2"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstmunit": {
    "weights": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_3:out0\"}",
      "def r_lstmunit_rule_get_param_weights(self, node, tensor, tensor_name): # 'C_2:out0'",
      "    return self.shape_pick(tensor[tensor_name])[1] / 4",
      ""
     ]
    ],
    "num_proj": [
     "ORIGIN",
     "VALUE",
     null
    ],
    "forget_bias": [
     "FLOAT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_1:out0\"}",
      "def r_lstmunit_rule_get_param_forget_bias(self, node, tensor, tensor_name): #",
      "    forget_bias = self.tensor_to_numpy(tensor[tensor_name])",
      "    return forget_bias",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "lstmunit": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_2:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_2:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_2:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_2:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": "len(self.shape_pick(tensor['C_2:out0'])) < 2 and self.attr_pick(node['MatMul'], 'transpose_a', False) == False"
 },
 {
  "ruler_name": "r_lstmunit_rule_2_0_0_x",
  "src_ops_alias": [
   "Mul",
   "AddV2",
   "Sigmoid",
   "Tanh",
   "Mul_1",
   "Mul_2",
   "Split",
   "Sigmoid_1",
   "Sigmoid_2",
   "Tanh_1",
   "C",
   "BiasAdd",
   "AddV2_1",
   "MatMul",
   "C_1",
   "C_2",
   "ConcatV2",
   "C_3",
   "C_4"
  ],
  "src_inter_flow": [
   [
    "Sigmoid:out0",
    "Mul:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "Mul_1:out0",
    "AddV2:in0"
   ],
   [
    "Mul_2:out0",
    "AddV2:in1"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "AddV2:out0",
    "Tanh:in0"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in0"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "C:out0",
    "Split:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "AddV2_1:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "C_1:out0",
    "BiasAdd:in1"
   ],
   [
    "Split:out2",
    "AddV2_1:in0"
   ],
   [
    "C_2:out0",
    "AddV2_1:in1"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "C_3:out0",
    "MatMul:in1"
   ],
   [
    "C_4:out0",
    "ConcatV2:in2"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ConcatV2:in0"
   ],
   [
    "I_1:out0",
    "ConcatV2:in1"
   ],
   [
    "I_2:out0",
    "Mul_1:in1"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0",
   "Mul:out0",
   "AddV2:out0"
  ],
  "acu_lys_alias": [
   "lstmunit"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstmunit:in0"
   ],
   [
    "I_1:out0",
    "lstmunit:in1"
   ],
   [
    "I_2:out0",
    "lstmunit:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "lstmunit:out0"
   ],
   [
    "Mul:out0",
    "lstmunit:out1"
   ],
   [
    "AddV2:out0",
    "lstmunit:out2"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstmunit": {
    "weights": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_3:out0\"}",
      "def r_lstmunit_rule_get_param_weights(self, node, tensor, tensor_name): # 'C_2:out0'",
      "    return self.shape_pick(tensor[tensor_name])[1] / 4",
      ""
     ]
    ],
    "num_proj": [
     "ORIGIN",
     "VALUE",
     null
    ],
    "forget_bias": [
     "FLOAT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_2:out0\"}",
      "def r_lstmunit_rule_get_param_forget_bias(self, node, tensor, tensor_name): #",
      "    forget_bias = self.tensor_to_numpy(tensor[tensor_name])",
      "    return forget_bias",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "lstmunit": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_3:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_1:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_1:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_1:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_1:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "stack_concat",
  "src_ops_alias": [
   "ConcatV2",
   "NextIteration",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "ConcatV2:in2"
   ],
   [
    "ConcatV2:out0",
    "NextIteration:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ConcatV2:in1"
   ],
   [
    "I_1:out0",
    "ConcatV2:in0"
   ]
  ],
  "src_out_tensor": [
   "ConcatV2:out0",
   "NextIteration:out0"
  ],
  "acu_lys_alias": [
   "stack_concat",
   "input"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "stack_concat:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ConcatV2:out0",
    "stack_concat:out0"
   ],
   [
    "NextIteration:out0",
    "stack_concat:out0"
   ]
  ],
  "acu_inter_flow": [
   [
    "input:out0",
    "stack_concat:in1"
   ]
  ],
  "param_map": {
   "stack_concat": {
    "axis": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"axis_tensor_name\": \"C:out0\"}",
      "def r_stack_concat_get_axis(self, node, tensor, axis_tensor_name):",
      "    axis = self.tensor_to_numpy(tensor[axis_tensor_name])",
      "    return axis",
      ""
     ]
    ],
    "shape": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"concat_out\": \"ConcatV2:out0\"}",
      "def r_stack_concat_get_shape(self, node, tensor, concat_out):",
      "    shape = self.shape_pick(tensor[concat_out])",
      "    return list(shape)",
      ""
     ]
    ]
   },
   "input": {
    "shape": [
     "ORIGIN",
     "VALUE",
     [
      1
     ]
    ]
   }
  },
  "blob_map": {},
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "multi_attention",
  "src_ops_alias": [
   "Reshape",
   "MatMul",
   "C",
   "Reshape_1",
   "C_1",
   "Transpose",
   "C_2",
   "Reshape_2",
   "C_3",
   "Transpose_1",
   "C_4",
   "BatchMatMul",
   "C_5",
   "Reshape_3",
   "Transpose_2",
   "Softmax",
   "C_6",
   "Reshape_4",
   "C_7",
   "Reshape_5",
   "ConcatV2",
   "C_8",
   "Add",
   "C_9",
   "Reshape_6",
   "C_10",
   "BatchMatMul_1",
   "MatMul_1",
   "C_11",
   "Mul",
   "Transpose_3",
   "Reshape_7",
   "C_12",
   "Transpose_4",
   "C_13",
   "Reshape_8",
   "C_14",
   "Transpose_5",
   "C_15",
   "Reshape_9",
   "C_16",
   "ConcatV2_1",
   "C_17",
   "C_18",
   "Reshape_10",
   "C_19",
   "Reshape_11",
   "C_20",
   "MatMul_2",
   "C_21",
   "MatMul_3",
   "C_22",
   "Reshape_12",
   "C_23",
   "Reshape_13",
   "C_24",
   "Transpose_6",
   "C_25",
   "Transpose_7",
   "C_26",
   "C_27",
   "C_28"
  ],
  "src_inter_flow": [
   [
    "MatMul:out0",
    "Reshape:in0"
   ],
   [
    "C:out0",
    "Reshape:in1"
   ],
   [
    "Reshape_1:out0",
    "MatMul:in0"
   ],
   [
    "C_1:out0",
    "MatMul:in1"
   ],
   [
    "Transpose:out0",
    "Reshape_1:in0"
   ],
   [
    "C_2:out0",
    "Reshape_1:in1"
   ],
   [
    "Reshape_2:out0",
    "Transpose:in0"
   ],
   [
    "C_3:out0",
    "Transpose:in1"
   ],
   [
    "Transpose_1:out0",
    "Reshape_2:in0"
   ],
   [
    "C_4:out0",
    "Reshape_2:in1"
   ],
   [
    "BatchMatMul:out0",
    "Transpose_1:in0"
   ],
   [
    "C_5:out0",
    "Transpose_1:in1"
   ],
   [
    "Reshape_3:out0",
    "BatchMatMul:in0"
   ],
   [
    "Transpose_2:out0",
    "BatchMatMul:in1"
   ],
   [
    "Softmax:out0",
    "Reshape_3:in0"
   ],
   [
    "C_6:out0",
    "Reshape_3:in1"
   ],
   [
    "Reshape_4:out0",
    "Transpose_2:in0"
   ],
   [
    "C_7:out0",
    "Transpose_2:in1"
   ],
   [
    "Reshape_5:out0",
    "Softmax:in0"
   ],
   [
    "ConcatV2:out0",
    "Reshape_4:in0"
   ],
   [
    "C_8:out0",
    "Reshape_4:in1"
   ],
   [
    "Add:out0",
    "Reshape_5:in0"
   ],
   [
    "C_9:out0",
    "Reshape_5:in1"
   ],
   [
    "Reshape_6:out0",
    "ConcatV2:in1"
   ],
   [
    "C_10:out0",
    "ConcatV2:in2"
   ],
   [
    "BatchMatMul_1:out0",
    "Add:in0"
   ],
   [
    "MatMul_1:out0",
    "Reshape_6:in0"
   ],
   [
    "C_11:out0",
    "Reshape_6:in1"
   ],
   [
    "Mul:out0",
    "BatchMatMul_1:in0"
   ],
   [
    "Transpose_3:out0",
    "BatchMatMul_1:in1"
   ],
   [
    "Reshape_7:out0",
    "MatMul_1:in0"
   ],
   [
    "C_12:out0",
    "MatMul_1:in1"
   ],
   [
    "Transpose_4:out0",
    "Mul:in0"
   ],
   [
    "C_13:out0",
    "Mul:in1"
   ],
   [
    "Reshape_8:out0",
    "Transpose_3:in0"
   ],
   [
    "C_14:out0",
    "Transpose_3:in1"
   ],
   [
    "Transpose_5:out0",
    "Reshape_7:in0"
   ],
   [
    "C_15:out0",
    "Reshape_7:in1"
   ],
   [
    "Reshape_9:out0",
    "Transpose_4:in0"
   ],
   [
    "C_16:out0",
    "Transpose_4:in1"
   ],
   [
    "ConcatV2_1:out0",
    "Reshape_8:in0"
   ],
   [
    "C_17:out0",
    "Reshape_8:in1"
   ],
   [
    "C_18:out0",
    "Transpose_5:in1"
   ],
   [
    "Reshape_10:out0",
    "Reshape_9:in0"
   ],
   [
    "C_19:out0",
    "Reshape_9:in1"
   ],
   [
    "Reshape_11:out0",
    "ConcatV2_1:in1"
   ],
   [
    "C_20:out0",
    "ConcatV2_1:in2"
   ],
   [
    "MatMul_2:out0",
    "Reshape_10:in0"
   ],
   [
    "C_21:out0",
    "Reshape_10:in1"
   ],
   [
    "MatMul_3:out0",
    "Reshape_11:in0"
   ],
   [
    "C_22:out0",
    "Reshape_11:in1"
   ],
   [
    "Reshape_12:out0",
    "MatMul_2:in0"
   ],
   [
    "C_23:out0",
    "MatMul_2:in1"
   ],
   [
    "Reshape_13:out0",
    "MatMul_3:in0"
   ],
   [
    "C_24:out0",
    "MatMul_3:in1"
   ],
   [
    "Transpose_6:out0",
    "Reshape_12:in0"
   ],
   [
    "C_25:out0",
    "Reshape_12:in1"
   ],
   [
    "Transpose_7:out0",
    "Reshape_13:in0"
   ],
   [
    "C_26:out0",
    "Reshape_13:in1"
   ],
   [
    "C_27:out0",
    "Transpose_6:in1"
   ],
   [
    "C_28:out0",
    "Transpose_7:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Transpose_5:in0"
   ],
   [
    "I:out0",
    "Transpose_7:in0"
   ],
   [
    "I:out0",
    "Transpose_6:in0"
   ],
   [
    "I_1:out0",
    "Add:in1"
   ],
   [
    "I_2:out0",
    "ConcatV2_1:in0"
   ],
   [
    "I_3:out0",
    "ConcatV2:in0"
   ]
  ],
  "src_out_tensor": [
   "Reshape:out0"
  ],
  "acu_lys_alias": [
   "variable",
   "input",
   "input_1",
   "reshape_11",
   "fullconnect_3",
   "reshape_10",
   "stack_concat_1",
   "reshape_9",
   "permute_3",
   "reshape_8",
   "fullconnect_2",
   "reshape_7",
   "stack_concat",
   "reshape_6",
   "permute_2",
   "reshape_5",
   "fullconnect_1",
   "reshape_4",
   "permute_1",
   "multiply",
   "matmul_1",
   "add",
   "reshape_3",
   "softmax",
   "reshape_2",
   "matmul",
   "permute",
   "reshape_1",
   "fullconnect",
   "reshape"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reshape_11:in0"
   ],
   [
    "I:out0",
    "reshape_8:in0"
   ],
   [
    "I:out0",
    "reshape_5:in0"
   ],
   [
    "I_1:out0",
    "add:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Reshape:out0",
    "reshape:out0"
   ]
  ],
  "acu_inter_flow": [
   [
    "variable:out0",
    "multiply:in1"
   ],
   [
    "reshape_11:out0",
    "fullconnect_3:in0"
   ],
   [
    "fullconnect_3:out0",
    "reshape_10:in0"
   ],
   [
    "reshape_10:out0",
    "stack_concat_1:in0"
   ],
   [
    "input_1:out0",
    "stack_concat_1:in1"
   ],
   [
    "stack_concat_1:out0",
    "reshape_9:in0"
   ],
   [
    "reshape_9:out0",
    "permute_3:in0"
   ],
   [
    "permute_3:out0",
    "matmul_1:in1"
   ],
   [
    "reshape_8:out0",
    "fullconnect_2:in0"
   ],
   [
    "fullconnect_2:out0",
    "reshape_7:in0"
   ],
   [
    "reshape_7:out0",
    "stack_concat:in0"
   ],
   [
    "input:out0",
    "stack_concat:in1"
   ],
   [
    "stack_concat:out0",
    "reshape_6:in0"
   ],
   [
    "reshape_6:out0",
    "permute_2:in0"
   ],
   [
    "permute_2:out0",
    "matmul:in1"
   ],
   [
    "reshape_5:out0",
    "fullconnect_1:in0"
   ],
   [
    "fullconnect_1:out0",
    "reshape_4:in0"
   ],
   [
    "reshape_4:out0",
    "permute_1:in0"
   ],
   [
    "permute_1:out0",
    "multiply:in0"
   ],
   [
    "multiply:out0",
    "matmul_1:in0"
   ],
   [
    "matmul_1:out0",
    "add:in0"
   ],
   [
    "add:out0",
    "reshape_3:in0"
   ],
   [
    "reshape_3:out0",
    "softmax:in0"
   ],
   [
    "softmax:out0",
    "reshape_2:in0"
   ],
   [
    "reshape_2:out0",
    "matmul:in0"
   ],
   [
    "matmul:out0",
    "permute:in0"
   ],
   [
    "permute:out0",
    "reshape_1:in0"
   ],
   [
    "reshape_1:out0",
    "fullconnect:in0"
   ],
   [
    "fullconnect:out0",
    "reshape:in0"
   ]
  ],
  "param_map": {
   "variable": {
    "shape": [
     "ORIGIN",
     "CODE",
     "self.shape_pick(tensor['C_13:out0'])"
    ]
   },
   "fullconnect": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_1:out0'])[1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ]
   },
   "fullconnect_1": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_23:out0'])[1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ]
   },
   "fullconnect_2": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_12:out0'])[1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ]
   },
   "fullconnect_3": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_24:out0'])[1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     false
    ]
   },
   "input": {
    "shape": [
     "ORIGIN",
     "VALUE",
     [
      1
     ]
    ]
   },
   "input_1": {
    "shape": [
     "ORIGIN",
     "VALUE",
     [
      1
     ]
    ]
   },
   "stack_concat": {},
   "stack_concat_1": {},
   "permute": {
    "perm": [
     "STRING",
     "CODE",
     "' '.join([str(perm) for perm in self.tensor_to_numpy(tensor['C_5:out0'])])"
    ]
   },
   "permute_1": {
    "perm": [
     "STRING",
     "CODE",
     "' '.join([str(perm) for perm in self.tensor_to_numpy(tensor['C_16:out0'])])"
    ]
   },
   "permute_2": {
    "perm": [
     "STRING",
     "CODE",
     "' '.join([str(perm) for perm in self.tensor_to_numpy(tensor['C_7:out0'])])"
    ]
   },
   "permute_3": {
    "perm": [
     "STRING",
     "CODE",
     "' '.join([str(perm) for perm in self.tensor_to_numpy(tensor['C_14:out0'])])"
    ]
   },
   "reshape": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C:out0'])"
    ]
   },
   "reshape_1": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C_2:out0'])"
    ]
   },
   "reshape_2": {
    "shape": [
     "INTS",
     "VALUE",
     [
      1,
      4,
      1,
      32
     ]
    ]
   },
   "reshape_3": {
    "shape": [
     "INTS",
     "VALUE",
     [
      -1,
      32
     ]
    ]
   },
   "reshape_4": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C_19:out0'])"
    ]
   },
   "reshape_5": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C_25:out0'])"
    ]
   },
   "reshape_6": {
    "shape": [
     "INTS",
     "VALUE",
     [
      1,
      32,
      4,
      64
     ]
    ]
   },
   "reshape_7": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C_11:out0'])"
    ]
   },
   "reshape_8": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C_15:out0'])"
    ]
   },
   "reshape_9": {
    "shape": [
     "INTS",
     "VALUE",
     [
      1,
      32,
      4,
      64
     ]
    ]
   },
   "reshape_10": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C_22:out0'])"
    ]
   },
   "reshape_11": {
    "shape": [
     "INTS",
     "CODE",
     "self.reshape_shape(tensor['C_26:out0'])"
    ]
   },
   "matmul": {
    "transpose_a": [
     "INT",
     "CODE",
     "self.attr_pick(node['BatchMatMul'], 'adj_x', False)"
    ],
    "transpose_b": [
     "INT",
     "CODE",
     "self.attr_pick(node['BatchMatMul'], 'adj_y', False)"
    ]
   },
   "matmul_1": {
    "transpose_a": [
     "INT",
     "CODE",
     "self.attr_pick(node['BatchMatMul_1'], 'adj_x', False)"
    ],
    "transpose_b": [
     "INT",
     "CODE",
     "self.attr_pick(node['BatchMatMul_1'], 'adj_y', False)"
    ]
   },
   "multiply": {},
   "softmax": {},
   "add": {}
  },
  "blob_map": {
   "variable": {
    "data": [
     "CODE",
     "np.array([self.tensor_to_numpy(tensor['C_13:out0'])], dtype=np.float32) if self.tensor_to_numpy(tensor['C_13:out0']).shape == () else self.tensor_to_numpy(tensor['C_13:out0'], dtype='float32')"
    ]
   },
   "variable_1": {
    "data": [
     "CODE",
     "np.zeros((1,32,256), dtype=np.float32)"
    ]
   },
   "variable_2": {
    "data": [
     "CODE",
     "np.zeros((1,32,256), dtype=np.float32)"
    ]
   },
   "fullconnect": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   },
   "fullconnect_1": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_23:out0'])"
    ]
   },
   "fullconnect_2": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_12:out0'])"
    ]
   },
   "fullconnect_3": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_24:out0'])"
    ]
   },
   "reshape_8": {},
   "stack_concat_1": {},
   "reshape_7": {},
   "reshape": {},
   "permute_2": {},
   "permute": {},
   "permute_3": {},
   "reshape_1": {},
   "reshape_2": {},
   "reshape_3": {},
   "permute_1": {},
   "multiply": {},
   "matmul_1": {},
   "reshape_9": {},
   "reshape_11": {},
   "matmul": {},
   "reshape_10": {},
   "reshape_4": {},
   "reshape_5": {},
   "reshape_6": {},
   "stack_concat": {},
   "softmax": {},
   "add": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "ffn_convs",
  "src_ops_alias": [
   "Squeeze",
   "BiasAdd",
   "Reshape",
   "C",
   "MatMul",
   "C_1",
   "Reshape_1",
   "C_2",
   "Transpose",
   "C_3",
   "Relu",
   "C_4",
   "BiasAdd_1",
   "Reshape_2",
   "C_5",
   "MatMul_1",
   "C_6",
   "Reshape_3",
   "C_7",
   "Transpose_1",
   "C_8",
   "ExpandDims",
   "C_9",
   "C_10"
  ],
  "src_inter_flow": [
   [
    "BiasAdd:out0",
    "Squeeze:in0"
   ],
   [
    "Reshape:out0",
    "BiasAdd:in0"
   ],
   [
    "C:out0",
    "BiasAdd:in1"
   ],
   [
    "MatMul:out0",
    "Reshape:in0"
   ],
   [
    "C_1:out0",
    "Reshape:in1"
   ],
   [
    "Reshape_1:out0",
    "MatMul:in0"
   ],
   [
    "C_2:out0",
    "MatMul:in1"
   ],
   [
    "Transpose:out0",
    "Reshape_1:in0"
   ],
   [
    "C_3:out0",
    "Reshape_1:in1"
   ],
   [
    "Relu:out0",
    "Transpose:in0"
   ],
   [
    "C_4:out0",
    "Transpose:in1"
   ],
   [
    "BiasAdd_1:out0",
    "Relu:in0"
   ],
   [
    "Reshape_2:out0",
    "BiasAdd_1:in0"
   ],
   [
    "C_5:out0",
    "BiasAdd_1:in1"
   ],
   [
    "MatMul_1:out0",
    "Reshape_2:in0"
   ],
   [
    "C_6:out0",
    "Reshape_2:in1"
   ],
   [
    "Reshape_3:out0",
    "MatMul_1:in0"
   ],
   [
    "C_7:out0",
    "MatMul_1:in1"
   ],
   [
    "Transpose_1:out0",
    "Reshape_3:in0"
   ],
   [
    "C_8:out0",
    "Reshape_3:in1"
   ],
   [
    "ExpandDims:out0",
    "Transpose_1:in0"
   ],
   [
    "C_9:out0",
    "Transpose_1:in1"
   ],
   [
    "C_10:out0",
    "ExpandDims:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ExpandDims:in0"
   ]
  ],
  "src_out_tensor": [
   "Squeeze:out0"
  ],
  "acu_lys_alias": [
   "fullconnect",
   "relu",
   "fullconnect_1"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "fullconnect:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Squeeze:out0",
    "fullconnect_1:out0"
   ]
  ],
  "acu_inter_flow": [
   [
    "fullconnect:out0",
    "relu:in0"
   ],
   [
    "relu:out0",
    "fullconnect_1:in0"
   ]
  ],
  "param_map": {
   "relu": {},
   "fullconnect": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_7:out0'])[1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ]
   },
   "fullconnect_1": {
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_2:out0'])[1]"
    ],
    "bias": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "relu": {},
   "fullconnect": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_7:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_5:out0'])"
    ]
   },
   "fullconnect_1": {
    "weight": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "bias": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": "len(self.shape_pick(tensor['C:out0'])) < 2 and len(self.shape_pick(tensor['C_5:out0'])) < 2"
 },
 {
  "ruler_name": "lstm_tf_1.10",
  "src_ops_alias": [
   "TensorArrayGatherV3",
   "TensorArrayV3",
   "Range",
   "Exit",
   "C",
   "C_1",
   "TensorArraySizeV3",
   "C_2",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "TensorArrayWriteV3",
   "Less",
   "Less_1",
   "Enter_1",
   "Identity",
   "Mul",
   "Identity_1",
   "Merge_1",
   "Enter_2",
   "Merge_2",
   "Enter_3",
   "Switch_1",
   "Sigmoid",
   "Tanh",
   "Enter_4",
   "NextIteration_1",
   "Enter_5",
   "NextIteration_2",
   "C_3",
   "Split",
   "Add",
   "C_4",
   "Add_1",
   "C_5",
   "Add_2",
   "C_6",
   "BiasAdd",
   "Mul_1",
   "Mul_2",
   "Identity_2",
   "C_7",
   "C_8",
   "MatMul",
   "Enter_6",
   "Sigmoid_1",
   "Identity_3",
   "Sigmoid_2",
   "Tanh_1",
   "Switch_2",
   "ConcatV2",
   "Enter_7",
   "C_9",
   "Add_3",
   "Switch_3",
   "TensorArrayReadV3",
   "Identity_4",
   "C_10",
   "C_11",
   "C_12",
   "Merge_3",
   "Enter_8",
   "Enter_9",
   "Switch_4",
   "Enter_10",
   "NextIteration_3",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "Merge_4",
   "C_13",
   "C_14",
   "Enter_11",
   "NextIteration_4",
   "C_15"
  ],
  "src_inter_flow": [
   [
    "Identity_2:out4096",
    "C_12:in0"
   ],
   [
    "Add:out0",
    "Tanh:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "Enter_8:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Identity_2:out0",
    "Add_1:in0"
   ],
   [
    "Enter_11:out0",
    "Merge_4:in0"
   ],
   [
    "Sigmoid:out0",
    "Mul:in0"
   ],
   [
    "Identity_2:out4096",
    "C_6:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "Enter_10:out0",
    "Merge_3:in0"
   ],
   [
    "C_15:out0",
    "Enter_11:in0"
   ],
   [
    "C_9:out0",
    "Enter_6:in0"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Switch_4:out1",
    "Identity_4:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_2:in0"
   ],
   [
    "C_11:out0",
    "Enter_7:in0"
   ],
   [
    "Exit:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in0"
   ],
   [
    "Identity_4:out0",
    "ConcatV2:in1"
   ],
   [
    "Add_1:out0",
    "NextIteration_1:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_4:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "Identity_3:out0",
    "Mul_1:in1"
   ],
   [
    "Merge_4:out0",
    "Switch_4:in0"
   ],
   [
    "Enter_2:out0",
    "Less:in1"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_9:in0"
   ],
   [
    "Exit:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "C_2:out0",
    "Range:in2"
   ],
   [
    "Mul:out0",
    "NextIteration_4:in0"
   ],
   [
    "C_1:out0",
    "Range:in0"
   ],
   [
    "Enter_3:out0",
    "Less_1:in1"
   ],
   [
    "Merge_1:out0",
    "Switch_2:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Enter_9:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_1:in0"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Switch_3:out1",
    "Identity_3:in0"
   ],
   [
    "Enter_4:out0",
    "Merge_1:in0"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "NextIteration_4:out0",
    "Merge_4:in1"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "Identity_2:out4096",
    "C_8:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "Enter_7:out0",
    "MatMul:in1"
   ],
   [
    "Add_3:out0",
    "Sigmoid_1:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_3:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "C_6:out0",
    "Split:in0"
   ],
   [
    "Add:out0",
    "NextIteration_3:in0"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "C_3:out0",
    "Enter_3:in0"
   ],
   [
    "Identity:out0",
    "Add_2:in0"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "Enter_6:out0",
    "BiasAdd:in1"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "C_13:out0",
    "Enter_10:in0"
   ],
   [
    "C_10:out0",
    "ConcatV2:in2"
   ],
   [
    "C_4:out0",
    "Enter_4:in0"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Identity_2:out4096",
    "C_7:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "C_7:out0",
    "Add_1:in1"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "Split:out2",
    "Add_3:in0"
   ],
   [
    "C_14:out0",
    "TensorArrayScatterV3:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "C_12:out0",
    "Add_3:in1"
   ],
   [
    "Identity_2:out4096",
    "C_10:in0"
   ],
   [
    "C_8:out0",
    "Add_2:in1"
   ],
   [
    "C:out0",
    "Enter_2:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in0"
   ],
   [
    "Enter_1:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "Mul:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Enter_5:out0",
    "Merge_2:in0"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_8:in0"
   ],
   [
    "C_5:out0",
    "Enter_5:in0"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_1:in0"
   ],
   [
    "Merge_2:out0",
    "Less_1:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayGatherV3:out0"
  ],
  "acu_lys_alias": [
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstm:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayGatherV3:out0",
    "lstm:out0"
   ]
  ],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_12:out0'])"
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter_7:out0'])[1] / 4"
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_9:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_9:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_9:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_9:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "time_major_lstm_false_seq",
  "src_ops_alias": [
   "Exit",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "Less",
   "C",
   "Mul",
   "Merge_1",
   "Enter_1",
   "Tanh",
   "Sigmoid",
   "Enter_2",
   "NextIteration_1",
   "C_1",
   "Add",
   "Split",
   "C_2",
   "Add_1",
   "Mul_1",
   "Mul_2",
   "C_3",
   "BiasAdd",
   "Identity",
   "C_4",
   "Identity_1",
   "Sigmoid_1",
   "Sigmoid_2",
   "Tanh_1",
   "MatMul",
   "Enter_3",
   "Switch_1",
   "Switch_2",
   "Add_2",
   "ConcatV2",
   "Enter_4",
   "C_5",
   "Merge_2",
   "C_6",
   "TensorArrayReadV3",
   "Identity_2",
   "C_7",
   "C_8",
   "Enter_5",
   "NextIteration_2",
   "Enter_6",
   "Enter_7",
   "C_9",
   "TensorArrayV3",
   "TensorArrayScatterV3",
   "C_10"
  ],
  "src_inter_flow": [
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "Less:out0",
    "LoopCond:in0"
   ],
   [
    "C:out0",
    "Enter:in0"
   ],
   [
    "Mul:out0",
    "NextIteration:in0"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Enter_1:out0",
    "Less:in1"
   ],
   [
    "Tanh:out0",
    "Mul:in0"
   ],
   [
    "Sigmoid:out0",
    "Mul:in1"
   ],
   [
    "Enter_2:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "C_1:out0",
    "Enter_1:in0"
   ],
   [
    "Add:out0",
    "Tanh:in0"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "C_2:out0",
    "Enter_2:in0"
   ],
   [
    "Add_1:out0",
    "NextIteration_1:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "C_3:out0",
    "Split:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Identity:out0",
    "Add_1:in0"
   ],
   [
    "C_4:out0",
    "Add_1:in1"
   ],
   [
    "Identity_1:out0",
    "Mul_1:in0"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in1"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "Identity:out4096",
    "C_3:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Enter_3:out0",
    "BiasAdd:in1"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "Identity:out4096",
    "C_4:in0"
   ],
   [
    "Switch_2:out1",
    "Identity_1:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "Add_2:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Merge_1:out0",
    "Switch_1:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Enter_4:out0",
    "MatMul:in1"
   ],
   [
    "C_5:out0",
    "Enter_3:in0"
   ],
   [
    "Split:out2",
    "Add_2:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_2:in0"
   ],
   [
    "C_6:out0",
    "Add_2:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Identity_2:out0",
    "ConcatV2:in1"
   ],
   [
    "C_7:out0",
    "ConcatV2:in2"
   ],
   [
    "C_8:out0",
    "Enter_4:in0"
   ],
   [
    "Switch:out1",
    "Identity_2:in0"
   ],
   [
    "Identity:out4096",
    "C_6:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "Enter_5:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "Identity:out4096",
    "C_7:in0"
   ],
   [
    "Enter_6:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_7:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "Add:out0",
    "NextIteration_2:in0"
   ],
   [
    "C_9:out0",
    "Enter_5:in0"
   ],
   [
    "C_1:out0",
    "TensorArrayV3:in0"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_6:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_7:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "TensorArrayV3:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "C_10:out0",
    "TensorArrayScatterV3:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "Exit:out0"
  ],
  "acu_lys_alias": [
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstm:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Exit:out0",
    "lstm:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_6:out0'])"
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter_4:out0'])[1] / 4"
    ],
    "return_sequences": [
     "BOOL",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensors_name\": [\"I:out0\", \"Exit:out0\"]}",
      "def r_lstm_get_seq(self, node, tensor, tensors_name):",
      "    in_shape = self.shape_pick(tensor[tensors_name[0]])",
      "    out_shape = self.shape_pick(tensor[tensors_name[1]])",
      "    if len(in_shape) == len(out_shape):",
      "        return True",
      "    else:",
      "        return False",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_8:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_5:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_5:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_5:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_5:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "lstm_time_major_no_seq_1_13",
  "src_ops_alias": [
   "Exit",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "C",
   "Mul",
   "Less",
   "Less_1",
   "Sigmoid",
   "Tanh",
   "Merge_1",
   "Enter_1",
   "Merge_2",
   "Enter_2",
   "Split",
   "Add",
   "Enter_3",
   "NextIteration_1",
   "C_1",
   "Enter_4",
   "NextIteration_2",
   "C_2",
   "C_3",
   "BiasAdd",
   "Mul_1",
   "Mul_2",
   "C_4",
   "Add_1",
   "C_5",
   "Add_2",
   "Identity",
   "MatMul",
   "Enter_5",
   "Sigmoid_1",
   "Identity_1",
   "Sigmoid_2",
   "Tanh_1",
   "C_6",
   "Identity_2",
   "C_7",
   "Switch_1",
   "ConcatV2",
   "Enter_6",
   "C_8",
   "Add_3",
   "Switch_2",
   "Switch_3",
   "TensorArrayReadV3",
   "Identity_3",
   "C_9",
   "C_10",
   "C_11",
   "Merge_3",
   "Enter_7",
   "Enter_8",
   "Enter_9",
   "NextIteration_3",
   "TensorArrayV3",
   "TensorArrayScatterV3",
   "C_12",
   "C_13"
  ],
  "src_inter_flow": [
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "C:out0",
    "Enter:in0"
   ],
   [
    "Mul:out0",
    "NextIteration:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "Sigmoid:out0",
    "Mul:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Enter_1:out0",
    "Less:in1"
   ],
   [
    "Merge_2:out0",
    "Less_1:in0"
   ],
   [
    "Enter_2:out0",
    "Less_1:in1"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "Add:out0",
    "Tanh:in0"
   ],
   [
    "Enter_3:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "C_1:out0",
    "Enter_1:in0"
   ],
   [
    "Enter_4:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "C_2:out0",
    "Enter_2:in0"
   ],
   [
    "C_3:out0",
    "Split:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Mul_1:out0",
    "Add:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "C_4:out0",
    "Enter_3:in0"
   ],
   [
    "Add_1:out0",
    "NextIteration_1:in0"
   ],
   [
    "C_5:out0",
    "Enter_4:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_2:in0"
   ],
   [
    "Identity:out4096",
    "C_3:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Enter_5:out0",
    "BiasAdd:in1"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in0"
   ],
   [
    "Identity_1:out0",
    "Mul_1:in1"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "Identity:out0",
    "Add_1:in0"
   ],
   [
    "C_6:out0",
    "Add_1:in1"
   ],
   [
    "Identity_2:out0",
    "Add_2:in0"
   ],
   [
    "C_7:out0",
    "Add_2:in1"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Enter_6:out0",
    "MatMul:in1"
   ],
   [
    "C_8:out0",
    "Enter_5:in0"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "Add_3:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "Switch_2:out1",
    "Identity_1:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Identity:out4096",
    "C_6:in0"
   ],
   [
    "Merge_1:out0",
    "Switch_1:in0"
   ],
   [
    "Identity:out4096",
    "C_7:in0"
   ],
   [
    "Switch_3:out1",
    "Identity_2:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Split:out2",
    "Add_3:in0"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Identity_3:out0",
    "ConcatV2:in1"
   ],
   [
    "C_9:out0",
    "ConcatV2:in2"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "C_10:out0",
    "Enter_6:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_3:in0"
   ],
   [
    "Switch:out1",
    "Identity_3:in0"
   ],
   [
    "C_11:out0",
    "Add_3:in1"
   ],
   [
    "Merge_3:out0",
    "Switch_2:in0"
   ],
   [
    "Identity_2:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "Identity:out4096",
    "C_9:in0"
   ],
   [
    "Enter_7:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_8:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "Identity:out4096",
    "C_11:in0"
   ],
   [
    "Add:out0",
    "NextIteration_3:in0"
   ],
   [
    "Enter_9:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_7:in0"
   ],
   [
    "C_1:out0",
    "TensorArrayV3:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_8:in0"
   ],
   [
    "C_12:out0",
    "Enter_9:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "TensorArrayV3:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "C_13:out0",
    "TensorArrayScatterV3:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "Exit:out0"
  ],
  "acu_lys_alias": [
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstm:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Exit:out0",
    "lstm:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_11:out0'])"
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter_6:out0'])[1] / 4"
    ],
    "return_sequences": [
     "BOOL",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensors_name\": [\"I:out0\", \"Exit:out0\"]}",
      "def r_lstm_get_seq(self, node, tensor, tensors_name):",
      "    in_shape = self.shape_pick(tensor[tensors_name[0]])",
      "    out_shape = self.shape_pick(tensor[tensors_name[1]])",
      "    if len(in_shape) == len(out_shape):",
      "        return True",
      "    else:",
      "        return False",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_10:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_10:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_10:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_10:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_10:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_10:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_10:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_10:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_8:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_8:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_8:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_8:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "gru_time_major_encoder",
  "src_ops_alias": [
   "TensorArrayGatherV3",
   "TensorArrayV3",
   "Range",
   "Exit",
   "C",
   "C_1",
   "TensorArraySizeV3",
   "C_2",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "TensorArrayWriteV3",
   "Less",
   "Less_1",
   "Enter_1",
   "Identity",
   "Select",
   "Identity_1",
   "Merge_1",
   "Enter_2",
   "Merge_2",
   "Enter_3",
   "Switch_1",
   "GreaterEqual",
   "Enter_4",
   "Add",
   "Enter_5",
   "NextIteration_1",
   "Enter_6",
   "NextIteration_2",
   "Minimum",
   "Enter_7",
   "C_3",
   "Mul",
   "Mul_1",
   "C_4",
   "Add_1",
   "C_5",
   "Add_2",
   "Maximum",
   "Identity_2",
   "Split",
   "Identity_3",
   "Sub",
   "Tanh",
   "Identity_4",
   "C_6",
   "C_7",
   "C_8",
   "Max",
   "C_9",
   "Sigmoid",
   "Switch_2",
   "C_10",
   "BiasAdd",
   "Switch_3",
   "C_11",
   "BiasAdd_1",
   "Merge_3",
   "MatMul",
   "Enter_8",
   "MatMul_1",
   "Enter_9",
   "Enter_10",
   "NextIteration_3",
   "ConcatV2",
   "Enter_11",
   "C_12",
   "ConcatV2_1",
   "Enter_12",
   "C_13",
   "C_14",
   "Select_1",
   "TensorArrayReadV3",
   "Mul_2",
   "C_15",
   "C_16",
   "C_17",
   "C_18",
   "Enter_13",
   "Enter_14",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "C_19"
  ],
  "src_inter_flow": [
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "Exit:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "C_1:out0",
    "Range:in0"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "C_2:out0",
    "Range:in2"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "Exit:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "Enter_1:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Select:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_1:in0"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Enter_2:out0",
    "Less:in1"
   ],
   [
    "Merge_2:out0",
    "Less_1:in0"
   ],
   [
    "Enter_3:out0",
    "Less_1:in1"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "GreaterEqual:out0",
    "Select:in0"
   ],
   [
    "Enter_4:out0",
    "Select:in1"
   ],
   [
    "Add:out0",
    "Select:in2"
   ],
   [
    "C:out0",
    "Enter_2:in0"
   ],
   [
    "Enter_5:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Enter_6:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "Minimum:out0",
    "Enter_3:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_1:in0"
   ],
   [
    "Identity:out0",
    "GreaterEqual:in0"
   ],
   [
    "Enter_7:out0",
    "GreaterEqual:in1"
   ],
   [
    "C_3:out0",
    "Enter_4:in0"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in1"
   ],
   [
    "C_4:out0",
    "Enter_5:in0"
   ],
   [
    "C:out0",
    "Minimum:in0"
   ],
   [
    "Add_1:out0",
    "NextIteration_1:in0"
   ],
   [
    "C_5:out0",
    "Enter_6:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_2:in0"
   ],
   [
    "Maximum:out0",
    "Minimum:in1"
   ],
   [
    "Identity_2:out0",
    "Enter_7:in0"
   ],
   [
    "Split:out1",
    "Mul:in0"
   ],
   [
    "Identity_3:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Tanh:out0",
    "Mul_1:in1"
   ],
   [
    "Identity:out0",
    "Add_2:in0"
   ],
   [
    "Identity_4:out0",
    "Add_1:in0"
   ],
   [
    "C_6:out0",
    "Add_1:in1"
   ],
   [
    "C_7:out0",
    "Add_2:in1"
   ],
   [
    "C_8:out0",
    "Maximum:in0"
   ],
   [
    "Max:out0",
    "Maximum:in1"
   ],
   [
    "C_9:out0",
    "Split:in0"
   ],
   [
    "Sigmoid:out0",
    "Split:in1"
   ],
   [
    "Split:out1",
    "Sub:in1"
   ],
   [
    "Switch_2:out1",
    "Identity_3:in0"
   ],
   [
    "C_10:out0",
    "Sub:in0"
   ],
   [
    "BiasAdd:out0",
    "Tanh:in0"
   ],
   [
    "Switch_3:out1",
    "Identity_4:in0"
   ],
   [
    "Identity_4:out4096",
    "C_6:in0"
   ],
   [
    "Identity_4:out4096",
    "C_7:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Identity_2:out0",
    "Max:in0"
   ],
   [
    "Identity_4:out4096",
    "C_9:in0"
   ],
   [
    "C_11:out0",
    "Max:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "BiasAdd_1:out0",
    "Sigmoid:in0"
   ],
   [
    "Merge_1:out0",
    "Switch_3:in0"
   ],
   [
    "Identity_4:out4096",
    "C_10:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_2:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Enter_8:out0",
    "BiasAdd:in1"
   ],
   [
    "MatMul_1:out0",
    "BiasAdd_1:in0"
   ],
   [
    "Enter_9:out0",
    "BiasAdd_1:in1"
   ],
   [
    "Enter_10:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Enter_11:out0",
    "MatMul:in1"
   ],
   [
    "C_12:out0",
    "Enter_8:in0"
   ],
   [
    "ConcatV2_1:out0",
    "MatMul_1:in0"
   ],
   [
    "Enter_12:out0",
    "MatMul_1:in1"
   ],
   [
    "C_13:out0",
    "Enter_9:in0"
   ],
   [
    "C_14:out0",
    "Enter_10:in0"
   ],
   [
    "Select_1:out0",
    "NextIteration_3:in0"
   ],
   [
    "Identity_3:out0",
    "ConcatV2_1:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Mul_2:out0",
    "ConcatV2:in1"
   ],
   [
    "C_15:out0",
    "ConcatV2:in2"
   ],
   [
    "C_16:out0",
    "Enter_11:in0"
   ],
   [
    "GreaterEqual:out0",
    "Select_1:in0"
   ],
   [
    "Add:out0",
    "Select_1:in2"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2_1:in0"
   ],
   [
    "C_17:out0",
    "ConcatV2_1:in2"
   ],
   [
    "Identity_3:out0",
    "Select_1:in1"
   ],
   [
    "C_18:out0",
    "Enter_12:in0"
   ],
   [
    "Split:out0",
    "Mul_2:in0"
   ],
   [
    "Identity_3:out0",
    "Mul_2:in1"
   ],
   [
    "Identity_4:out4096",
    "C_15:in0"
   ],
   [
    "Enter_13:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_14:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "Identity_4:out4096",
    "C_17:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_13:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_14:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "C_19:out0",
    "TensorArrayScatterV3:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ],
   [
    "I_1:out0",
    "Identity_2:in0"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayGatherV3:out0"
  ],
  "acu_lys_alias": [
   "gru"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gru:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayGatherV3:out0",
    "gru:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gru": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "num_units": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter_12:out0'])[1] / 2"
    ],
    "reset_after": [
     "BOOL",
     "VALUE",
     false
    ]
   }
  },
  "blob_map": {
   "gru": {
    "k_iz": [
     "CODE",
     "self.gru_kernel(tensor['C_18:out0'], tensor['C_16:out0'])[0][0]"
    ],
    "k_ir": [
     "CODE",
     "self.gru_kernel(tensor['C_18:out0'], tensor['C_16:out0'])[0][1]"
    ],
    "k_ih": [
     "CODE",
     "self.gru_kernel(tensor['C_18:out0'], tensor['C_16:out0'])[0][2]"
    ],
    "k_rz": [
     "CODE",
     "self.gru_kernel(tensor['C_18:out0'], tensor['C_16:out0'])[1][0]"
    ],
    "k_rr": [
     "CODE",
     "self.gru_kernel(tensor['C_18:out0'], tensor['C_16:out0'])[1][1]"
    ],
    "k_rh": [
     "CODE",
     "self.gru_kernel(tensor['C_18:out0'], tensor['C_16:out0'])[1][2]"
    ],
    "b_iz": [
     "CODE",
     "self.gru_bias(tensor['C_13:out0'], tensor['C_12:out0'])[0][0]"
    ],
    "b_ir": [
     "CODE",
     "self.gru_bias(tensor['C_13:out0'], tensor['C_12:out0'])[0][1]"
    ],
    "b_ih": [
     "CODE",
     "self.gru_bias(tensor['C_13:out0'], tensor['C_12:out0'])[0][2]"
    ],
    "b_rz": [
     "CODE",
     "self.gru_bias(tensor['C_13:out0'], tensor['C_12:out0'])[1][0]"
    ],
    "b_rr": [
     "CODE",
     "self.gru_bias(tensor['C_13:out0'], tensor['C_12:out0'])[1][1]"
    ],
    "b_rh": [
     "CODE",
     "self.gru_bias(tensor['C_13:out0'], tensor['C_12:out0'])[1][2]"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "gru_time_major_post",
  "src_ops_alias": [
   "TensorArrayGatherV3",
   "TensorArrayV3",
   "Range",
   "Exit",
   "C",
   "C_1",
   "TensorArraySizeV3",
   "C_2",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "TensorArrayWriteV3",
   "Less",
   "Less_1",
   "Enter_1",
   "Identity",
   "Add",
   "Identity_1",
   "Merge_1",
   "Enter_2",
   "Merge_2",
   "Enter_3",
   "Switch_1",
   "Mul",
   "Mul_1",
   "Enter_4",
   "NextIteration_1",
   "Enter_5",
   "NextIteration_2",
   "C_3",
   "Split",
   "Identity_2",
   "Sub",
   "Tanh",
   "C_4",
   "Add_1",
   "C_5",
   "Add_2",
   "C_6",
   "Sigmoid",
   "Switch_2",
   "C_7",
   "BiasAdd",
   "Identity_3",
   "C_8",
   "C_9",
   "BiasAdd_1",
   "Merge_3",
   "MatMul",
   "Enter_6",
   "Switch_3",
   "MatMul_1",
   "Enter_7",
   "Enter_8",
   "NextIteration_3",
   "ConcatV2",
   "Enter_9",
   "C_10",
   "ConcatV2_1",
   "Enter_10",
   "C_11",
   "C_12",
   "TensorArrayReadV3",
   "Mul_2",
   "C_13",
   "C_14",
   "C_15",
   "C_16",
   "Enter_11",
   "Enter_12",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "C_17"
  ],
  "src_inter_flow": [
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "Exit:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "C_1:out0",
    "Range:in0"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "C_2:out0",
    "Range:in2"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "Exit:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "Enter_1:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Add:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_1:in0"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Enter_2:out0",
    "Less:in1"
   ],
   [
    "Merge_2:out0",
    "Less_1:in0"
   ],
   [
    "Enter_3:out0",
    "Less_1:in1"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in1"
   ],
   [
    "C:out0",
    "Enter_2:in0"
   ],
   [
    "Enter_4:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Enter_5:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "C_3:out0",
    "Enter_3:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_1:in0"
   ],
   [
    "Split:out1",
    "Mul:in0"
   ],
   [
    "Identity_2:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Tanh:out0",
    "Mul_1:in1"
   ],
   [
    "C_4:out0",
    "Enter_4:in0"
   ],
   [
    "Add_1:out0",
    "NextIteration_1:in0"
   ],
   [
    "C_5:out0",
    "Enter_5:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_2:in0"
   ],
   [
    "C_6:out0",
    "Split:in0"
   ],
   [
    "Sigmoid:out0",
    "Split:in1"
   ],
   [
    "Split:out1",
    "Sub:in1"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "C_7:out0",
    "Sub:in0"
   ],
   [
    "BiasAdd:out0",
    "Tanh:in0"
   ],
   [
    "Identity:out0",
    "Add_2:in0"
   ],
   [
    "Identity_3:out0",
    "Add_1:in0"
   ],
   [
    "C_8:out0",
    "Add_1:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "C_9:out0",
    "Add_2:in1"
   ],
   [
    "Identity_3:out4096",
    "C_6:in0"
   ],
   [
    "BiasAdd_1:out0",
    "Sigmoid:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_2:in0"
   ],
   [
    "Identity_3:out4096",
    "C_7:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Enter_6:out0",
    "BiasAdd:in1"
   ],
   [
    "Switch_3:out1",
    "Identity_3:in0"
   ],
   [
    "Identity_3:out4096",
    "C_8:in0"
   ],
   [
    "Identity_3:out4096",
    "C_9:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "MatMul_1:out0",
    "BiasAdd_1:in0"
   ],
   [
    "Enter_7:out0",
    "BiasAdd_1:in1"
   ],
   [
    "Merge_1:out0",
    "Switch_3:in0"
   ],
   [
    "Enter_8:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Enter_9:out0",
    "MatMul:in1"
   ],
   [
    "C_10:out0",
    "Enter_6:in0"
   ],
   [
    "Add:out0",
    "NextIteration_3:in0"
   ],
   [
    "ConcatV2_1:out0",
    "MatMul_1:in0"
   ],
   [
    "Enter_10:out0",
    "MatMul_1:in1"
   ],
   [
    "C_11:out0",
    "Enter_7:in0"
   ],
   [
    "C_12:out0",
    "Enter_8:in0"
   ],
   [
    "Identity_2:out0",
    "ConcatV2_1:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Mul_2:out0",
    "ConcatV2:in1"
   ],
   [
    "C_13:out0",
    "ConcatV2:in2"
   ],
   [
    "C_14:out0",
    "Enter_9:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2_1:in0"
   ],
   [
    "C_15:out0",
    "ConcatV2_1:in2"
   ],
   [
    "C_16:out0",
    "Enter_10:in0"
   ],
   [
    "Split:out0",
    "Mul_2:in0"
   ],
   [
    "Identity_2:out0",
    "Mul_2:in1"
   ],
   [
    "Identity_3:out4096",
    "C_13:in0"
   ],
   [
    "Enter_11:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_12:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "Identity_3:out4096",
    "C_15:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_11:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_12:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "C_17:out0",
    "TensorArrayScatterV3:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayGatherV3:out0"
  ],
  "acu_lys_alias": [
   "gru"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gru:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayGatherV3:out0",
    "gru:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gru": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "num_units": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter_10:out0'])[1] / 2"
    ],
    "reset_after": [
     "BOOL",
     "VALUE",
     false
    ]
   }
  },
  "blob_map": {
   "gru": {
    "k_iz": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[0][0]"
    ],
    "k_ir": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[0][1]"
    ],
    "k_ih": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[0][2]"
    ],
    "k_rz": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[1][0]"
    ],
    "k_rr": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[1][1]"
    ],
    "k_rh": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[1][2]"
    ],
    "b_iz": [
     "CODE",
     "self.gru_bias(tensor['C_11:out0'], tensor['C_10:out0'])[0][0]"
    ],
    "b_ir": [
     "CODE",
     "self.gru_bias(tensor['C_11:out0'], tensor['C_10:out0'])[0][1]"
    ],
    "b_ih": [
     "CODE",
     "self.gru_bias(tensor['C_11:out0'], tensor['C_10:out0'])[0][2]"
    ],
    "b_rz": [
     "CODE",
     "self.gru_bias(tensor['C_11:out0'], tensor['C_10:out0'])[1][0]"
    ],
    "b_rr": [
     "CODE",
     "self.gru_bias(tensor['C_11:out0'], tensor['C_10:out0'])[1][1]"
    ],
    "b_rh": [
     "CODE",
     "self.gru_bias(tensor['C_11:out0'], tensor['C_10:out0'])[1][2]"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "gru_out_hstate_1_13_2",
  "src_ops_alias": [
   "TensorArrayGatherV3",
   "Exit",
   "TensorArrayV3",
   "Range",
   "Exit_1",
   "Switch",
   "C",
   "C_1",
   "TensorArraySizeV3",
   "C_2",
   "Switch_1",
   "Merge",
   "LoopCond",
   "Merge_1",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "Enter_1",
   "NextIteration_1",
   "C_3",
   "Add",
   "Less",
   "Less_1",
   "TensorArrayWriteV3",
   "Mul",
   "Mul_1",
   "Merge_2",
   "Enter_2",
   "Merge_3",
   "Enter_3",
   "Enter_4",
   "Identity",
   "Identity_1",
   "Split",
   "Identity_2",
   "Sub",
   "Tanh",
   "Enter_5",
   "NextIteration_2",
   "Enter_6",
   "NextIteration_3",
   "C_4",
   "Switch_2",
   "C_5",
   "Sigmoid",
   "C_6",
   "BiasAdd",
   "C_7",
   "Add_1",
   "C_8",
   "Add_2",
   "Identity_3",
   "BiasAdd_1",
   "MatMul",
   "Enter_7",
   "C_9",
   "C_10",
   "Switch_3",
   "MatMul_1",
   "Enter_8",
   "ConcatV2",
   "Enter_9",
   "C_11",
   "ConcatV2_1",
   "Enter_10",
   "C_12",
   "TensorArrayReadV3",
   "Mul_2",
   "C_13",
   "C_14",
   "C_15",
   "C_16",
   "Enter_11",
   "Enter_12",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "C_17"
  ],
  "src_inter_flow": [
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "Exit_1:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "C_1:out0",
    "Range:in0"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "C_2:out0",
    "Range:in2"
   ],
   [
    "Switch_1:out0",
    "Exit_1:in0"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "Exit_1:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Merge_1:out0",
    "Switch_1:in0"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "Enter_1:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "C_3:out0",
    "Enter:in0"
   ],
   [
    "Add:out0",
    "NextIteration:in0"
   ],
   [
    "TensorArrayV3:out1",
    "Enter_1:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration_1:in0"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in1"
   ],
   [
    "Merge_2:out0",
    "Less:in0"
   ],
   [
    "Enter_2:out0",
    "Less:in1"
   ],
   [
    "Merge_3:out0",
    "Less_1:in0"
   ],
   [
    "Enter_3:out0",
    "Less_1:in1"
   ],
   [
    "Add:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Enter_4:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "Split:out1",
    "Mul:in0"
   ],
   [
    "Identity_2:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Enter_2:in0"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Tanh:out0",
    "Mul_1:in1"
   ],
   [
    "Enter_5:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_4:in0"
   ],
   [
    "Enter_6:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "Switch_1:out1",
    "Identity_1:in0"
   ],
   [
    "C_4:out0",
    "Enter_3:in0"
   ],
   [
    "Switch:out1",
    "Identity_2:in0"
   ],
   [
    "Switch_2:out1",
    "Identity:in0"
   ],
   [
    "C_5:out0",
    "Split:in0"
   ],
   [
    "Sigmoid:out0",
    "Split:in1"
   ],
   [
    "Split:out1",
    "Sub:in1"
   ],
   [
    "C_6:out0",
    "Sub:in0"
   ],
   [
    "BiasAdd:out0",
    "Tanh:in0"
   ],
   [
    "C_7:out0",
    "Enter_5:in0"
   ],
   [
    "Add_1:out0",
    "NextIteration_2:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "C_8:out0",
    "Enter_6:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_2:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_3:in0"
   ],
   [
    "Identity_3:out4096",
    "C_5:in0"
   ],
   [
    "BiasAdd_1:out0",
    "Sigmoid:in0"
   ],
   [
    "Identity_3:out4096",
    "C_6:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Enter_7:out0",
    "BiasAdd:in1"
   ],
   [
    "Identity_3:out0",
    "Add_1:in0"
   ],
   [
    "Identity:out0",
    "Add_2:in0"
   ],
   [
    "C_9:out0",
    "Add_1:in1"
   ],
   [
    "C_10:out0",
    "Add_2:in1"
   ],
   [
    "Switch_3:out1",
    "Identity_3:in0"
   ],
   [
    "MatMul_1:out0",
    "BiasAdd_1:in0"
   ],
   [
    "Enter_8:out0",
    "BiasAdd_1:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Enter_9:out0",
    "MatMul:in1"
   ],
   [
    "Merge_2:out0",
    "Switch_3:in0"
   ],
   [
    "Identity_3:out4096",
    "C_9:in0"
   ],
   [
    "C_11:out0",
    "Enter_7:in0"
   ],
   [
    "Identity_3:out4096",
    "C_10:in0"
   ],
   [
    "ConcatV2_1:out0",
    "MatMul_1:in0"
   ],
   [
    "Enter_10:out0",
    "MatMul_1:in1"
   ],
   [
    "C_12:out0",
    "Enter_8:in0"
   ],
   [
    "Identity_2:out0",
    "ConcatV2_1:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Mul_2:out0",
    "ConcatV2:in1"
   ],
   [
    "C_13:out0",
    "ConcatV2:in2"
   ],
   [
    "C_14:out0",
    "Enter_9:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2_1:in0"
   ],
   [
    "C_15:out0",
    "ConcatV2_1:in2"
   ],
   [
    "Split:out0",
    "Mul_2:in0"
   ],
   [
    "Identity_2:out0",
    "Mul_2:in1"
   ],
   [
    "C_16:out0",
    "Enter_10:in0"
   ],
   [
    "Identity_3:out4096",
    "C_13:in0"
   ],
   [
    "Enter_11:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_12:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "Identity_3:out4096",
    "C_15:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_11:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_12:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "C_17:out0",
    "TensorArrayScatterV3:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayGatherV3:out0",
   "Exit:out0"
  ],
  "acu_lys_alias": [
   "gru"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gru:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayGatherV3:out0",
    "gru:out0"
   ],
   [
    "Exit:out0",
    "gru:out1"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gru": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "num_units": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter_10:out0'])[1] / 2"
    ]
   }
  },
  "reset_after": [
   "BOOL",
   "VALUE",
   false
  ],
  "blob_map": {
   "gru": {
    "k_iz": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[0][0]"
    ],
    "k_ir": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[0][1]"
    ],
    "k_ih": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[0][2]"
    ],
    "k_rz": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[1][0]"
    ],
    "k_rr": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[1][1]"
    ],
    "k_rh": [
     "CODE",
     "self.gru_kernel(tensor['C_16:out0'], tensor['C_14:out0'])[1][2]"
    ],
    "b_iz": [
     "CODE",
     "self.gru_bias(tensor['C_12:out0'], tensor['C_11:out0'])[0][0]"
    ],
    "b_ir": [
     "CODE",
     "self.gru_bias(tensor['C_12:out0'], tensor['C_11:out0'])[0][1]"
    ],
    "b_ih": [
     "CODE",
     "self.gru_bias(tensor['C_12:out0'], tensor['C_11:out0'])[0][2]"
    ],
    "b_rz": [
     "CODE",
     "self.gru_bias(tensor['C_12:out0'], tensor['C_11:out0'])[1][0]"
    ],
    "b_rr": [
     "CODE",
     "self.gru_bias(tensor['C_12:out0'], tensor['C_11:out0'])[1][1]"
    ],
    "b_rh": [
     "CODE",
     "self.gru_bias(tensor['C_12:out0'], tensor['C_11:out0'])[1][2]"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "gru_false_seq_1_13_2",
  "src_ops_alias": [
   "Exit",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "C",
   "Add",
   "Less",
   "Less_1",
   "Mul",
   "Mul_1",
   "Merge_1",
   "Enter_1",
   "Merge_2",
   "Enter_2",
   "Split",
   "Identity",
   "Sub",
   "Tanh",
   "Enter_3",
   "NextIteration_1",
   "C_1",
   "Enter_4",
   "NextIteration_2",
   "C_2",
   "C_3",
   "Sigmoid",
   "C_4",
   "BiasAdd",
   "C_5",
   "Add_1",
   "C_6",
   "Add_2",
   "Identity_1",
   "BiasAdd_1",
   "MatMul",
   "Enter_5",
   "C_7",
   "Identity_2",
   "C_8",
   "Switch_1",
   "MatMul_1",
   "Enter_6",
   "ConcatV2",
   "Enter_7",
   "C_9",
   "Switch_2",
   "ConcatV2_1",
   "Enter_8",
   "C_10",
   "TensorArrayReadV3",
   "Mul_2",
   "C_11",
   "C_12",
   "C_13",
   "C_14",
   "Enter_9",
   "Enter_10",
   "TensorArrayV3",
   "TensorArrayScatterV3",
   "C_15"
  ],
  "src_inter_flow": [
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "C:out0",
    "Enter:in0"
   ],
   [
    "Add:out0",
    "NextIteration:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in1"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Enter_1:out0",
    "Less:in1"
   ],
   [
    "Merge_2:out0",
    "Less_1:in0"
   ],
   [
    "Enter_2:out0",
    "Less_1:in1"
   ],
   [
    "Split:out1",
    "Mul:in0"
   ],
   [
    "Identity:out0",
    "Mul:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Tanh:out0",
    "Mul_1:in1"
   ],
   [
    "Enter_3:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "C_1:out0",
    "Enter_1:in0"
   ],
   [
    "Enter_4:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "Switch:out1",
    "Identity:in0"
   ],
   [
    "C_2:out0",
    "Enter_2:in0"
   ],
   [
    "C_3:out0",
    "Split:in0"
   ],
   [
    "Sigmoid:out0",
    "Split:in1"
   ],
   [
    "Split:out1",
    "Sub:in1"
   ],
   [
    "C_4:out0",
    "Sub:in0"
   ],
   [
    "BiasAdd:out0",
    "Tanh:in0"
   ],
   [
    "C_5:out0",
    "Enter_3:in0"
   ],
   [
    "Add_1:out0",
    "NextIteration_1:in0"
   ],
   [
    "C_6:out0",
    "Enter_4:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_2:in0"
   ],
   [
    "Identity_1:out4096",
    "C_3:in0"
   ],
   [
    "BiasAdd_1:out0",
    "Sigmoid:in0"
   ],
   [
    "Identity_1:out4096",
    "C_4:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Enter_5:out0",
    "BiasAdd:in1"
   ],
   [
    "Identity_1:out0",
    "Add_1:in0"
   ],
   [
    "C_7:out0",
    "Add_1:in1"
   ],
   [
    "Identity_2:out0",
    "Add_2:in0"
   ],
   [
    "C_8:out0",
    "Add_2:in1"
   ],
   [
    "Switch_1:out1",
    "Identity_1:in0"
   ],
   [
    "MatMul_1:out0",
    "BiasAdd_1:in0"
   ],
   [
    "Enter_6:out0",
    "BiasAdd_1:in1"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Enter_7:out0",
    "MatMul:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Identity_1:out4096",
    "C_7:in0"
   ],
   [
    "C_9:out0",
    "Enter_5:in0"
   ],
   [
    "Merge_1:out0",
    "Switch_1:in0"
   ],
   [
    "Identity_1:out4096",
    "C_8:in0"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "ConcatV2_1:out0",
    "MatMul_1:in0"
   ],
   [
    "Enter_8:out0",
    "MatMul_1:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "C_10:out0",
    "Enter_6:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_2:in0"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Mul_2:out0",
    "ConcatV2:in1"
   ],
   [
    "C_11:out0",
    "ConcatV2:in2"
   ],
   [
    "Identity:out0",
    "ConcatV2_1:in1"
   ],
   [
    "C_12:out0",
    "Enter_7:in0"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2_1:in0"
   ],
   [
    "C_13:out0",
    "ConcatV2_1:in2"
   ],
   [
    "Split:out0",
    "Mul_2:in0"
   ],
   [
    "Identity:out0",
    "Mul_2:in1"
   ],
   [
    "C_14:out0",
    "Enter_8:in0"
   ],
   [
    "Identity_2:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "Identity_1:out4096",
    "C_11:in0"
   ],
   [
    "Enter_9:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_10:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "Identity_1:out4096",
    "C_13:in0"
   ],
   [
    "C_1:out0",
    "TensorArrayV3:in0"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_9:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_10:in0"
   ],
   [
    "TensorArrayV3:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "C_15:out0",
    "TensorArrayScatterV3:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "Exit:out0"
  ],
  "acu_lys_alias": [
   "gru"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gru:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Exit:out0",
    "gru:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gru": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "num_units": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter_8:out0'])[1] / 2"
    ],
    "return_sequences": [
     "BOOL",
     "VALUE",
     false
    ]
   }
  },
  "reset_after": [
   "BOOL",
   "VALUE",
   false
  ],
  "blob_map": {
   "gru": {
    "k_iz": [
     "CODE",
     "self.gru_kernel(tensor['C_14:out0'], tensor['C_12:out0'])[0][0]"
    ],
    "k_ir": [
     "CODE",
     "self.gru_kernel(tensor['C_14:out0'], tensor['C_12:out0'])[0][1]"
    ],
    "k_ih": [
     "CODE",
     "self.gru_kernel(tensor['C_14:out0'], tensor['C_12:out0'])[0][2]"
    ],
    "k_rz": [
     "CODE",
     "self.gru_kernel(tensor['C_14:out0'], tensor['C_12:out0'])[1][0]"
    ],
    "k_rr": [
     "CODE",
     "self.gru_kernel(tensor['C_14:out0'], tensor['C_12:out0'])[1][1]"
    ],
    "k_rh": [
     "CODE",
     "self.gru_kernel(tensor['C_14:out0'], tensor['C_12:out0'])[1][2]"
    ],
    "b_iz": [
     "CODE",
     "self.gru_bias(tensor['C_10:out0'], tensor['C_9:out0'])[0][0]"
    ],
    "b_ir": [
     "CODE",
     "self.gru_bias(tensor['C_10:out0'], tensor['C_9:out0'])[0][1]"
    ],
    "b_ih": [
     "CODE",
     "self.gru_bias(tensor['C_10:out0'], tensor['C_9:out0'])[0][2]"
    ],
    "b_rz": [
     "CODE",
     "self.gru_bias(tensor['C_10:out0'], tensor['C_9:out0'])[1][0]"
    ],
    "b_rr": [
     "CODE",
     "self.gru_bias(tensor['C_10:out0'], tensor['C_9:out0'])[1][1]"
    ],
    "b_rh": [
     "CODE",
     "self.gru_bias(tensor['C_10:out0'], tensor['C_9:out0'])[1][2]"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "gru_cell",
  "src_ops_alias": [
   "Add",
   "Mul",
   "Mul_1",
   "Split",
   "Sub",
   "Tanh",
   "C",
   "Sigmoid",
   "C_1",
   "BiasAdd",
   "BiasAdd_1",
   "MatMul",
   "C_2",
   "MatMul_1",
   "C_3",
   "ConcatV2",
   "C_4",
   "ConcatV2_1",
   "C_5",
   "Mul_2",
   "C_6",
   "C_7"
  ],
  "src_inter_flow": [
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in1"
   ],
   [
    "Split:out1",
    "Mul:in0"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Tanh:out0",
    "Mul_1:in1"
   ],
   [
    "C:out0",
    "Split:in0"
   ],
   [
    "Sigmoid:out0",
    "Split:in1"
   ],
   [
    "Split:out1",
    "Sub:in1"
   ],
   [
    "C_1:out0",
    "Sub:in0"
   ],
   [
    "BiasAdd:out0",
    "Tanh:in0"
   ],
   [
    "BiasAdd_1:out0",
    "Sigmoid:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "C_2:out0",
    "BiasAdd:in1"
   ],
   [
    "MatMul_1:out0",
    "BiasAdd_1:in0"
   ],
   [
    "C_3:out0",
    "BiasAdd_1:in1"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "C_4:out0",
    "MatMul:in1"
   ],
   [
    "ConcatV2_1:out0",
    "MatMul_1:in0"
   ],
   [
    "C_5:out0",
    "MatMul_1:in1"
   ],
   [
    "Mul_2:out0",
    "ConcatV2:in1"
   ],
   [
    "C_6:out0",
    "ConcatV2:in2"
   ],
   [
    "C_7:out0",
    "ConcatV2_1:in2"
   ],
   [
    "Split:out0",
    "Mul_2:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ConcatV2_1:in0"
   ],
   [
    "I:out0",
    "ConcatV2:in0"
   ],
   [
    "I_1:out0",
    "ConcatV2_1:in1"
   ],
   [
    "I_1:out0",
    "Mul_2:in1"
   ],
   [
    "I_1:out0",
    "Mul:in1"
   ]
  ],
  "src_out_tensor": [
   "Add:out0"
  ],
  "acu_lys_alias": [
   "gru_cell"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "gru_cell:in0"
   ],
   [
    "I_1:out0",
    "gru_cell:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Add:out0",
    "gru_cell:out0"
   ],
   [
    "Add:out0",
    "gru_cell:out1"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "gru_cell": {
    "num_units": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_5:out0'])[1] / 2"
    ],
    "reset_after": [
     "BOOL",
     "VALUE",
     false
    ]
   }
  },
  "blob_map": {
   "gru_cell": {
    "k_iz": [
     "CODE",
     "self.gru_kernel(tensor['C_5:out0'], tensor['C_4:out0'])[0][0]"
    ],
    "k_ir": [
     "CODE",
     "self.gru_kernel(tensor['C_5:out0'], tensor['C_4:out0'])[0][1]"
    ],
    "k_ih": [
     "CODE",
     "self.gru_kernel(tensor['C_5:out0'], tensor['C_4:out0'])[0][2]"
    ],
    "k_rz": [
     "CODE",
     "self.gru_kernel(tensor['C_5:out0'], tensor['C_4:out0'])[1][0]"
    ],
    "k_rr": [
     "CODE",
     "self.gru_kernel(tensor['C_5:out0'], tensor['C_4:out0'])[1][1]"
    ],
    "k_rh": [
     "CODE",
     "self.gru_kernel(tensor['C_5:out0'], tensor['C_4:out0'])[1][2]"
    ],
    "b_iz": [
     "CODE",
     "self.gru_bias(tensor['C_3:out0'], tensor['C_2:out0'])[0][0]"
    ],
    "b_ir": [
     "CODE",
     "self.gru_bias(tensor['C_3:out0'], tensor['C_2:out0'])[0][1]"
    ],
    "b_ih": [
     "CODE",
     "self.gru_bias(tensor['C_3:out0'], tensor['C_2:out0'])[0][2]"
    ],
    "b_rz": [
     "CODE",
     "self.gru_bias(tensor['C_3:out0'], tensor['C_2:out0'])[1][0]"
    ],
    "b_rr": [
     "CODE",
     "self.gru_bias(tensor['C_3:out0'], tensor['C_2:out0'])[1][1]"
    ],
    "b_rh": [
     "CODE",
     "self.gru_bias(tensor['C_3:out0'], tensor['C_2:out0'])[1][2]"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "reverse_sequence_to_reverse",
  "src_ops_alias": [
   "ReverseSequence"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "ReverseSequence:in0"
   ],
   [
    "I_1:out0",
    "ReverseSequence:in1"
   ]
  ],
  "src_out_tensor": [
   "ReverseSequence:out0"
  ],
  "acu_lys_alias": [
   "reverse"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reverse:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "ReverseSequence:out0",
    "reverse:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "reverse": {
    "axis": [
     "INTS",
     "CODE",
     "[self.attr_pick(node['ReverseSequence'], 'seq_dim')]"
    ]
   }
  },
  "blob_map": {
   "reverse": {}
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_name\": \"I_1:out0\"}",
   "def r_reverse_sequence_pre_condition(self, node, tensor, input_name):",
   "    input_tensor_shape = self.shape_pick(tensor[input_name])",
   "    return len(input_tensor_shape) == 1",
   ""
  ]
 },
 {
  "ruler_name": "relu20",
  "src_ops_alias": [
   "Mul",
   "RealDiv",
   "C",
   "Relu6",
   "C_1",
   "Mul_1",
   "RealDiv_1",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Mul:in1"
   ],
   [
    "C_1:out0",
    "RealDiv:in1"
   ],
   [
    "RealDiv_1:out0",
    "Mul_1:in0"
   ],
   [
    "Mul_1:out0",
    "Relu6:in0"
   ],
   [
    "C_3:out0",
    "RealDiv_1:in1"
   ],
   [
    "RealDiv:out0",
    "Mul:in0"
   ],
   [
    "Relu6:out0",
    "RealDiv:in0"
   ],
   [
    "C_2:out0",
    "Mul_1:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "RealDiv_1:in0"
   ]
  ],
  "src_out_tensor": [
   "Mul:out0"
  ],
  "acu_lys_alias": [
   "relun"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "relun:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Mul:out0",
    "relun:out0"
   ]
  ],
  "param_map": {
   "relun": {
    "relu_clamp_top": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_3:out0\"}",
      "def r_relu20_get_n(self, node, tensor, tensor_name):",
      "    return self.tensor_to_numpy(tensor[tensor_name])",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "relun": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "single_timestep_lstm_to_lstmunit",
  "src_ops_alias": [
   "Transpose",
   "TensorArrayGatherV3",
   "C",
   "TensorArrayV3",
   "Range",
   "Exit",
   "C_1",
   "C_2",
   "TensorArraySizeV3",
   "C_3",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "TensorArrayWriteV3",
   "Less",
   "Less_1",
   "Enter_1",
   "Identity",
   "Mul",
   "Identity_1",
   "Merge_1",
   "Enter_2",
   "Merge_2",
   "Enter_3",
   "Switch_1",
   "Sigmoid",
   "Tanh",
   "Enter_4",
   "NextIteration_1",
   "Enter_5",
   "NextIteration_2",
   "C_4",
   "Split",
   "Add",
   "C_5",
   "Add_1",
   "C_6",
   "Add_2",
   "C_7",
   "BiasAdd",
   "Mul_1",
   "Mul_2",
   "Identity_2",
   "C_8",
   "C_9",
   "MatMul",
   "Enter_6",
   "Sigmoid_1",
   "Identity_3",
   "Sigmoid_2",
   "Tanh_1",
   "Switch_2",
   "ConcatV2",
   "Enter_7",
   "C_10",
   "Add_3",
   "Switch_3",
   "TensorArrayReadV3",
   "Identity_4",
   "C_11",
   "C_12",
   "C_13",
   "Merge_3",
   "Enter_8",
   "Enter_9",
   "Switch_4",
   "Enter_10",
   "NextIteration_3",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "Merge_4",
   "C_14",
   "Transpose_1",
   "Enter_11",
   "NextIteration_4",
   "C_15",
   "Exit_1",
   "Exit_2"
  ],
  "src_inter_flow": [
   [
    "C_1:out0",
    "TensorArrayV3:in0"
   ],
   [
    "Exit:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Merge_3:out0",
    "Switch_3:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "C_6:out0",
    "Enter_5:in0"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "Identity_4:out0",
    "ConcatV2:in1"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "Enter_5:out0",
    "Merge_2:in0"
   ],
   [
    "C_14:out0",
    "TensorArrayScatterV3:in1"
   ],
   [
    "C_9:out0",
    "Add_2:in1"
   ],
   [
    "Add:out0",
    "Tanh:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "Sigmoid:out0",
    "Mul:in0"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Add:out0",
    "NextIteration_3:in0"
   ],
   [
    "Add_3:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Identity:out0",
    "Add_2:in0"
   ],
   [
    "NextIteration_4:out0",
    "Merge_4:in1"
   ],
   [
    "Merge_2:out0",
    "Less_1:in0"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in0"
   ],
   [
    "Enter_11:out0",
    "Merge_4:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_2:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Enter_6:out0",
    "BiasAdd:in1"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "Switch_3:out1",
    "Identity_3:in0"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "C_12:out0",
    "Enter_7:in0"
   ],
   [
    "Enter_4:out0",
    "Merge_1:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "C_13:out0",
    "Add_3:in1"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_4:in1"
   ],
   [
    "Split:out2",
    "Add_3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "C_10:out0",
    "Enter_6:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "Enter_7:out0",
    "MatMul:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "C_1:out0",
    "Enter_2:in0"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "C_11:out0",
    "ConcatV2:in2"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "C_8:out0",
    "Add_1:in1"
   ],
   [
    "Identity_2:out0",
    "Add_1:in0"
   ],
   [
    "Enter_3:out0",
    "Less_1:in1"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_8:in0"
   ],
   [
    "C_4:out0",
    "Enter_3:in0"
   ],
   [
    "C_3:out0",
    "Range:in2"
   ],
   [
    "Exit:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Identity_2:out4096",
    "C_11:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "C_7:out0",
    "Split:in0"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "C_5:out0",
    "Enter_4:in0"
   ],
   [
    "C_15:out0",
    "Transpose_1:in1"
   ],
   [
    "Switch_4:out0",
    "Exit_1:in0"
   ],
   [
    "Mul_1:out0",
    "Add:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "Identity_2:out4096",
    "C_7:in0"
   ],
   [
    "Merge_4:out0",
    "Switch_4:in0"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "Switch_3:out0",
    "Exit_2:in0"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "Identity_2:out4096",
    "C_9:in0"
   ],
   [
    "C_2:out0",
    "Range:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "Merge_2:out0",
    "Switch_1:in0"
   ],
   [
    "Switch_4:out1",
    "Identity_4:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_9:in0"
   ],
   [
    "Transpose_1:out0",
    "TensorArrayScatterV3:in2"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "Enter_10:out0",
    "Merge_3:in0"
   ],
   [
    "Identity_2:out4096",
    "C_13:in0"
   ],
   [
    "Identity_3:out0",
    "Mul_1:in1"
   ],
   [
    "C:out0",
    "Transpose:in1"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "C_1:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "Add_1:out0",
    "NextIteration_1:in0"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_1:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Enter_8:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_9:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "Mul:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Identity_2:out4096",
    "C_8:in0"
   ],
   [
    "TensorArrayGatherV3:out0",
    "Transpose:in0"
   ],
   [
    "Enter_1:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "Enter_2:out0",
    "Less:in1"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "Mul:out0",
    "NextIteration_4:in0"
   ],
   [
    "Merge_1:out0",
    "Switch_2:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Transpose_1:in0"
   ],
   [
    "I_1:out0",
    "Enter_11:in0"
   ],
   [
    "I_2:out0",
    "Enter_10:in0"
   ]
  ],
  "src_out_tensor": [
   "Transpose:out0",
   "Exit_1:out0",
   "Exit_2:out0"
  ],
  "acu_lys_alias": [
   "reshape_1",
   "lstmunit",
   "reshape_2"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reshape_1:in0"
   ],
   [
    "I_1:out0",
    "lstmunit:in1"
   ],
   [
    "I_2:out0",
    "lstmunit:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Transpose:out0",
    "reshape_2:out0"
   ],
   [
    "Exit_1:out0",
    "lstmunit:out1"
   ],
   [
    "Exit_2:out0",
    "lstmunit:out2"
   ]
  ],
  "param_map": {
   "lstmunit": {
    "weights": [
     "INT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_12:out0\"}",
      "def r_lstmunit_rule_get_param_weights(self, node, tensor, tensor_name): # 'C_2:out0'",
      "    return self.shape_pick(tensor[tensor_name])[1] / 4",
      ""
     ]
    ],
    "num_proj": [
     "ORIGIN",
     "VALUE",
     null
    ],
    "forget_bias": [
     "FLOAT",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"tensor_name\": \"C_13:out0\"}",
      "def r_lstmunit_rule_get_param_forget_bias(self, node, tensor, tensor_name): #",
      "    forget_bias = self.tensor_to_numpy(tensor[tensor_name])",
      "    return forget_bias",
      ""
     ]
    ]
   },
   "reshape_1": {
    "shape": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"input_name\": \"I:out0\"}",
      "def r_lstmunit_rule_lstm_to_lstmunit_reshape1_shape(self, node, tensor, input_name):",
      "    input_tensor_shape = self.shape_pick(tensor[input_name])",
      "    batch = 0 if input_tensor_shape[0] == 1 else input_tensor_shape[0]",
      "    return [batch, input_tensor_shape[2]]",
      ""
     ]
    ]
   },
   "reshape_2": {
    "shape": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"input_name\": \"I:out0\", \"kernel_name\": \"C_12:out0\"}",
      "def r_lstmunit_rule_lstm_to_lstmunit_reshape2_shape(self, node, tensor, input_name, kernel_name):",
      "    input_tensor_shape = self.shape_pick(tensor[input_name])",
      "    weights = self.shape_pick(tensor[kernel_name])[1] / 4",
      "    batch = 0 if input_tensor_shape[0] == 1 else input_tensor_shape[0]",
      "    return [batch, input_tensor_shape[1], int(weights)]",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "lstmunit": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_12:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_10:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_10:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_10:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_10:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "acu_inter_flow": [
   [
    "reshape_1:out0",
    "lstmunit:in0"
   ],
   [
    "lstmunit:out0",
    "reshape_2:in0"
   ]
  ],
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_name\": \"I:out0\"}",
   "def r_lstmunit_rule_lstm_to_lstmunit_pre_condition(self, node, tensor, input_name):",
   "    input_tensor_shape = self.shape_pick(tensor[input_name])",
   "    return input_tensor_shape[1] == 1 # convert single timestep lstm to lstmunit",
   ""
  ]
 },
 {
  "ruler_name": "r_keras_time_major_lstmlayer",
  "src_ops_alias": [
   "TensorArrayReadV3",
   "TensorArrayV3",
   "Sub",
   "Exit",
   "C",
   "Exit_1",
   "C_1",
   "Switch",
   "Switch_1",
   "Merge",
   "LoopCond",
   "Merge_1",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "Enter_1",
   "NextIteration_1",
   "TensorArrayWriteV3",
   "Less",
   "Less_1",
   "C_2",
   "Add",
   "Enter_2",
   "Identity",
   "Mul",
   "Identity_1",
   "Merge_2",
   "Enter_3",
   "Enter_4",
   "C_3",
   "ClipByValue",
   "Tanh",
   "Enter_5",
   "NextIteration_2",
   "C_4",
   "Identity_2",
   "Add_1",
   "C_5",
   "C_6",
   "Add_2",
   "C_7",
   "Add_3",
   "Switch_2",
   "Mul_1",
   "C_8",
   "Mul_2",
   "Mul_3",
   "C_9",
   "C_10",
   "Add_4",
   "ClipByValue_1",
   "Identity_3",
   "ClipByValue_2",
   "Tanh_1",
   "BiasAdd",
   "MatMul",
   "Add_5",
   "C_11",
   "C_12",
   "Switch_3",
   "Add_6",
   "C_13",
   "C_14",
   "Add_7",
   "MatMul_1",
   "Enter_6",
   "Identity_4",
   "Enter_7",
   "Mul_4",
   "C_15",
   "Merge_3",
   "Mul_5",
   "C_16",
   "BiasAdd_1",
   "MatMul_2",
   "TensorArrayReadV3_1",
   "Enter_8",
   "C_17",
   "Switch_4",
   "C_18",
   "C_19",
   "Add_8",
   "Enter_9",
   "NextIteration_3",
   "C_20",
   "Add_9",
   "MatMul_3",
   "Enter_10",
   "Enter_11",
   "Enter_12",
   "Enter_13",
   "C_21",
   "Merge_4",
   "BiasAdd_2",
   "MatMul_4",
   "C_22",
   "BiasAdd_3",
   "MatMul_5",
   "Enter_14",
   "C_23",
   "C_24",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "Enter_15",
   "NextIteration_4",
   "MatMul_6",
   "Enter_16",
   "Enter_17",
   "MatMul_7",
   "Enter_18",
   "Enter_19",
   "C_25",
   "C_26",
   "C_27",
   "Enter_20",
   "C_28",
   "C_29",
   "Enter_21",
   "C_30",
   "C_31",
   "C_32",
   "C_33"
  ],
  "src_inter_flow": [
   [
    "TensorArrayV3:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Sub:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "Exit:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "Exit_1:out0",
    "Sub:in0"
   ],
   [
    "C_1:out0",
    "Sub:in1"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "Switch_1:out0",
    "Exit_1:in0"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Merge_1:out0",
    "Switch_1:in0"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "Enter_1:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "C_2:out0",
    "Enter_1:in0"
   ],
   [
    "Add:out0",
    "NextIteration_1:in0"
   ],
   [
    "Enter_2:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Mul:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "Merge_1:out0",
    "Less_1:in0"
   ],
   [
    "Merge_2:out0",
    "Less:in0"
   ],
   [
    "Enter_3:out0",
    "Less:in1"
   ],
   [
    "Enter_4:out0",
    "Less_1:in1"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_2:in0"
   ],
   [
    "Identity:out0",
    "Add:in0"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "C_3:out0",
    "Add:in1"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "ClipByValue:out0",
    "Mul:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Enter_4:in0"
   ],
   [
    "Enter_5:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "C_4:out0",
    "Enter_3:in0"
   ],
   [
    "Identity_2:out4096",
    "C_3:in0"
   ],
   [
    "Add_1:out0",
    "ClipByValue:in0"
   ],
   [
    "C_5:out0",
    "ClipByValue:in1"
   ],
   [
    "C_6:out0",
    "ClipByValue:in2"
   ],
   [
    "Add_2:out0",
    "Tanh:in0"
   ],
   [
    "C_7:out0",
    "Enter_5:in0"
   ],
   [
    "Add_3:out0",
    "NextIteration_2:in0"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "Mul_1:out0",
    "Add_1:in0"
   ],
   [
    "C_8:out0",
    "Add_1:in1"
   ],
   [
    "Identity_2:out4096",
    "C_5:in0"
   ],
   [
    "Identity_2:out4096",
    "C_6:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Mul_2:out0",
    "Add_2:in0"
   ],
   [
    "Mul_3:out0",
    "Add_2:in1"
   ],
   [
    "Identity_2:out0",
    "Add_3:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_2:in0"
   ],
   [
    "C_9:out0",
    "Add_3:in1"
   ],
   [
    "Identity_2:out4096",
    "C_8:in0"
   ],
   [
    "C_10:out0",
    "Mul_1:in0"
   ],
   [
    "Add_4:out0",
    "Mul_1:in1"
   ],
   [
    "ClipByValue_1:out0",
    "Mul_2:in0"
   ],
   [
    "Identity_3:out0",
    "Mul_2:in1"
   ],
   [
    "Identity_2:out4096",
    "C_9:in0"
   ],
   [
    "ClipByValue_2:out0",
    "Mul_3:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_3:in1"
   ],
   [
    "Identity_2:out4096",
    "C_10:in0"
   ],
   [
    "BiasAdd:out0",
    "Add_4:in0"
   ],
   [
    "MatMul:out0",
    "Add_4:in1"
   ],
   [
    "Add_5:out0",
    "ClipByValue_1:in0"
   ],
   [
    "C_11:out0",
    "ClipByValue_1:in1"
   ],
   [
    "C_12:out0",
    "ClipByValue_1:in2"
   ],
   [
    "Switch_3:out1",
    "Identity_3:in0"
   ],
   [
    "Add_6:out0",
    "ClipByValue_2:in0"
   ],
   [
    "C_13:out0",
    "ClipByValue_2:in1"
   ],
   [
    "C_14:out0",
    "ClipByValue_2:in2"
   ],
   [
    "Add_7:out0",
    "Tanh_1:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "MatMul_1:out0",
    "BiasAdd:in0"
   ],
   [
    "Identity_2:out4096",
    "C_11:in0"
   ],
   [
    "Enter_6:out0",
    "BiasAdd:in1"
   ],
   [
    "Identity_2:out4096",
    "C_12:in0"
   ],
   [
    "Identity_4:out0",
    "MatMul:in0"
   ],
   [
    "Enter_7:out0",
    "MatMul:in1"
   ],
   [
    "Mul_4:out0",
    "Add_5:in0"
   ],
   [
    "C_15:out0",
    "Add_5:in1"
   ],
   [
    "Identity_2:out4096",
    "C_13:in0"
   ],
   [
    "Identity_2:out4096",
    "C_14:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_3:in0"
   ],
   [
    "Mul_5:out0",
    "Add_6:in0"
   ],
   [
    "C_16:out0",
    "Add_6:in1"
   ],
   [
    "BiasAdd_1:out0",
    "Add_7:in0"
   ],
   [
    "MatMul_2:out0",
    "Add_7:in1"
   ],
   [
    "TensorArrayReadV3_1:out0",
    "MatMul_1:in0"
   ],
   [
    "Enter_8:out0",
    "MatMul_1:in1"
   ],
   [
    "Identity_2:out4096",
    "C_15:in0"
   ],
   [
    "C_17:out0",
    "Enter_6:in0"
   ],
   [
    "Switch_4:out1",
    "Identity_4:in0"
   ],
   [
    "C_18:out0",
    "Enter_7:in0"
   ],
   [
    "Identity_2:out4096",
    "C_16:in0"
   ],
   [
    "C_19:out0",
    "Mul_4:in0"
   ],
   [
    "Add_8:out0",
    "Mul_4:in1"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3_1:in1"
   ],
   [
    "Enter_9:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_4:in1"
   ],
   [
    "C_20:out0",
    "Mul_5:in0"
   ],
   [
    "Add_9:out0",
    "Mul_5:in1"
   ],
   [
    "Identity_4:out0",
    "MatMul_2:in0"
   ],
   [
    "MatMul_3:out0",
    "BiasAdd_1:in0"
   ],
   [
    "Enter_10:out0",
    "BiasAdd_1:in1"
   ],
   [
    "Enter_11:out0",
    "MatMul_2:in1"
   ],
   [
    "Identity_2:out4096",
    "C_19:in0"
   ],
   [
    "Enter_12:out0",
    "TensorArrayReadV3_1:in0"
   ],
   [
    "Enter_13:out0",
    "TensorArrayReadV3_1:in2"
   ],
   [
    "C_21:out0",
    "Enter_8:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_3:in0"
   ],
   [
    "Merge_4:out0",
    "Switch_4:in0"
   ],
   [
    "Identity_2:out4096",
    "C_20:in0"
   ],
   [
    "BiasAdd_2:out0",
    "Add_8:in0"
   ],
   [
    "MatMul_4:out0",
    "Add_8:in1"
   ],
   [
    "C_22:out0",
    "Enter_9:in0"
   ],
   [
    "TensorArrayReadV3_1:out0",
    "MatMul_3:in0"
   ],
   [
    "BiasAdd_3:out0",
    "Add_9:in0"
   ],
   [
    "MatMul_5:out0",
    "Add_9:in1"
   ],
   [
    "Enter_14:out0",
    "MatMul_3:in1"
   ],
   [
    "C_23:out0",
    "Enter_10:in0"
   ],
   [
    "C_24:out0",
    "Enter_11:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_12:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_13:in0"
   ],
   [
    "Identity_4:out0",
    "MatMul_4:in0"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "Enter_15:out0",
    "Merge_4:in0"
   ],
   [
    "NextIteration_4:out0",
    "Merge_4:in1"
   ],
   [
    "MatMul_6:out0",
    "BiasAdd_2:in0"
   ],
   [
    "Enter_16:out0",
    "BiasAdd_2:in1"
   ],
   [
    "Identity_4:out0",
    "MatMul_5:in0"
   ],
   [
    "Enter_17:out0",
    "MatMul_4:in1"
   ],
   [
    "Mul:out0",
    "NextIteration_4:in0"
   ],
   [
    "MatMul_7:out0",
    "BiasAdd_3:in0"
   ],
   [
    "Enter_18:out0",
    "BiasAdd_3:in1"
   ],
   [
    "Enter_19:out0",
    "MatMul_5:in1"
   ],
   [
    "C_25:out0",
    "Enter_14:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "TensorArrayReadV3_1:out0",
    "MatMul_6:in0"
   ],
   [
    "C_26:out0",
    "TensorArrayScatterV3:in1"
   ],
   [
    "C_27:out0",
    "Enter_15:in0"
   ],
   [
    "TensorArrayReadV3_1:out0",
    "MatMul_7:in0"
   ],
   [
    "Enter_20:out0",
    "MatMul_6:in1"
   ],
   [
    "C_28:out0",
    "Enter_16:in0"
   ],
   [
    "C_29:out0",
    "Enter_17:in0"
   ],
   [
    "Enter_21:out0",
    "MatMul_7:in1"
   ],
   [
    "C_30:out0",
    "Enter_18:in0"
   ],
   [
    "C_31:out0",
    "Enter_19:in0"
   ],
   [
    "C_32:out0",
    "Enter_20:in0"
   ],
   [
    "C_33:out0",
    "Enter_21:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayReadV3:out0"
  ],
  "acu_lys_alias": [
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstm:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayReadV3:out0",
    "lstm:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "VALUE",
     0
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_33:out0'])[1]"
    ],
    "recurrent_activation": [
     "STRING",
     "VALUE",
     "hard_sigmoid"
    ],
    "return_sequences": [
     "BOOL",
     "VALUE",
     false
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_33:out0'])"
    ],
    "wf": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_32:out0'])"
    ],
    "wc": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_25:out0'])"
    ],
    "wo": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_21:out0'])"
    ],
    "hi": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_31:out0'])"
    ],
    "hf": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_29:out0'])"
    ],
    "hc": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_24:out0'])"
    ],
    "ho": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_18:out0'])"
    ],
    "bi": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_30:out0'])"
    ],
    "bf": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_28:out0'])"
    ],
    "bc": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_23:out0'])"
    ],
    "bo": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_17:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_keras_time_major_reverse_lstmlayer",
  "src_ops_alias": [
   "TensorArrayReadV3",
   "TensorArrayV3",
   "Sub",
   "Exit",
   "C",
   "Exit_1",
   "C_1",
   "Switch",
   "Switch_1",
   "Merge",
   "LoopCond",
   "Merge_1",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "Enter_1",
   "NextIteration_1",
   "TensorArrayWriteV3",
   "Less",
   "Less_1",
   "C_2",
   "Add",
   "Enter_2",
   "Identity",
   "Mul",
   "Identity_1",
   "Merge_2",
   "Enter_3",
   "Enter_4",
   "C_3",
   "ClipByValue",
   "Tanh",
   "Enter_5",
   "NextIteration_2",
   "C_4",
   "Identity_2",
   "Add_1",
   "C_5",
   "C_6",
   "Add_2",
   "C_7",
   "Add_3",
   "Switch_2",
   "Mul_1",
   "C_8",
   "Mul_2",
   "Mul_3",
   "C_9",
   "C_10",
   "Add_4",
   "ClipByValue_1",
   "Identity_3",
   "ClipByValue_2",
   "Tanh_1",
   "BiasAdd",
   "MatMul",
   "Add_5",
   "C_11",
   "C_12",
   "Switch_3",
   "Add_6",
   "C_13",
   "C_14",
   "Add_7",
   "MatMul_1",
   "Enter_6",
   "Identity_4",
   "Enter_7",
   "Mul_4",
   "C_15",
   "Merge_3",
   "Mul_5",
   "C_16",
   "BiasAdd_1",
   "MatMul_2",
   "TensorArrayReadV3_1",
   "Enter_8",
   "C_17",
   "Switch_4",
   "C_18",
   "C_19",
   "Add_8",
   "Enter_9",
   "NextIteration_3",
   "C_20",
   "Add_9",
   "MatMul_3",
   "Enter_10",
   "Enter_11",
   "Enter_12",
   "Enter_13",
   "C_21",
   "Merge_4",
   "BiasAdd_2",
   "MatMul_4",
   "C_22",
   "BiasAdd_3",
   "MatMul_5",
   "Enter_14",
   "C_23",
   "C_24",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "Enter_15",
   "NextIteration_4",
   "MatMul_6",
   "Enter_16",
   "Enter_17",
   "MatMul_7",
   "Enter_18",
   "Enter_19",
   "C_25",
   "C_26",
   "ReverseV2",
   "C_27",
   "Enter_20",
   "C_28",
   "C_29",
   "Enter_21",
   "C_30",
   "C_31",
   "C_32",
   "C_33",
   "C_34"
  ],
  "src_inter_flow": [
   [
    "TensorArrayV3:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Sub:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "Exit:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "Exit_1:out0",
    "Sub:in0"
   ],
   [
    "C_1:out0",
    "Sub:in1"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "Switch_1:out0",
    "Exit_1:in0"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Merge_1:out0",
    "Switch_1:in0"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "Enter_1:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "C_2:out0",
    "Enter_1:in0"
   ],
   [
    "Add:out0",
    "NextIteration_1:in0"
   ],
   [
    "Enter_2:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Mul:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "Merge_1:out0",
    "Less_1:in0"
   ],
   [
    "Merge_2:out0",
    "Less:in0"
   ],
   [
    "Enter_3:out0",
    "Less:in1"
   ],
   [
    "Enter_4:out0",
    "Less_1:in1"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_2:in0"
   ],
   [
    "Identity:out0",
    "Add:in0"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "C_3:out0",
    "Add:in1"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "ClipByValue:out0",
    "Mul:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Enter_4:in0"
   ],
   [
    "Enter_5:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "C_4:out0",
    "Enter_3:in0"
   ],
   [
    "Identity_2:out4096",
    "C_3:in0"
   ],
   [
    "Add_1:out0",
    "ClipByValue:in0"
   ],
   [
    "C_5:out0",
    "ClipByValue:in1"
   ],
   [
    "C_6:out0",
    "ClipByValue:in2"
   ],
   [
    "Add_2:out0",
    "Tanh:in0"
   ],
   [
    "C_7:out0",
    "Enter_5:in0"
   ],
   [
    "Add_3:out0",
    "NextIteration_2:in0"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "Mul_1:out0",
    "Add_1:in0"
   ],
   [
    "C_8:out0",
    "Add_1:in1"
   ],
   [
    "Identity_2:out4096",
    "C_5:in0"
   ],
   [
    "Identity_2:out4096",
    "C_6:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Mul_2:out0",
    "Add_2:in0"
   ],
   [
    "Mul_3:out0",
    "Add_2:in1"
   ],
   [
    "Identity_2:out0",
    "Add_3:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_2:in0"
   ],
   [
    "C_9:out0",
    "Add_3:in1"
   ],
   [
    "Identity_2:out4096",
    "C_8:in0"
   ],
   [
    "C_10:out0",
    "Mul_1:in0"
   ],
   [
    "Add_4:out0",
    "Mul_1:in1"
   ],
   [
    "ClipByValue_1:out0",
    "Mul_2:in0"
   ],
   [
    "Identity_3:out0",
    "Mul_2:in1"
   ],
   [
    "Identity_2:out4096",
    "C_9:in0"
   ],
   [
    "ClipByValue_2:out0",
    "Mul_3:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_3:in1"
   ],
   [
    "Identity_2:out4096",
    "C_10:in0"
   ],
   [
    "BiasAdd:out0",
    "Add_4:in0"
   ],
   [
    "MatMul:out0",
    "Add_4:in1"
   ],
   [
    "Add_5:out0",
    "ClipByValue_1:in0"
   ],
   [
    "C_11:out0",
    "ClipByValue_1:in1"
   ],
   [
    "C_12:out0",
    "ClipByValue_1:in2"
   ],
   [
    "Switch_3:out1",
    "Identity_3:in0"
   ],
   [
    "Add_6:out0",
    "ClipByValue_2:in0"
   ],
   [
    "C_13:out0",
    "ClipByValue_2:in1"
   ],
   [
    "C_14:out0",
    "ClipByValue_2:in2"
   ],
   [
    "Add_7:out0",
    "Tanh_1:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "MatMul_1:out0",
    "BiasAdd:in0"
   ],
   [
    "Identity_2:out4096",
    "C_11:in0"
   ],
   [
    "Enter_6:out0",
    "BiasAdd:in1"
   ],
   [
    "Identity_2:out4096",
    "C_12:in0"
   ],
   [
    "Identity_4:out0",
    "MatMul:in0"
   ],
   [
    "Enter_7:out0",
    "MatMul:in1"
   ],
   [
    "Mul_4:out0",
    "Add_5:in0"
   ],
   [
    "C_15:out0",
    "Add_5:in1"
   ],
   [
    "Identity_2:out4096",
    "C_13:in0"
   ],
   [
    "Identity_2:out4096",
    "C_14:in0"
   ],
   [
    "Merge_3:out0",
    "Switch_3:in0"
   ],
   [
    "Mul_5:out0",
    "Add_6:in0"
   ],
   [
    "C_16:out0",
    "Add_6:in1"
   ],
   [
    "BiasAdd_1:out0",
    "Add_7:in0"
   ],
   [
    "MatMul_2:out0",
    "Add_7:in1"
   ],
   [
    "TensorArrayReadV3_1:out0",
    "MatMul_1:in0"
   ],
   [
    "Enter_8:out0",
    "MatMul_1:in1"
   ],
   [
    "Identity_2:out4096",
    "C_15:in0"
   ],
   [
    "C_17:out0",
    "Enter_6:in0"
   ],
   [
    "Switch_4:out1",
    "Identity_4:in0"
   ],
   [
    "C_18:out0",
    "Enter_7:in0"
   ],
   [
    "Identity_2:out4096",
    "C_16:in0"
   ],
   [
    "C_19:out0",
    "Mul_4:in0"
   ],
   [
    "Add_8:out0",
    "Mul_4:in1"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3_1:in1"
   ],
   [
    "Enter_9:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_4:in1"
   ],
   [
    "C_20:out0",
    "Mul_5:in0"
   ],
   [
    "Add_9:out0",
    "Mul_5:in1"
   ],
   [
    "Identity_4:out0",
    "MatMul_2:in0"
   ],
   [
    "MatMul_3:out0",
    "BiasAdd_1:in0"
   ],
   [
    "Enter_10:out0",
    "BiasAdd_1:in1"
   ],
   [
    "Enter_11:out0",
    "MatMul_2:in1"
   ],
   [
    "Identity_2:out4096",
    "C_19:in0"
   ],
   [
    "Enter_12:out0",
    "TensorArrayReadV3_1:in0"
   ],
   [
    "Enter_13:out0",
    "TensorArrayReadV3_1:in2"
   ],
   [
    "C_21:out0",
    "Enter_8:in0"
   ],
   [
    "Add_2:out0",
    "NextIteration_3:in0"
   ],
   [
    "Merge_4:out0",
    "Switch_4:in0"
   ],
   [
    "Identity_2:out4096",
    "C_20:in0"
   ],
   [
    "BiasAdd_2:out0",
    "Add_8:in0"
   ],
   [
    "MatMul_4:out0",
    "Add_8:in1"
   ],
   [
    "C_22:out0",
    "Enter_9:in0"
   ],
   [
    "TensorArrayReadV3_1:out0",
    "MatMul_3:in0"
   ],
   [
    "BiasAdd_3:out0",
    "Add_9:in0"
   ],
   [
    "MatMul_5:out0",
    "Add_9:in1"
   ],
   [
    "Enter_14:out0",
    "MatMul_3:in1"
   ],
   [
    "C_23:out0",
    "Enter_10:in0"
   ],
   [
    "C_24:out0",
    "Enter_11:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_12:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_13:in0"
   ],
   [
    "Identity_4:out0",
    "MatMul_4:in0"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "Enter_15:out0",
    "Merge_4:in0"
   ],
   [
    "NextIteration_4:out0",
    "Merge_4:in1"
   ],
   [
    "MatMul_6:out0",
    "BiasAdd_2:in0"
   ],
   [
    "Identity_4:out0",
    "MatMul_5:in0"
   ],
   [
    "Enter_16:out0",
    "BiasAdd_2:in1"
   ],
   [
    "Enter_17:out0",
    "MatMul_4:in1"
   ],
   [
    "Mul:out0",
    "NextIteration_4:in0"
   ],
   [
    "MatMul_7:out0",
    "BiasAdd_3:in0"
   ],
   [
    "Enter_18:out0",
    "BiasAdd_3:in1"
   ],
   [
    "Enter_19:out0",
    "MatMul_5:in1"
   ],
   [
    "C_25:out0",
    "Enter_14:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "TensorArrayReadV3_1:out0",
    "MatMul_6:in0"
   ],
   [
    "C_26:out0",
    "TensorArrayScatterV3:in1"
   ],
   [
    "ReverseV2:out0",
    "TensorArrayScatterV3:in2"
   ],
   [
    "C_27:out0",
    "Enter_15:in0"
   ],
   [
    "TensorArrayReadV3_1:out0",
    "MatMul_7:in0"
   ],
   [
    "Enter_20:out0",
    "MatMul_6:in1"
   ],
   [
    "C_28:out0",
    "Enter_16:in0"
   ],
   [
    "C_29:out0",
    "Enter_17:in0"
   ],
   [
    "Enter_21:out0",
    "MatMul_7:in1"
   ],
   [
    "C_30:out0",
    "Enter_18:in0"
   ],
   [
    "C_31:out0",
    "Enter_19:in0"
   ],
   [
    "C_32:out0",
    "ReverseV2:in1"
   ],
   [
    "C_33:out0",
    "Enter_20:in0"
   ],
   [
    "C_34:out0",
    "Enter_21:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ReverseV2:in0"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayReadV3:out0"
  ],
  "acu_lys_alias": [
   "reverse",
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reverse:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayReadV3:out0",
    "lstm:out0"
   ]
  ],
  "acu_inter_flow": [
   [
    "reverse:out0",
    "lstm:in0"
   ]
  ],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "VALUE",
     0
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C_34:out0'])[1]"
    ],
    "recurrent_activation": [
     "STRING",
     "VALUE",
     "hard_sigmoid"
    ],
    "return_sequences": [
     "BOOL",
     "VALUE",
     false
    ]
   },
   "reverse": {
    "axis": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_32:out0'])"
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_34:out0'])"
    ],
    "wf": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_33:out0'])"
    ],
    "wc": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_25:out0'])"
    ],
    "wo": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_21:out0'])"
    ],
    "hi": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_31:out0'])"
    ],
    "hf": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_29:out0'])"
    ],
    "hc": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_24:out0'])"
    ],
    "ho": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_18:out0'])"
    ],
    "bi": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_30:out0'])"
    ],
    "bf": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_28:out0'])"
    ],
    "bc": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_23:out0'])"
    ],
    "bo": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_17:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "reducesum",
  "src_ops_alias": [
   "Sum",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Sum:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Sum:in0"
   ]
  ],
  "src_out_tensor": [
   "Sum:out0"
  ],
  "acu_lys_alias": [
   "reducesum"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reducesum:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Sum:out0",
    "reducesum:out0"
   ]
  ],
  "param_map": {
   "reducesum": {
    "axis_list": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"axis_tensor_name\": \"C:out0\"}",
      "def r_reduce_sum_get_axis(self, node, tensor, axis_tensor_name):",
      "    axis = self.tensor_to_numpy(tensor[axis_tensor_name]).tolist()",
      "    if isinstance(axis, list):",
      "        return axis",
      "    else:",
      "        return [axis]",
      ""
     ]
    ],
    "keep_dims": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['Sum'], 'keep_dims')"
    ]
   }
  },
  "blob_map": {
   "reducesum": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "reducesum_output_scalar",
  "src_ops_alias": [
   "Pack",
   "Sum",
   "C"
  ],
  "src_inter_flow": [
   [
    "Sum:out0",
    "Pack:in0"
   ],
   [
    "C:out0",
    "Sum:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Sum:in0"
   ]
  ],
  "src_out_tensor": [
   "Pack:out0"
  ],
  "acu_lys_alias": [
   "reducesum"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reducesum:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Pack:out0",
    "reducesum:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "reducesum": {
    "axis_list": [
     "INTS",
     "VALUE",
     [
      0
     ]
    ],
    "keep_dims": [
     "BOOL",
     "VALUE",
     true
    ]
   }
  },
  "blob_map": {
   "reducesum": {}
  },
  "priority_tip": 0,
  "pre_condition": [
   "__rule_func_additional_args = {\"input_tensor_name\": \"I:out0\"}",
   "def r_reducesum_output_scalar_pre_condition(self, node, tensor, input_tensor_name):",
   "    input_tensor_shape = self.shape_pick(tensor[input_tensor_name])",
   "    if len(input_tensor_shape) == 1:",
   "        print(input_tensor_shape)",
   "        return True",
   "    return False",
   ""
  ]
 },
 {
  "ruler_name": "reduceany",
  "src_ops_alias": [
   "Any",
   "C"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Any:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Any:in0"
   ]
  ],
  "src_out_tensor": [
   "Any:out0"
  ],
  "acu_lys_alias": [
   "reduceany"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "reduceany:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Any:out0",
    "reduceany:out0"
   ]
  ],
  "param_map": {
   "reduceany": {
    "axis_list": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"axis_tensor_name\": \"C:out0\"}",
      "def r_reduce_sum_get_axis(self, node, tensor, axis_tensor_name):",
      "    axis = self.tensor_to_numpy(tensor[axis_tensor_name]).tolist()",
      "    if isinstance(axis, list):",
      "        return axis",
      "    else:",
      "        return [axis]",
      ""
     ]
    ],
    "keep_dims": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['Any'], 'keep_dims')"
    ]
   }
  },
  "blob_map": {
   "reduceany": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "nms_v5",
  "src_ops_alias": [
   "NonMaxSuppressionV5",
   "C",
   "C_1",
   "C_2",
   "C_3"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "NonMaxSuppressionV5:in2"
   ],
   [
    "C_1:out0",
    "NonMaxSuppressionV5:in3"
   ],
   [
    "C_2:out0",
    "NonMaxSuppressionV5:in4"
   ],
   [
    "C_3:out0",
    "NonMaxSuppressionV5:in5"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "NonMaxSuppressionV5:in0"
   ],
   [
    "I_1:out0",
    "NonMaxSuppressionV5:in1"
   ]
  ],
  "src_out_tensor": [
   "NonMaxSuppressionV5:out0",
   "NonMaxSuppressionV5:out1",
   "NonMaxSuppressionV5:out2"
  ],
  "acu_lys_alias": [
   "nms"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "nms:in0"
   ],
   [
    "I_1:out0",
    "nms:in1"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "NonMaxSuppressionV5:out0",
    "nms:out0"
   ],
   [
    "NonMaxSuppressionV5:out1",
    "nms:out1"
   ],
   [
    "NonMaxSuppressionV5:out2",
    "nms:out2"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "nms": {
    "max_output_size": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "iou_threshold": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "score_threshold": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "soft_nms_sigma": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_3:out0'])"
    ],
    "pad_to_max_output_size": [
     "BOOL",
     "CODE",
     "self.attr_pick(node['NonMaxSuppressionV5'],'pad_to_max_output_size')"
    ]
   }
  },
  "blob_map": {
   "nms": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "swish",
  "src_ops_alias": [
   "swish_f32"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "swish_f32:in0"
   ]
  ],
  "src_out_tensor": [
   "swish_f32:out0"
  ],
  "acu_lys_alias": [
   "swish"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "swish:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "swish_f32:out0",
    "swish:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "swish": {}
  },
  "blob_map": {
   "swish": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "swish_tf23",
  "src_ops_alias": [
   "Sigmoid",
   "Mul",
   "IdentityN"
  ],
  "src_inter_flow": [
   [
    "Sigmoid:out0",
    "Mul:in1"
   ],
   [
    "Mul:out0",
    "IdentityN:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Sigmoid:in0"
   ],
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "IdentityN:in1"
   ]
  ],
  "src_out_tensor": [
   "IdentityN:out0"
  ],
  "acu_lys_alias": [
   "swish"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "swish:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "IdentityN:out0",
    "swish:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "swish": {}
  },
  "blob_map": {
   "swish": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "swish_tf28",
  "src_ops_alias": [
   "Identity",
   "IdentityN",
   "Mul",
   "Sigmoid",
   "Mul_1",
   "C"
  ],
  "src_inter_flow": [
   [
    "IdentityN:out0",
    "Identity:in0"
   ],
   [
    "Mul:out0",
    "IdentityN:in0"
   ],
   [
    "Sigmoid:out0",
    "Mul:in1"
   ],
   [
    "Mul_1:out0",
    "Sigmoid:in0"
   ],
   [
    "C:out0",
    "Mul_1:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul_1:in1"
   ],
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "IdentityN:in1"
   ]
  ],
  "src_out_tensor": [
   "Identity:out0"
  ],
  "acu_lys_alias": [
   "swish"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "swish:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Identity:out0",
    "swish:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "swish": {
    "beta": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "swish": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "swish_tf_2_12_0",
  "src_ops_alias": [
   "Identity",
   "IdentityN",
   "Mul",
   "Sigmoid",
   "Mul_1",
   "C"
  ],
  "src_inter_flow": [
   [
    "IdentityN:out0",
    "Identity:in0"
   ],
   [
    "Mul:out0",
    "IdentityN:in0"
   ],
   [
    "Sigmoid:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "IdentityN:in2"
   ],
   [
    "Mul_1:out0",
    "Sigmoid:in0"
   ],
   [
    "C:out0",
    "Mul_1:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Mul_1:in1"
   ],
   [
    "I:out0",
    "Mul:in0"
   ],
   [
    "I:out0",
    "IdentityN:in1"
   ]
  ],
  "src_out_tensor": [
   "Identity:out0"
  ],
  "acu_lys_alias": [
   "swish"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "swish:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Identity:out0",
    "swish:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "swish": {
    "beta": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ]
   }
  },
  "blob_map": {
   "swish": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "frame",
  "src_ops_alias": [
   "Reshape",
   "GatherV2",
   "C",
   "Reshape_1",
   "C_1",
   "C_2",
   "StridedSlice",
   "C_3",
   "PadV2",
   "C_4",
   "C_5",
   "C_6",
   "C_7",
   "C_8"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Reshape:in1"
   ],
   [
    "C_6:out0",
    "StridedSlice:in3"
   ],
   [
    "PadV2:out0",
    "StridedSlice:in0"
   ],
   [
    "StridedSlice:out0",
    "Reshape_1:in0"
   ],
   [
    "C_2:out0",
    "GatherV2:in2"
   ],
   [
    "C_4:out0",
    "StridedSlice:in1"
   ],
   [
    "C_1:out0",
    "GatherV2:in1"
   ],
   [
    "C_3:out0",
    "Reshape_1:in1"
   ],
   [
    "Reshape_1:out0",
    "GatherV2:in0"
   ],
   [
    "C_7:out0",
    "PadV2:in1"
   ],
   [
    "C_5:out0",
    "StridedSlice:in2"
   ],
   [
    "C_8:out0",
    "PadV2:in2"
   ],
   [
    "GatherV2:out0",
    "Reshape:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "PadV2:in0"
   ]
  ],
  "src_out_tensor": [
   "Reshape:out0"
  ],
  "acu_lys_alias": [
   "signalframe"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "signalframe:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Reshape:out0",
    "signalframe:out0"
   ]
  ],
  "param_map": {
   "signalframe": {
    "frame_length": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"paddings_tensor_name\": \"C_7:out0\", \"output_tensor_name\": \"Reshape:out0\"}",
      "def r_signal_frame_get_frame_len(self, node, tensor, paddings_tensor_name, output_tensor_name):",
      "    axis = None",
      "    paddings = self.tensor_to_numpy(tensor[paddings_tensor_name])",
      "    for i, pad in enumerate(paddings):",
      "        if (pad != [0, 0]).any():",
      "            if axis is None:",
      "                axis = i",
      "            else:",
      "                raise ValueError('Cannot determine axis')",
      "    if axis is None:",
      "        raise ValueError('Cannot determine axis')",
      "",
      "    output_tensor_shape = self.shape_pick(tensor[output_tensor_name])",
      "    return output_tensor_shape[axis + 1]",
      ""
     ]
    ],
    "frame_step": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"paddings_tensor_name\": \"C_7:out0\", \"input_tensor_name\": \"I:out0\", \"output_tensor_name\": \"Reshape:out0\"}",
      "def r_signal_frame_get_frame_step(self, node, tensor, paddings_tensor_name, input_tensor_name,",
      "                            output_tensor_name):",
      "    axis = None",
      "    pad_value = None",
      "    paddings = self.tensor_to_numpy(tensor[paddings_tensor_name])",
      "    for i, pad in enumerate(paddings):",
      "        if (pad != [0, 0]).any():",
      "            if axis is None:",
      "                axis = i",
      "                pad_value = pad[1]",
      "            else:",
      "                raise ValueError('Cannot determine axis')",
      "    if axis is None or pad_value is None:",
      "        raise ValueError('Cannot determine axis')",
      "",
      "    input_tensor_shape = self.shape_pick(tensor[input_tensor_name])",
      "    output_tensor_shape = self.shape_pick(tensor[output_tensor_name])",
      "    output_len = output_tensor_shape[axis]",
      "    frame_length = output_tensor_shape[axis + 1]",
      "    if (input_tensor_shape[axis] + pad_value - frame_length) == 0:",
      "        frame_step = 1",
      "    else:",
      "        frame_step = (input_tensor_shape[axis] + pad_value - frame_length) / (output_len - 1)",
      "",
      "    return int(frame_step)",
      ""
     ]
    ],
    "pad_end": [
     "BOOL",
     "VALUE",
     true
    ],
    "pad_value": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_8:out0'])"
    ],
    "axis": [
     "ORIGIN",
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"paddings_tensor_name\": \"C_7:out0\"}",
      "def r_signal_frame_get_axis(self, node, tensor, paddings_tensor_name):",
      "    axis = None",
      "    paddings = self.tensor_to_numpy(tensor[paddings_tensor_name])",
      "    for i, pad in enumerate(paddings):",
      "        if (pad != [0, 0]).any():",
      "            if axis is None:",
      "                axis = i",
      "            else:",
      "                raise ValueError('Cannot determine axis')",
      "    if axis is None:",
      "        raise ValueError('Cannot determine axis')",
      "",
      "    return axis",
      ""
     ]
    ]
   }
  },
  "blob_map": {
   "signalframe": {}
  },
  "acu_inter_flow": [],
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "mrcnn-proposal",
  "src_ops_alias": [
   "Pack",
   "Pad",
   "GatherV2",
   "C",
   "StridedSlice",
   "NonMaxSuppressionV3",
   "C_1",
   "Pack_1",
   "C_2",
   "C_3",
   "C_4",
   "StridedSlice_1",
   "C_5",
   "C_6",
   "C_7",
   "ConcatV2",
   "Pack_2",
   "C_8",
   "C_9",
   "C_10",
   "Maximum",
   "Maximum_1",
   "Maximum_2",
   "Maximum_3",
   "C_11",
   "GatherV2_1",
   "Minimum",
   "C_12",
   "Minimum_1",
   "C_13",
   "Minimum_2",
   "Minimum_3",
   "StridedSlice_2",
   "StridedSlice_3",
   "C_14",
   "Split",
   "C_15",
   "C_16",
   "StridedSlice_4",
   "C_17",
   "C_18",
   "C_19",
   "TopKV2",
   "C_20",
   "C_21",
   "C_22",
   "C_23",
   "StridedSlice_5",
   "C_24",
   "C_25",
   "C_26",
   "C_27",
   "Pack_3",
   "C_28",
   "C_29",
   "C_30",
   "Pack_4",
   "Sub",
   "Sub_1",
   "Add",
   "Add_1",
   "Add_2",
   "Mul",
   "Add_3",
   "Mul_1",
   "Mul_2",
   "Mul_3",
   "Add_4",
   "Mul_4",
   "C_31",
   "Add_5",
   "Mul_5",
   "C_32",
   "Sub_2",
   "Exp",
   "Sub_3",
   "Exp_1",
   "StridedSlice_6",
   "Mul_6",
   "StridedSlice_7",
   "StridedSlice_8",
   "Mul_7",
   "StridedSlice_9",
   "StridedSlice_10",
   "StridedSlice_11",
   "StridedSlice_12",
   "StridedSlice_13",
   "StridedSlice_14",
   "StridedSlice_15",
   "StridedSlice_16",
   "C_33",
   "C_34",
   "C_35",
   "C_36",
   "StridedSlice_17",
   "C_37",
   "C_38",
   "C_39",
   "C_40",
   "C_41",
   "C_42",
   "C_43",
   "C_44",
   "C_45",
   "C_46",
   "C_47",
   "C_48",
   "C_49",
   "C_50",
   "C_51",
   "C_52",
   "C_53",
   "C_54",
   "C_55",
   "C_56",
   "C_57",
   "C_58",
   "C_59",
   "C_60",
   "C_61",
   "C_62",
   "C_63",
   "C_64",
   "Pack_5",
   "C_65",
   "C_66",
   "C_67",
   "Pack_6",
   "C_68",
   "C_69",
   "C_70",
   "GatherV2_2",
   "GatherV2_3",
   "StridedSlice_18",
   "StridedSlice_19",
   "C_71",
   "StridedSlice_20",
   "StridedSlice_21",
   "C_72",
   "C_73",
   "C_74",
   "C_75",
   "C_76",
   "C_77",
   "C_78",
   "Mul_8",
   "C_79",
   "C_80",
   "C_81",
   "C_82",
   "C_83",
   "C_84",
   "C_85"
  ],
  "src_inter_flow": [
   [
    "Pad:out0",
    "Pack:in0"
   ],
   [
    "GatherV2:out0",
    "Pad:in0"
   ],
   [
    "C:out0",
    "Pad:in1"
   ],
   [
    "StridedSlice:out0",
    "GatherV2:in0"
   ],
   [
    "NonMaxSuppressionV3:out0",
    "GatherV2:in1"
   ],
   [
    "C_1:out0",
    "GatherV2:in2"
   ],
   [
    "Pack_1:out0",
    "StridedSlice:in0"
   ],
   [
    "C_2:out0",
    "StridedSlice:in1"
   ],
   [
    "C_3:out0",
    "StridedSlice:in2"
   ],
   [
    "C_4:out0",
    "StridedSlice:in3"
   ],
   [
    "StridedSlice:out0",
    "NonMaxSuppressionV3:in0"
   ],
   [
    "StridedSlice_1:out0",
    "NonMaxSuppressionV3:in1"
   ],
   [
    "C_5:out0",
    "NonMaxSuppressionV3:in2"
   ],
   [
    "C_6:out0",
    "NonMaxSuppressionV3:in3"
   ],
   [
    "C_7:out0",
    "NonMaxSuppressionV3:in4"
   ],
   [
    "ConcatV2:out0",
    "Pack_1:in0"
   ],
   [
    "Pack_2:out0",
    "StridedSlice_1:in0"
   ],
   [
    "C_8:out0",
    "StridedSlice_1:in1"
   ],
   [
    "C_9:out0",
    "StridedSlice_1:in2"
   ],
   [
    "C_10:out0",
    "StridedSlice_1:in3"
   ],
   [
    "Maximum:out0",
    "ConcatV2:in0"
   ],
   [
    "Maximum_1:out0",
    "ConcatV2:in1"
   ],
   [
    "Maximum_2:out0",
    "ConcatV2:in2"
   ],
   [
    "Maximum_3:out0",
    "ConcatV2:in3"
   ],
   [
    "C_11:out0",
    "ConcatV2:in4"
   ],
   [
    "GatherV2_1:out0",
    "Pack_2:in0"
   ],
   [
    "Minimum:out0",
    "Maximum:in0"
   ],
   [
    "C_12:out0",
    "Maximum:in1"
   ],
   [
    "Minimum_1:out0",
    "Maximum_1:in0"
   ],
   [
    "C_13:out0",
    "Maximum_1:in1"
   ],
   [
    "C_12:out0",
    "Maximum_2:in1"
   ],
   [
    "Minimum_2:out0",
    "Maximum_2:in0"
   ],
   [
    "C_13:out0",
    "Maximum_3:in1"
   ],
   [
    "Minimum_3:out0",
    "Maximum_3:in0"
   ],
   [
    "StridedSlice_2:out0",
    "GatherV2_1:in0"
   ],
   [
    "StridedSlice_3:out0",
    "GatherV2_1:in1"
   ],
   [
    "C_14:out0",
    "GatherV2_1:in2"
   ],
   [
    "Split:out0",
    "Minimum:in0"
   ],
   [
    "C_15:out0",
    "Minimum:in1"
   ],
   [
    "Split:out1",
    "Minimum_1:in0"
   ],
   [
    "C_16:out0",
    "Minimum_1:in1"
   ],
   [
    "Split:out2",
    "Minimum_2:in0"
   ],
   [
    "C_15:out0",
    "Minimum_2:in1"
   ],
   [
    "Split:out3",
    "Minimum_3:in0"
   ],
   [
    "C_16:out0",
    "Minimum_3:in1"
   ],
   [
    "StridedSlice_4:out0",
    "StridedSlice_2:in0"
   ],
   [
    "C_17:out0",
    "StridedSlice_2:in1"
   ],
   [
    "C_18:out0",
    "StridedSlice_2:in2"
   ],
   [
    "C_19:out0",
    "StridedSlice_2:in3"
   ],
   [
    "TopKV2:out1",
    "StridedSlice_3:in0"
   ],
   [
    "C_20:out0",
    "StridedSlice_3:in1"
   ],
   [
    "C_21:out0",
    "StridedSlice_3:in2"
   ],
   [
    "C_22:out0",
    "StridedSlice_3:in3"
   ],
   [
    "C_23:out0",
    "Split:in0"
   ],
   [
    "StridedSlice_5:out0",
    "Split:in1"
   ],
   [
    "C_24:out0",
    "StridedSlice_4:in1"
   ],
   [
    "C_25:out0",
    "StridedSlice_4:in2"
   ],
   [
    "C_26:out0",
    "StridedSlice_4:in3"
   ],
   [
    "StridedSlice_4:out0",
    "TopKV2:in0"
   ],
   [
    "C_27:out0",
    "TopKV2:in1"
   ],
   [
    "Pack_3:out0",
    "StridedSlice_5:in0"
   ],
   [
    "C_28:out0",
    "StridedSlice_5:in1"
   ],
   [
    "C_29:out0",
    "StridedSlice_5:in2"
   ],
   [
    "C_30:out0",
    "StridedSlice_5:in3"
   ],
   [
    "Pack_4:out0",
    "Pack_3:in0"
   ],
   [
    "Sub:out0",
    "Pack_4:in0"
   ],
   [
    "Sub_1:out0",
    "Pack_4:in1"
   ],
   [
    "Add:out0",
    "Pack_4:in2"
   ],
   [
    "Add_1:out0",
    "Pack_4:in3"
   ],
   [
    "Add_2:out0",
    "Sub:in0"
   ],
   [
    "Mul:out0",
    "Sub:in1"
   ],
   [
    "Add_3:out0",
    "Sub_1:in0"
   ],
   [
    "Mul_1:out0",
    "Sub_1:in1"
   ],
   [
    "Sub:out0",
    "Add:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "Sub_1:out0",
    "Add_1:in0"
   ],
   [
    "Mul_3:out0",
    "Add_1:in1"
   ],
   [
    "Add_4:out0",
    "Add_2:in0"
   ],
   [
    "Mul_4:out0",
    "Add_2:in1"
   ],
   [
    "Mul_2:out0",
    "Mul:in1"
   ],
   [
    "C_31:out0",
    "Mul:in0"
   ],
   [
    "Add_5:out0",
    "Add_3:in0"
   ],
   [
    "Mul_5:out0",
    "Add_3:in1"
   ],
   [
    "Mul_3:out0",
    "Mul_1:in1"
   ],
   [
    "C_32:out0",
    "Mul_1:in0"
   ],
   [
    "Sub_2:out0",
    "Mul_2:in0"
   ],
   [
    "Exp:out0",
    "Mul_2:in1"
   ],
   [
    "Sub_3:out0",
    "Mul_3:in0"
   ],
   [
    "Exp_1:out0",
    "Mul_3:in1"
   ],
   [
    "StridedSlice_6:out0",
    "Add_4:in0"
   ],
   [
    "Mul_6:out0",
    "Add_4:in1"
   ],
   [
    "Sub_2:out0",
    "Mul_4:in1"
   ],
   [
    "StridedSlice_7:out0",
    "Mul_4:in0"
   ],
   [
    "StridedSlice_8:out0",
    "Add_5:in0"
   ],
   [
    "Mul_7:out0",
    "Add_5:in1"
   ],
   [
    "Sub_3:out0",
    "Mul_5:in1"
   ],
   [
    "StridedSlice_9:out0",
    "Mul_5:in0"
   ],
   [
    "StridedSlice_10:out0",
    "Sub_2:in0"
   ],
   [
    "StridedSlice_11:out0",
    "Sub_2:in1"
   ],
   [
    "StridedSlice_12:out0",
    "Exp:in0"
   ],
   [
    "StridedSlice_13:out0",
    "Sub_3:in0"
   ],
   [
    "StridedSlice_14:out0",
    "Sub_3:in1"
   ],
   [
    "StridedSlice_15:out0",
    "Exp_1:in0"
   ],
   [
    "Sub_2:out0",
    "Mul_6:in1"
   ],
   [
    "StridedSlice_16:out0",
    "StridedSlice_6:in0"
   ],
   [
    "C_33:out0",
    "StridedSlice_6:in1"
   ],
   [
    "C_34:out0",
    "StridedSlice_6:in2"
   ],
   [
    "C_35:out0",
    "StridedSlice_6:in3"
   ],
   [
    "C_36:out0",
    "Mul_6:in0"
   ],
   [
    "StridedSlice_17:out0",
    "StridedSlice_7:in0"
   ],
   [
    "Sub_3:out0",
    "Mul_7:in1"
   ],
   [
    "C_37:out0",
    "StridedSlice_7:in1"
   ],
   [
    "C_38:out0",
    "StridedSlice_7:in2"
   ],
   [
    "C_39:out0",
    "StridedSlice_7:in3"
   ],
   [
    "StridedSlice_16:out0",
    "StridedSlice_8:in0"
   ],
   [
    "C_40:out0",
    "StridedSlice_8:in1"
   ],
   [
    "C_41:out0",
    "StridedSlice_8:in2"
   ],
   [
    "C_42:out0",
    "StridedSlice_8:in3"
   ],
   [
    "C_43:out0",
    "Mul_7:in0"
   ],
   [
    "StridedSlice_17:out0",
    "StridedSlice_9:in0"
   ],
   [
    "StridedSlice_16:out0",
    "StridedSlice_10:in0"
   ],
   [
    "C_44:out0",
    "StridedSlice_9:in1"
   ],
   [
    "C_45:out0",
    "StridedSlice_9:in2"
   ],
   [
    "C_46:out0",
    "StridedSlice_9:in3"
   ],
   [
    "StridedSlice_16:out0",
    "StridedSlice_11:in0"
   ],
   [
    "C_47:out0",
    "StridedSlice_10:in1"
   ],
   [
    "C_48:out0",
    "StridedSlice_10:in2"
   ],
   [
    "C_49:out0",
    "StridedSlice_10:in3"
   ],
   [
    "StridedSlice_17:out0",
    "StridedSlice_12:in0"
   ],
   [
    "C_50:out0",
    "StridedSlice_11:in1"
   ],
   [
    "StridedSlice_16:out0",
    "StridedSlice_13:in0"
   ],
   [
    "C_51:out0",
    "StridedSlice_11:in2"
   ],
   [
    "C_52:out0",
    "StridedSlice_11:in3"
   ],
   [
    "StridedSlice_16:out0",
    "StridedSlice_14:in0"
   ],
   [
    "C_53:out0",
    "StridedSlice_12:in1"
   ],
   [
    "C_54:out0",
    "StridedSlice_12:in2"
   ],
   [
    "C_55:out0",
    "StridedSlice_12:in3"
   ],
   [
    "C_56:out0",
    "StridedSlice_13:in1"
   ],
   [
    "StridedSlice_17:out0",
    "StridedSlice_15:in0"
   ],
   [
    "C_57:out0",
    "StridedSlice_13:in2"
   ],
   [
    "C_58:out0",
    "StridedSlice_13:in3"
   ],
   [
    "C_59:out0",
    "StridedSlice_14:in1"
   ],
   [
    "C_60:out0",
    "StridedSlice_14:in2"
   ],
   [
    "C_61:out0",
    "StridedSlice_14:in3"
   ],
   [
    "C_62:out0",
    "StridedSlice_15:in1"
   ],
   [
    "C_63:out0",
    "StridedSlice_15:in2"
   ],
   [
    "C_64:out0",
    "StridedSlice_15:in3"
   ],
   [
    "Pack_5:out0",
    "StridedSlice_16:in0"
   ],
   [
    "C_65:out0",
    "StridedSlice_16:in1"
   ],
   [
    "C_66:out0",
    "StridedSlice_16:in2"
   ],
   [
    "C_67:out0",
    "StridedSlice_16:in3"
   ],
   [
    "Pack_6:out0",
    "StridedSlice_17:in0"
   ],
   [
    "C_68:out0",
    "StridedSlice_17:in1"
   ],
   [
    "C_69:out0",
    "StridedSlice_17:in2"
   ],
   [
    "C_70:out0",
    "StridedSlice_17:in3"
   ],
   [
    "GatherV2_2:out0",
    "Pack_5:in0"
   ],
   [
    "TopKV2:out1",
    "StridedSlice_19:in0"
   ],
   [
    "GatherV2_3:out0",
    "Pack_6:in0"
   ],
   [
    "TopKV2:out1",
    "StridedSlice_21:in0"
   ],
   [
    "StridedSlice_18:out0",
    "GatherV2_2:in0"
   ],
   [
    "StridedSlice_19:out0",
    "GatherV2_2:in1"
   ],
   [
    "C_71:out0",
    "GatherV2_2:in2"
   ],
   [
    "StridedSlice_20:out0",
    "GatherV2_3:in0"
   ],
   [
    "StridedSlice_21:out0",
    "GatherV2_3:in1"
   ],
   [
    "C_72:out0",
    "GatherV2_3:in2"
   ],
   [
    "C_73:out0",
    "StridedSlice_18:in1"
   ],
   [
    "C_74:out0",
    "StridedSlice_18:in2"
   ],
   [
    "C_75:out0",
    "StridedSlice_18:in3"
   ],
   [
    "C_76:out0",
    "StridedSlice_19:in1"
   ],
   [
    "C_77:out0",
    "StridedSlice_19:in2"
   ],
   [
    "C_78:out0",
    "StridedSlice_19:in3"
   ],
   [
    "Mul_8:out0",
    "StridedSlice_20:in0"
   ],
   [
    "C_79:out0",
    "StridedSlice_20:in1"
   ],
   [
    "C_80:out0",
    "StridedSlice_20:in2"
   ],
   [
    "C_81:out0",
    "StridedSlice_20:in3"
   ],
   [
    "C_82:out0",
    "StridedSlice_21:in1"
   ],
   [
    "C_83:out0",
    "StridedSlice_21:in2"
   ],
   [
    "C_84:out0",
    "StridedSlice_21:in3"
   ],
   [
    "C_85:out0",
    "Mul_8:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "StridedSlice_4:in0"
   ],
   [
    "I_1:out0",
    "Mul_8:in0"
   ],
   [
    "I_2:out0",
    "StridedSlice_18:in0"
   ]
  ],
  "src_out_tensor": [
   "Pack:out0"
  ],
  "acu_lys_alias": [
   "mrcnn_proposal"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "mrcnn_proposal:in0"
   ],
   [
    "I_1:out0",
    "mrcnn_proposal:in1"
   ],
   [
    "I_2:out0",
    "mrcnn_proposal:in2"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Pack:out0",
    "mrcnn_proposal:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "mrcnn_proposal": {}
  },
  "blob_map": {
   "mrcnn_proposal": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "mrcnn-detection",
  "src_ops_alias": [
   "Reshape",
   "Pack",
   "C",
   "Pad",
   "ConcatV2",
   "C_1",
   "GatherV2",
   "StridedSlice",
   "StridedSlice_1",
   "C_2",
   "ConcatV2_1",
   "GatherV2_1",
   "C_3",
   "Cast",
   "C_4",
   "C_5",
   "C_6",
   "GatherV2_2",
   "C_7",
   "C_8",
   "C_9",
   "Maximum",
   "Maximum_1",
   "Maximum_2",
   "Maximum_3",
   "C_10",
   "StridedSlice_2",
   "TopKV2",
   "C_11",
   "GatherV2_3",
   "GatherNd",
   "C_12",
   "Minimum",
   "Split",
   "Minimum_1",
   "Minimum_2",
   "Minimum_3",
   "SparseToDense",
   "C_13",
   "C_14",
   "C_15",
   "GatherV2_4",
   "C_16",
   "ArgMax",
   "C_17",
   "StridedSlice_3",
   "Pack_1",
   "Split_1",
   "C_18",
   "StridedSlice_4",
   "DenseToDenseSetOperation",
   "C_19",
   "C_20",
   "C_21",
   "C_22",
   "C_23",
   "C_24",
   "C_25",
   "C_26",
   "Pack_2",
   "RealDiv",
   "C_27",
   "C_28",
   "C_29",
   "ExpandDims",
   "ExpandDims_1",
   "Sub",
   "Sub_1",
   "Add",
   "Add_1",
   "Sub_2",
   "Sub_3",
   "StridedSlice_5",
   "C_30",
   "GatherV2_5",
   "C_31",
   "Add_2",
   "Mul",
   "Add_3",
   "Mul_1",
   "Mul_2",
   "Mul_3",
   "StridedSlice_6",
   "C_32",
   "ConcatV2_2",
   "C_33",
   "Where",
   "C_34",
   "C_35",
   "C_36",
   "Reshape_1",
   "StridedSlice_7",
   "C_37",
   "Add_4",
   "Mul_4",
   "C_38",
   "Add_5",
   "Mul_5",
   "C_39",
   "Sub_4",
   "Exp",
   "Sub_5",
   "Exp_1",
   "C_40",
   "C_41",
   "C_42",
   "Split_2",
   "C_43",
   "Greater",
   "TensorArrayGatherV3",
   "C_44",
   "Where_1",
   "C_45",
   "C_46",
   "C_47",
   "StridedSlice_8",
   "Mul_6",
   "StridedSlice_9",
   "StridedSlice_10",
   "Mul_7",
   "StridedSlice_11",
   "StridedSlice_12",
   "StridedSlice_13",
   "StridedSlice_14",
   "StridedSlice_15",
   "StridedSlice_16",
   "StridedSlice_17",
   "C_48",
   "StridedSlice_18",
   "C_49",
   "TensorArrayV3",
   "Range",
   "Exit",
   "Greater_1",
   "StridedSlice_19",
   "C_50",
   "C_51",
   "C_52",
   "C_53",
   "Mul_8",
   "C_54",
   "C_55",
   "C_56",
   "C_57",
   "C_58",
   "C_59",
   "C_60",
   "C_61",
   "C_62",
   "C_63",
   "C_64",
   "C_65",
   "C_66",
   "C_67",
   "C_68",
   "C_69",
   "C_70",
   "C_71",
   "C_72",
   "C_73",
   "C_74",
   "C_75",
   "C_76",
   "C_77",
   "C_78",
   "C_79",
   "C_80",
   "C_81",
   "StridedSlice_20",
   "C_82",
   "C_83",
   "C_84",
   "C_85",
   "C_86",
   "TensorArraySizeV3",
   "C_87",
   "Switch",
   "C_88",
   "C_89",
   "C_90",
   "C_91",
   "GatherNd_1",
   "C_92",
   "StridedSlice_21",
   "C_93",
   "C_94",
   "C_95",
   "Merge",
   "LoopCond",
   "StridedSlice_22",
   "C_96",
   "C_97",
   "C_98",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "C_99",
   "C_100",
   "C_101",
   "TensorArrayWriteV3",
   "Less",
   "Less_1",
   "Enter_1",
   "Identity",
   "PadV2",
   "Identity_1",
   "Merge_1",
   "Enter_2",
   "Merge_2",
   "Switch_1",
   "GatherV2_6",
   "Pack_3",
   "C_102",
   "Enter_3",
   "NextIteration_1",
   "Enter_4",
   "NextIteration_2",
   "Enter_5",
   "GatherV2_7",
   "C_103",
   "Pack_4",
   "Identity_2",
   "C_104",
   "Add_6",
   "C_105",
   "Add_7",
   "StridedSlice_23",
   "NonMaxSuppressionV3",
   "C_106",
   "C_107",
   "Sub_6",
   "Switch_2",
   "C_108",
   "C_109",
   "Where_2",
   "C_110",
   "C_111",
   "C_112",
   "GatherV2_8",
   "GatherV2_9",
   "C_113",
   "C_114",
   "C_115",
   "C_116",
   "StridedSlice_24",
   "Equal",
   "Enter_6",
   "C_117",
   "Enter_7",
   "C_118",
   "Shape",
   "C_119",
   "C_120",
   "C_121",
   "Enter_8",
   "TensorArrayReadV3",
   "GatherV2_10",
   "GatherV2_11",
   "GatherV2_12",
   "Enter_9",
   "Enter_10",
   "C_122",
   "C_123",
   "C_124",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "C_125",
   "Unique"
  ],
  "src_inter_flow": [
   [
    "Pack:out0",
    "Reshape:in0"
   ],
   [
    "C:out0",
    "Reshape:in1"
   ],
   [
    "Pad:out0",
    "Pack:in0"
   ],
   [
    "ConcatV2:out0",
    "Pad:in0"
   ],
   [
    "C_1:out0",
    "Pad:in1"
   ],
   [
    "GatherV2:out0",
    "ConcatV2:in0"
   ],
   [
    "StridedSlice:out0",
    "ConcatV2:in1"
   ],
   [
    "StridedSlice_1:out0",
    "ConcatV2:in2"
   ],
   [
    "C_2:out0",
    "ConcatV2:in3"
   ],
   [
    "ConcatV2_1:out0",
    "GatherV2:in0"
   ],
   [
    "GatherV2_1:out0",
    "GatherV2:in1"
   ],
   [
    "C_3:out0",
    "GatherV2:in2"
   ],
   [
    "Cast:out0",
    "StridedSlice:in0"
   ],
   [
    "C_4:out0",
    "StridedSlice:in1"
   ],
   [
    "C_5:out0",
    "StridedSlice:in2"
   ],
   [
    "C_6:out0",
    "StridedSlice:in3"
   ],
   [
    "GatherV2_2:out0",
    "StridedSlice_1:in0"
   ],
   [
    "C_7:out0",
    "StridedSlice_1:in1"
   ],
   [
    "C_8:out0",
    "StridedSlice_1:in2"
   ],
   [
    "C_9:out0",
    "StridedSlice_1:in3"
   ],
   [
    "Maximum:out0",
    "ConcatV2_1:in0"
   ],
   [
    "Maximum_1:out0",
    "ConcatV2_1:in1"
   ],
   [
    "Maximum_2:out0",
    "ConcatV2_1:in2"
   ],
   [
    "Maximum_3:out0",
    "ConcatV2_1:in3"
   ],
   [
    "C_10:out0",
    "ConcatV2_1:in4"
   ],
   [
    "StridedSlice_2:out0",
    "GatherV2_1:in0"
   ],
   [
    "TopKV2:out1",
    "GatherV2_1:in1"
   ],
   [
    "C_11:out0",
    "GatherV2_1:in2"
   ],
   [
    "GatherV2_3:out0",
    "Cast:in0"
   ],
   [
    "GatherV2_1:out0",
    "GatherV2_2:in1"
   ],
   [
    "GatherNd:out0",
    "GatherV2_2:in0"
   ],
   [
    "C_12:out0",
    "GatherV2_2:in2"
   ],
   [
    "Minimum:out0",
    "Maximum:in0"
   ],
   [
    "Split:out0",
    "Maximum:in1"
   ],
   [
    "Split:out1",
    "Maximum_1:in1"
   ],
   [
    "Minimum_1:out0",
    "Maximum_1:in0"
   ],
   [
    "Split:out0",
    "Maximum_2:in1"
   ],
   [
    "Minimum_2:out0",
    "Maximum_2:in0"
   ],
   [
    "Split:out1",
    "Maximum_3:in1"
   ],
   [
    "Minimum_3:out0",
    "Maximum_3:in0"
   ],
   [
    "SparseToDense:out0",
    "StridedSlice_2:in0"
   ],
   [
    "C_13:out0",
    "StridedSlice_2:in1"
   ],
   [
    "C_14:out0",
    "StridedSlice_2:in2"
   ],
   [
    "C_15:out0",
    "StridedSlice_2:in3"
   ],
   [
    "GatherV2_1:out0",
    "GatherV2_3:in1"
   ],
   [
    "GatherV2_4:out0",
    "TopKV2:in0"
   ],
   [
    "C_16:out0",
    "TopKV2:in1"
   ],
   [
    "ArgMax:out0",
    "GatherV2_3:in0"
   ],
   [
    "C_17:out0",
    "GatherV2_3:in2"
   ],
   [
    "StridedSlice_3:out0",
    "GatherNd:in0"
   ],
   [
    "Pack_1:out0",
    "GatherNd:in1"
   ],
   [
    "Split:out2",
    "Minimum:in1"
   ],
   [
    "Split_1:out0",
    "Minimum:in0"
   ],
   [
    "Split:out3",
    "Minimum_1:in1"
   ],
   [
    "C_18:out0",
    "Split:in0"
   ],
   [
    "StridedSlice_4:out0",
    "Split:in1"
   ],
   [
    "Split:out2",
    "Minimum_2:in1"
   ],
   [
    "Split_1:out1",
    "Minimum_1:in0"
   ],
   [
    "Split:out3",
    "Minimum_3:in1"
   ],
   [
    "Split_1:out2",
    "Minimum_2:in0"
   ],
   [
    "Split_1:out3",
    "Minimum_3:in0"
   ],
   [
    "DenseToDenseSetOperation:out1",
    "SparseToDense:in2"
   ],
   [
    "DenseToDenseSetOperation:out2",
    "SparseToDense:in1"
   ],
   [
    "DenseToDenseSetOperation:out0",
    "SparseToDense:in0"
   ],
   [
    "C_19:out0",
    "SparseToDense:in3"
   ],
   [
    "StridedSlice_2:out0",
    "GatherV2_4:in1"
   ],
   [
    "GatherNd:out0",
    "GatherV2_4:in0"
   ],
   [
    "C_20:out0",
    "GatherV2_4:in2"
   ],
   [
    "StridedSlice_3:out0",
    "ArgMax:in0"
   ],
   [
    "C_21:out0",
    "ArgMax:in1"
   ],
   [
    "ArgMax:out0",
    "Pack_1:in1"
   ],
   [
    "C_22:out0",
    "StridedSlice_3:in1"
   ],
   [
    "C_23:out0",
    "StridedSlice_3:in2"
   ],
   [
    "C_24:out0",
    "StridedSlice_3:in3"
   ],
   [
    "C_25:out0",
    "Pack_1:in0"
   ],
   [
    "C_26:out0",
    "Split_1:in0"
   ],
   [
    "Pack_2:out0",
    "Split_1:in1"
   ],
   [
    "RealDiv:out0",
    "StridedSlice_4:in0"
   ],
   [
    "C_27:out0",
    "StridedSlice_4:in1"
   ],
   [
    "C_28:out0",
    "StridedSlice_4:in2"
   ],
   [
    "C_29:out0",
    "StridedSlice_4:in3"
   ],
   [
    "ExpandDims:out0",
    "DenseToDenseSetOperation:in0"
   ],
   [
    "ExpandDims_1:out0",
    "DenseToDenseSetOperation:in1"
   ],
   [
    "Sub:out0",
    "Pack_2:in0"
   ],
   [
    "Sub_1:out0",
    "Pack_2:in1"
   ],
   [
    "Add:out0",
    "Pack_2:in2"
   ],
   [
    "Add_1:out0",
    "Pack_2:in3"
   ],
   [
    "Sub_2:out0",
    "RealDiv:in0"
   ],
   [
    "Sub_3:out0",
    "RealDiv:in1"
   ],
   [
    "StridedSlice_5:out0",
    "ExpandDims:in0"
   ],
   [
    "C_30:out0",
    "ExpandDims:in1"
   ],
   [
    "GatherV2_5:out0",
    "ExpandDims_1:in0"
   ],
   [
    "C_31:out0",
    "ExpandDims_1:in1"
   ],
   [
    "Add_2:out0",
    "Sub:in0"
   ],
   [
    "Mul:out0",
    "Sub:in1"
   ],
   [
    "Sub:out0",
    "Add:in0"
   ],
   [
    "Add_3:out0",
    "Sub_1:in0"
   ],
   [
    "Mul_1:out0",
    "Sub_1:in1"
   ],
   [
    "Sub_1:out0",
    "Add_1:in0"
   ],
   [
    "Mul_2:out0",
    "Add:in1"
   ],
   [
    "Mul_3:out0",
    "Add_1:in1"
   ],
   [
    "StridedSlice_6:out0",
    "Sub_2:in0"
   ],
   [
    "C_32:out0",
    "Sub_2:in1"
   ],
   [
    "ConcatV2_2:out0",
    "Sub_3:in0"
   ],
   [
    "C_33:out0",
    "Sub_3:in1"
   ],
   [
    "Where:out0",
    "StridedSlice_5:in0"
   ],
   [
    "C_34:out0",
    "StridedSlice_5:in1"
   ],
   [
    "C_35:out0",
    "StridedSlice_5:in2"
   ],
   [
    "C_36:out0",
    "StridedSlice_5:in3"
   ],
   [
    "Reshape_1:out0",
    "GatherV2_5:in0"
   ],
   [
    "StridedSlice_7:out0",
    "GatherV2_5:in1"
   ],
   [
    "C_37:out0",
    "GatherV2_5:in2"
   ],
   [
    "Mul_2:out0",
    "Mul:in1"
   ],
   [
    "Add_4:out0",
    "Add_2:in0"
   ],
   [
    "Mul_4:out0",
    "Add_2:in1"
   ],
   [
    "C_38:out0",
    "Mul:in0"
   ],
   [
    "Mul_3:out0",
    "Mul_1:in1"
   ],
   [
    "Add_5:out0",
    "Add_3:in0"
   ],
   [
    "Mul_5:out0",
    "Add_3:in1"
   ],
   [
    "C_39:out0",
    "Mul_1:in0"
   ],
   [
    "Sub_4:out0",
    "Mul_2:in0"
   ],
   [
    "Exp:out0",
    "Mul_2:in1"
   ],
   [
    "Sub_5:out0",
    "Mul_3:in0"
   ],
   [
    "Exp_1:out0",
    "Mul_3:in1"
   ],
   [
    "C_40:out0",
    "StridedSlice_6:in1"
   ],
   [
    "C_41:out0",
    "StridedSlice_6:in2"
   ],
   [
    "C_42:out0",
    "StridedSlice_6:in3"
   ],
   [
    "Split_2:out0",
    "ConcatV2_2:in2"
   ],
   [
    "Split_2:out1",
    "ConcatV2_2:in3"
   ],
   [
    "Split_2:out0",
    "ConcatV2_2:in0"
   ],
   [
    "Split_2:out1",
    "ConcatV2_2:in1"
   ],
   [
    "C_43:out0",
    "ConcatV2_2:in4"
   ],
   [
    "Greater:out0",
    "Where:in0"
   ],
   [
    "TensorArrayGatherV3:out0",
    "Reshape_1:in0"
   ],
   [
    "C_44:out0",
    "Reshape_1:in1"
   ],
   [
    "Where_1:out0",
    "StridedSlice_7:in0"
   ],
   [
    "C_45:out0",
    "StridedSlice_7:in1"
   ],
   [
    "C_46:out0",
    "StridedSlice_7:in2"
   ],
   [
    "C_47:out0",
    "StridedSlice_7:in3"
   ],
   [
    "Sub_4:out0",
    "Mul_4:in1"
   ],
   [
    "StridedSlice_8:out0",
    "Add_4:in0"
   ],
   [
    "Mul_6:out0",
    "Add_4:in1"
   ],
   [
    "StridedSlice_9:out0",
    "Mul_4:in0"
   ],
   [
    "Sub_5:out0",
    "Mul_5:in1"
   ],
   [
    "StridedSlice_10:out0",
    "Add_5:in0"
   ],
   [
    "Mul_7:out0",
    "Add_5:in1"
   ],
   [
    "StridedSlice_11:out0",
    "Mul_5:in0"
   ],
   [
    "StridedSlice_12:out0",
    "Sub_4:in0"
   ],
   [
    "StridedSlice_13:out0",
    "Sub_4:in1"
   ],
   [
    "StridedSlice_14:out0",
    "Exp:in0"
   ],
   [
    "ArgMax:out0",
    "Greater:in0"
   ],
   [
    "StridedSlice_15:out0",
    "Sub_5:in0"
   ],
   [
    "StridedSlice_16:out0",
    "Sub_5:in1"
   ],
   [
    "StridedSlice_17:out0",
    "Exp_1:in0"
   ],
   [
    "C_48:out0",
    "Split_2:in0"
   ],
   [
    "StridedSlice_18:out0",
    "Split_2:in1"
   ],
   [
    "C_49:out0",
    "Greater:in1"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "Exit:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "Greater_1:out0",
    "Where_1:in0"
   ],
   [
    "Sub_4:out0",
    "Mul_6:in1"
   ],
   [
    "StridedSlice_19:out0",
    "StridedSlice_8:in0"
   ],
   [
    "C_50:out0",
    "StridedSlice_8:in1"
   ],
   [
    "C_51:out0",
    "StridedSlice_8:in2"
   ],
   [
    "C_52:out0",
    "StridedSlice_8:in3"
   ],
   [
    "Sub_5:out0",
    "Mul_7:in1"
   ],
   [
    "C_53:out0",
    "Mul_6:in0"
   ],
   [
    "Mul_8:out0",
    "StridedSlice_9:in0"
   ],
   [
    "C_54:out0",
    "StridedSlice_9:in1"
   ],
   [
    "C_55:out0",
    "StridedSlice_9:in2"
   ],
   [
    "C_56:out0",
    "StridedSlice_9:in3"
   ],
   [
    "StridedSlice_19:out0",
    "StridedSlice_10:in0"
   ],
   [
    "C_57:out0",
    "StridedSlice_10:in1"
   ],
   [
    "C_58:out0",
    "StridedSlice_10:in2"
   ],
   [
    "C_59:out0",
    "StridedSlice_10:in3"
   ],
   [
    "C_60:out0",
    "Mul_7:in0"
   ],
   [
    "Mul_8:out0",
    "StridedSlice_11:in0"
   ],
   [
    "StridedSlice_19:out0",
    "StridedSlice_12:in0"
   ],
   [
    "C_61:out0",
    "StridedSlice_11:in1"
   ],
   [
    "C_62:out0",
    "StridedSlice_11:in2"
   ],
   [
    "C_63:out0",
    "StridedSlice_11:in3"
   ],
   [
    "StridedSlice_19:out0",
    "StridedSlice_13:in0"
   ],
   [
    "C_64:out0",
    "StridedSlice_12:in1"
   ],
   [
    "C_65:out0",
    "StridedSlice_12:in2"
   ],
   [
    "C_66:out0",
    "StridedSlice_12:in3"
   ],
   [
    "Mul_8:out0",
    "StridedSlice_14:in0"
   ],
   [
    "C_67:out0",
    "StridedSlice_13:in1"
   ],
   [
    "StridedSlice_19:out0",
    "StridedSlice_15:in0"
   ],
   [
    "C_68:out0",
    "StridedSlice_13:in2"
   ],
   [
    "C_69:out0",
    "StridedSlice_13:in3"
   ],
   [
    "StridedSlice_19:out0",
    "StridedSlice_16:in0"
   ],
   [
    "C_70:out0",
    "StridedSlice_14:in1"
   ],
   [
    "C_71:out0",
    "StridedSlice_14:in2"
   ],
   [
    "C_72:out0",
    "StridedSlice_14:in3"
   ],
   [
    "C_73:out0",
    "StridedSlice_15:in1"
   ],
   [
    "Mul_8:out0",
    "StridedSlice_17:in0"
   ],
   [
    "C_74:out0",
    "StridedSlice_15:in2"
   ],
   [
    "C_75:out0",
    "StridedSlice_15:in3"
   ],
   [
    "C_76:out0",
    "StridedSlice_16:in1"
   ],
   [
    "C_77:out0",
    "StridedSlice_16:in2"
   ],
   [
    "C_78:out0",
    "StridedSlice_16:in3"
   ],
   [
    "Reshape_1:out0",
    "Greater_1:in0"
   ],
   [
    "C_79:out0",
    "StridedSlice_17:in1"
   ],
   [
    "C_80:out0",
    "StridedSlice_17:in2"
   ],
   [
    "C_81:out0",
    "StridedSlice_17:in3"
   ],
   [
    "StridedSlice_20:out0",
    "StridedSlice_18:in0"
   ],
   [
    "C_82:out0",
    "StridedSlice_18:in1"
   ],
   [
    "C_83:out0",
    "StridedSlice_18:in2"
   ],
   [
    "C_84:out0",
    "StridedSlice_18:in3"
   ],
   [
    "C_85:out0",
    "TensorArrayV3:in0"
   ],
   [
    "C_86:out0",
    "Range:in0"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "C_87:out0",
    "Range:in2"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "C_88:out0",
    "Greater_1:in1"
   ],
   [
    "C_89:out0",
    "StridedSlice_19:in1"
   ],
   [
    "C_90:out0",
    "StridedSlice_19:in2"
   ],
   [
    "C_91:out0",
    "StridedSlice_19:in3"
   ],
   [
    "GatherNd_1:out0",
    "Mul_8:in0"
   ],
   [
    "C_92:out0",
    "Mul_8:in1"
   ],
   [
    "Pack_1:out0",
    "GatherNd_1:in1"
   ],
   [
    "StridedSlice_21:out0",
    "StridedSlice_20:in0"
   ],
   [
    "C_93:out0",
    "StridedSlice_20:in1"
   ],
   [
    "C_94:out0",
    "StridedSlice_20:in2"
   ],
   [
    "C_95:out0",
    "StridedSlice_20:in3"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "Exit:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "StridedSlice_22:out0",
    "GatherNd_1:in0"
   ],
   [
    "C_96:out0",
    "StridedSlice_21:in1"
   ],
   [
    "C_97:out0",
    "StridedSlice_21:in2"
   ],
   [
    "C_98:out0",
    "StridedSlice_21:in3"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "C_99:out0",
    "StridedSlice_22:in1"
   ],
   [
    "C_100:out0",
    "StridedSlice_22:in2"
   ],
   [
    "C_101:out0",
    "StridedSlice_22:in3"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_1:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "Enter_1:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "PadV2:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "Enter_2:out0",
    "Less:in1"
   ],
   [
    "Enter_2:out0",
    "Less_1:in1"
   ],
   [
    "Merge_2:out0",
    "Less_1:in0"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "StridedSlice_5:out0",
    "Enter_5:in0"
   ],
   [
    "C_85:out0",
    "Enter_2:in0"
   ],
   [
    "GatherV2_6:out0",
    "PadV2:in0"
   ],
   [
    "Pack_3:out0",
    "PadV2:in1"
   ],
   [
    "C_102:out0",
    "PadV2:in2"
   ],
   [
    "Enter_3:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Enter_4:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "Merge_2:out0",
    "Switch_1:in0"
   ],
   [
    "Enter_5:out0",
    "GatherV2_6:in0"
   ],
   [
    "GatherV2_7:out0",
    "GatherV2_6:in1"
   ],
   [
    "C_103:out0",
    "GatherV2_6:in2"
   ],
   [
    "Pack_4:out0",
    "Pack_3:in0"
   ],
   [
    "Identity_2:out4096",
    "C_102:in0"
   ],
   [
    "C_104:out0",
    "Enter_3:in0"
   ],
   [
    "Add_6:out0",
    "NextIteration_1:in0"
   ],
   [
    "C_105:out0",
    "Enter_4:in0"
   ],
   [
    "Add_7:out0",
    "NextIteration_2:in0"
   ],
   [
    "StridedSlice_23:out0",
    "GatherV2_7:in0"
   ],
   [
    "NonMaxSuppressionV3:out0",
    "GatherV2_7:in1"
   ],
   [
    "C_106:out0",
    "GatherV2_7:in2"
   ],
   [
    "Identity_2:out4096",
    "C_103:in0"
   ],
   [
    "C_107:out0",
    "Pack_4:in0"
   ],
   [
    "Sub_6:out0",
    "Pack_4:in1"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "Identity_2:out0",
    "Add_6:in0"
   ],
   [
    "Identity:out0",
    "Add_7:in0"
   ],
   [
    "C_108:out0",
    "Add_6:in1"
   ],
   [
    "C_109:out0",
    "Add_7:in1"
   ],
   [
    "Where_2:out0",
    "StridedSlice_23:in0"
   ],
   [
    "C_110:out0",
    "StridedSlice_23:in1"
   ],
   [
    "C_111:out0",
    "StridedSlice_23:in2"
   ],
   [
    "C_112:out0",
    "StridedSlice_23:in3"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Identity_2:out4096",
    "C_106:in0"
   ],
   [
    "GatherV2_8:out0",
    "NonMaxSuppressionV3:in0"
   ],
   [
    "GatherV2_9:out0",
    "NonMaxSuppressionV3:in1"
   ],
   [
    "C_113:out0",
    "NonMaxSuppressionV3:in2"
   ],
   [
    "C_114:out0",
    "NonMaxSuppressionV3:in3"
   ],
   [
    "Identity_2:out4096",
    "C_107:in0"
   ],
   [
    "C_115:out0",
    "NonMaxSuppressionV3:in4"
   ],
   [
    "Merge_1:out0",
    "Switch_2:in0"
   ],
   [
    "Identity_2:out4096",
    "C_108:in0"
   ],
   [
    "C_116:out0",
    "Sub_6:in0"
   ],
   [
    "StridedSlice_24:out0",
    "Sub_6:in1"
   ],
   [
    "Identity_2:out4096",
    "C_109:in0"
   ],
   [
    "ConcatV2_1:out0",
    "GatherV2_10:in0"
   ],
   [
    "Identity_2:out4096",
    "C_110:in0"
   ],
   [
    "Identity_2:out4096",
    "C_111:in0"
   ],
   [
    "Equal:out0",
    "Where_2:in0"
   ],
   [
    "Identity_2:out4096",
    "C_112:in0"
   ],
   [
    "GatherNd:out0",
    "GatherV2_11:in0"
   ],
   [
    "StridedSlice_23:out0",
    "GatherV2_8:in1"
   ],
   [
    "StridedSlice_23:out0",
    "GatherV2_9:in1"
   ],
   [
    "Identity_2:out4096",
    "C_113:in0"
   ],
   [
    "ArgMax:out0",
    "GatherV2_12:in0"
   ],
   [
    "Enter_6:out0",
    "GatherV2_8:in0"
   ],
   [
    "C_117:out0",
    "GatherV2_8:in2"
   ],
   [
    "Identity_2:out4096",
    "C_114:in0"
   ],
   [
    "StridedSlice_5:out0",
    "GatherV2_10:in1"
   ],
   [
    "Enter_7:out0",
    "GatherV2_9:in0"
   ],
   [
    "C_118:out0",
    "GatherV2_9:in2"
   ],
   [
    "Identity_2:out4096",
    "C_115:in0"
   ],
   [
    "StridedSlice_5:out0",
    "GatherV2_11:in1"
   ],
   [
    "Identity_2:out4096",
    "C_116:in0"
   ],
   [
    "StridedSlice_5:out0",
    "GatherV2_12:in1"
   ],
   [
    "Shape:out0",
    "StridedSlice_24:in0"
   ],
   [
    "Identity_2:out4096",
    "C_117:in0"
   ],
   [
    "C_119:out0",
    "StridedSlice_24:in1"
   ],
   [
    "C_120:out0",
    "StridedSlice_24:in2"
   ],
   [
    "C_121:out0",
    "StridedSlice_24:in3"
   ],
   [
    "Enter_8:out0",
    "Equal:in0"
   ],
   [
    "TensorArrayReadV3:out0",
    "Equal:in1"
   ],
   [
    "GatherV2_6:out0",
    "Shape:in0"
   ],
   [
    "Identity_2:out4096",
    "C_118:in0"
   ],
   [
    "GatherV2_10:out0",
    "Enter_6:in0"
   ],
   [
    "Identity_2:out4096",
    "C_119:in0"
   ],
   [
    "GatherV2_11:out0",
    "Enter_7:in0"
   ],
   [
    "Identity_2:out4096",
    "C_120:in0"
   ],
   [
    "Identity_2:out4096",
    "C_121:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "GatherV2_12:out0",
    "Enter_8:in0"
   ],
   [
    "Enter_9:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_10:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "C_85:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "C_122:out0",
    "GatherV2_10:in2"
   ],
   [
    "C_123:out0",
    "GatherV2_11:in2"
   ],
   [
    "C_124:out0",
    "GatherV2_12:in2"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_9:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_10:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "C_125:out0",
    "TensorArrayScatterV3:in1"
   ],
   [
    "Unique:out0",
    "TensorArrayScatterV3:in2"
   ],
   [
    "GatherV2_12:out0",
    "Unique:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "StridedSlice_21:in0"
   ],
   [
    "I:out0",
    "StridedSlice_6:in0"
   ],
   [
    "I_1:out0",
    "StridedSlice_19:in0"
   ],
   [
    "I_2:out0",
    "StridedSlice_3:in0"
   ],
   [
    "I_3:out0",
    "StridedSlice_22:in0"
   ]
  ],
  "src_out_tensor": [
   "Reshape:out0"
  ],
  "acu_lys_alias": [
   "mrcnn_detection"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "mrcnn_detection:in0"
   ],
   [
    "I_1:out0",
    "mrcnn_detection:in1"
   ],
   [
    "I_2:out0",
    "mrcnn_detection:in2"
   ],
   [
    "I_3:out0",
    "mrcnn_detection:in3"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Reshape:out0",
    "mrcnn_detection:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "mrcnn_detection": {}
  },
  "blob_map": {
   "mrcnn_detection": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "mrcnn-roi_align",
  "src_ops_alias": [
   "Reshape",
   "GatherV2",
   "C",
   "ConcatV2",
   "GatherV2_1",
   "C_1",
   "CropAndResize",
   "CropAndResize_1",
   "CropAndResize_2",
   "CropAndResize_3",
   "C_2",
   "StridedSlice",
   "StridedSlice_1",
   "C_3",
   "StopGradient",
   "StopGradient_1",
   "C_4",
   "StopGradient_2",
   "StopGradient_3",
   "C_5",
   "StopGradient_4",
   "StopGradient_5",
   "C_6",
   "StopGradient_6",
   "StopGradient_7",
   "C_7",
   "ConcatV2_1",
   "C_8",
   "C_9",
   "C_10",
   "TopKV2",
   "C_11",
   "C_12",
   "C_13",
   "GatherNd",
   "Cast",
   "GatherNd_1",
   "Cast_1",
   "GatherNd_2",
   "Cast_2",
   "GatherNd_3",
   "Cast_3",
   "Cast_4",
   "C_14",
   "C_15",
   "Add",
   "C_16",
   "Where",
   "StridedSlice_2",
   "Where_1",
   "StridedSlice_3",
   "Where_2",
   "StridedSlice_4",
   "Where_3",
   "StridedSlice_5",
   "ConcatV2_2",
   "Mul",
   "StridedSlice_6",
   "Equal",
   "C_17",
   "C_18",
   "C_19",
   "Equal_1",
   "C_20",
   "C_21",
   "C_22",
   "Equal_2",
   "C_23",
   "C_24",
   "C_25",
   "Equal_3",
   "C_26",
   "C_27",
   "C_28",
   "C_29",
   "StridedSlice_7",
   "C_30",
   "C_31",
   "C_32",
   "C_33",
   "Squeeze",
   "C_34",
   "C_35",
   "C_36",
   "C_37",
   "C_38",
   "C_39",
   "C_40",
   "Minimum",
   "C_41",
   "Maximum",
   "C_42",
   "Add_1",
   "C_43",
   "Cast_5",
   "Round",
   "RealDiv",
   "Log",
   "C_44",
   "RealDiv_1",
   "Sqrt",
   "RealDiv_2",
   "Mul_1",
   "C_45",
   "Sqrt_1",
   "Sub",
   "Sub_1",
   "Mul_2",
   "Split",
   "StridedSlice_8",
   "StridedSlice_9",
   "C_46",
   "StridedSlice_10",
   "C_47",
   "C_48",
   "C_49",
   "C_50",
   "C_51",
   "C_52",
   "StridedSlice_11",
   "C_53",
   "C_54",
   "C_55",
   "C_56",
   "C_57",
   "C_58"
  ],
  "src_inter_flow": [
   [
    "GatherV2:out0",
    "Reshape:in0"
   ],
   [
    "C:out0",
    "Reshape:in1"
   ],
   [
    "ConcatV2:out0",
    "GatherV2:in0"
   ],
   [
    "GatherV2_1:out0",
    "GatherV2:in1"
   ],
   [
    "C_1:out0",
    "GatherV2:in2"
   ],
   [
    "CropAndResize:out0",
    "ConcatV2:in0"
   ],
   [
    "CropAndResize_1:out0",
    "ConcatV2:in1"
   ],
   [
    "CropAndResize_2:out0",
    "ConcatV2:in2"
   ],
   [
    "CropAndResize_3:out0",
    "ConcatV2:in3"
   ],
   [
    "C_2:out0",
    "ConcatV2:in4"
   ],
   [
    "StridedSlice:out0",
    "GatherV2_1:in0"
   ],
   [
    "StridedSlice_1:out0",
    "GatherV2_1:in1"
   ],
   [
    "C_3:out0",
    "GatherV2_1:in2"
   ],
   [
    "StopGradient:out0",
    "CropAndResize:in1"
   ],
   [
    "StopGradient_1:out0",
    "CropAndResize:in2"
   ],
   [
    "C_4:out0",
    "CropAndResize:in3"
   ],
   [
    "StopGradient_2:out0",
    "CropAndResize_1:in1"
   ],
   [
    "StopGradient_3:out0",
    "CropAndResize_1:in2"
   ],
   [
    "C_5:out0",
    "CropAndResize_1:in3"
   ],
   [
    "StopGradient_4:out0",
    "CropAndResize_2:in1"
   ],
   [
    "StopGradient_5:out0",
    "CropAndResize_2:in2"
   ],
   [
    "C_6:out0",
    "CropAndResize_2:in3"
   ],
   [
    "StopGradient_6:out0",
    "CropAndResize_3:in1"
   ],
   [
    "StopGradient_7:out0",
    "CropAndResize_3:in2"
   ],
   [
    "C_7:out0",
    "CropAndResize_3:in3"
   ],
   [
    "ConcatV2_1:out0",
    "StridedSlice:in0"
   ],
   [
    "C_8:out0",
    "StridedSlice:in1"
   ],
   [
    "C_9:out0",
    "StridedSlice:in2"
   ],
   [
    "C_10:out0",
    "StridedSlice:in3"
   ],
   [
    "TopKV2:out1",
    "StridedSlice_1:in0"
   ],
   [
    "C_11:out0",
    "StridedSlice_1:in1"
   ],
   [
    "C_12:out0",
    "StridedSlice_1:in2"
   ],
   [
    "C_13:out0",
    "StridedSlice_1:in3"
   ],
   [
    "GatherNd:out0",
    "StopGradient:in0"
   ],
   [
    "Cast:out0",
    "StopGradient_1:in0"
   ],
   [
    "GatherNd_1:out0",
    "StopGradient_2:in0"
   ],
   [
    "Cast_1:out0",
    "StopGradient_3:in0"
   ],
   [
    "GatherNd_2:out0",
    "StopGradient_4:in0"
   ],
   [
    "Cast_2:out0",
    "StopGradient_5:in0"
   ],
   [
    "GatherNd_3:out0",
    "StopGradient_6:in0"
   ],
   [
    "Cast_3:out0",
    "StopGradient_7:in0"
   ],
   [
    "Cast_4:out0",
    "ConcatV2_1:in0"
   ],
   [
    "C_14:out0",
    "ConcatV2_1:in1"
   ],
   [
    "C_15:out0",
    "ConcatV2_1:in2"
   ],
   [
    "Add:out0",
    "TopKV2:in0"
   ],
   [
    "C_16:out0",
    "TopKV2:in1"
   ],
   [
    "Where:out0",
    "GatherNd:in1"
   ],
   [
    "StridedSlice_2:out0",
    "Cast:in0"
   ],
   [
    "Where_1:out0",
    "GatherNd_1:in1"
   ],
   [
    "StridedSlice_3:out0",
    "Cast_1:in0"
   ],
   [
    "Where_2:out0",
    "GatherNd_2:in1"
   ],
   [
    "StridedSlice_4:out0",
    "Cast_2:in0"
   ],
   [
    "Where_3:out0",
    "GatherNd_3:in1"
   ],
   [
    "StridedSlice_5:out0",
    "Cast_3:in0"
   ],
   [
    "ConcatV2_2:out0",
    "Cast_4:in0"
   ],
   [
    "Mul:out0",
    "Add:in0"
   ],
   [
    "StridedSlice_6:out0",
    "Add:in1"
   ],
   [
    "Where:out0",
    "StridedSlice_2:in0"
   ],
   [
    "Equal:out0",
    "Where:in0"
   ],
   [
    "C_17:out0",
    "StridedSlice_2:in1"
   ],
   [
    "C_18:out0",
    "StridedSlice_2:in2"
   ],
   [
    "C_19:out0",
    "StridedSlice_2:in3"
   ],
   [
    "Where_1:out0",
    "StridedSlice_3:in0"
   ],
   [
    "Equal_1:out0",
    "Where_1:in0"
   ],
   [
    "C_20:out0",
    "StridedSlice_3:in1"
   ],
   [
    "C_21:out0",
    "StridedSlice_3:in2"
   ],
   [
    "C_22:out0",
    "StridedSlice_3:in3"
   ],
   [
    "Where_2:out0",
    "StridedSlice_4:in0"
   ],
   [
    "Equal_2:out0",
    "Where_2:in0"
   ],
   [
    "C_23:out0",
    "StridedSlice_4:in1"
   ],
   [
    "C_24:out0",
    "StridedSlice_4:in2"
   ],
   [
    "C_25:out0",
    "StridedSlice_4:in3"
   ],
   [
    "Where_3:out0",
    "StridedSlice_5:in0"
   ],
   [
    "ConcatV2_1:out0",
    "StridedSlice_6:in0"
   ],
   [
    "Where:out0",
    "ConcatV2_2:in0"
   ],
   [
    "Where_1:out0",
    "ConcatV2_2:in1"
   ],
   [
    "Equal_3:out0",
    "Where_3:in0"
   ],
   [
    "Where_2:out0",
    "ConcatV2_2:in2"
   ],
   [
    "Where_3:out0",
    "ConcatV2_2:in3"
   ],
   [
    "C_26:out0",
    "StridedSlice_5:in1"
   ],
   [
    "C_27:out0",
    "StridedSlice_5:in2"
   ],
   [
    "C_28:out0",
    "StridedSlice_5:in3"
   ],
   [
    "C_29:out0",
    "ConcatV2_2:in4"
   ],
   [
    "StridedSlice_7:out0",
    "Mul:in0"
   ],
   [
    "C_30:out0",
    "Mul:in1"
   ],
   [
    "C_31:out0",
    "StridedSlice_6:in1"
   ],
   [
    "C_32:out0",
    "StridedSlice_6:in2"
   ],
   [
    "C_33:out0",
    "StridedSlice_6:in3"
   ],
   [
    "Squeeze:out0",
    "Equal:in0"
   ],
   [
    "C_34:out0",
    "Equal:in1"
   ],
   [
    "Squeeze:out0",
    "Equal_1:in0"
   ],
   [
    "C_35:out0",
    "Equal_1:in1"
   ],
   [
    "Squeeze:out0",
    "Equal_2:in0"
   ],
   [
    "C_36:out0",
    "Equal_2:in1"
   ],
   [
    "ConcatV2_1:out0",
    "StridedSlice_7:in0"
   ],
   [
    "Squeeze:out0",
    "Equal_3:in0"
   ],
   [
    "C_37:out0",
    "Equal_3:in1"
   ],
   [
    "C_38:out0",
    "StridedSlice_7:in1"
   ],
   [
    "C_39:out0",
    "StridedSlice_7:in2"
   ],
   [
    "C_40:out0",
    "StridedSlice_7:in3"
   ],
   [
    "Minimum:out0",
    "Squeeze:in0"
   ],
   [
    "C_41:out0",
    "Minimum:in0"
   ],
   [
    "Maximum:out0",
    "Minimum:in1"
   ],
   [
    "C_42:out0",
    "Maximum:in0"
   ],
   [
    "Add_1:out0",
    "Maximum:in1"
   ],
   [
    "C_43:out0",
    "Add_1:in0"
   ],
   [
    "Cast_5:out0",
    "Add_1:in1"
   ],
   [
    "Round:out0",
    "Cast_5:in0"
   ],
   [
    "RealDiv:out0",
    "Round:in0"
   ],
   [
    "Log:out0",
    "RealDiv:in0"
   ],
   [
    "C_44:out0",
    "RealDiv:in1"
   ],
   [
    "RealDiv_1:out0",
    "Log:in0"
   ],
   [
    "Sqrt:out0",
    "RealDiv_1:in0"
   ],
   [
    "RealDiv_2:out0",
    "RealDiv_1:in1"
   ],
   [
    "Mul_1:out0",
    "Sqrt:in0"
   ],
   [
    "C_45:out0",
    "RealDiv_2:in0"
   ],
   [
    "Sqrt_1:out0",
    "RealDiv_2:in1"
   ],
   [
    "Sub:out0",
    "Mul_1:in0"
   ],
   [
    "Sub_1:out0",
    "Mul_1:in1"
   ],
   [
    "Mul_2:out0",
    "Sqrt_1:in0"
   ],
   [
    "Split:out0",
    "Sub:in1"
   ],
   [
    "Split:out2",
    "Sub:in0"
   ],
   [
    "Split:out3",
    "Sub_1:in0"
   ],
   [
    "Split:out1",
    "Sub_1:in1"
   ],
   [
    "StridedSlice_8:out0",
    "Mul_2:in0"
   ],
   [
    "StridedSlice_9:out0",
    "Mul_2:in1"
   ],
   [
    "C_46:out0",
    "Split:in0"
   ],
   [
    "StridedSlice_10:out0",
    "StridedSlice_8:in0"
   ],
   [
    "C_47:out0",
    "StridedSlice_8:in1"
   ],
   [
    "C_48:out0",
    "StridedSlice_8:in2"
   ],
   [
    "C_49:out0",
    "StridedSlice_8:in3"
   ],
   [
    "StridedSlice_10:out0",
    "StridedSlice_9:in0"
   ],
   [
    "C_50:out0",
    "StridedSlice_9:in1"
   ],
   [
    "C_51:out0",
    "StridedSlice_9:in2"
   ],
   [
    "C_52:out0",
    "StridedSlice_9:in3"
   ],
   [
    "StridedSlice_11:out0",
    "StridedSlice_10:in0"
   ],
   [
    "C_53:out0",
    "StridedSlice_10:in1"
   ],
   [
    "C_54:out0",
    "StridedSlice_10:in2"
   ],
   [
    "C_55:out0",
    "StridedSlice_10:in3"
   ],
   [
    "C_56:out0",
    "StridedSlice_11:in1"
   ],
   [
    "C_57:out0",
    "StridedSlice_11:in2"
   ],
   [
    "C_58:out0",
    "StridedSlice_11:in3"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "Split:in1"
   ],
   [
    "I:out0",
    "GatherNd:in0"
   ],
   [
    "I:out0",
    "GatherNd_1:in0"
   ],
   [
    "I:out0",
    "GatherNd_2:in0"
   ],
   [
    "I:out0",
    "GatherNd_3:in0"
   ],
   [
    "I_1:out0",
    "StridedSlice_11:in0"
   ],
   [
    "I_2:out0",
    "CropAndResize:in0"
   ],
   [
    "I_3:out0",
    "CropAndResize_1:in0"
   ],
   [
    "I_4:out0",
    "CropAndResize_2:in0"
   ],
   [
    "I_5:out0",
    "CropAndResize_3:in0"
   ]
  ],
  "src_out_tensor": [
   "Reshape:out0"
  ],
  "acu_lys_alias": [
   "mrcnn_roi_align"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "mrcnn_roi_align:in0"
   ],
   [
    "I_1:out0",
    "mrcnn_roi_align:in1"
   ],
   [
    "I_2:out0",
    "mrcnn_roi_align:in2"
   ],
   [
    "I_3:out0",
    "mrcnn_roi_align:in3"
   ],
   [
    "I_4:out0",
    "mrcnn_roi_align:in4"
   ],
   [
    "I_5:out0",
    "mrcnn_roi_align:in5"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Reshape:out0",
    "mrcnn_roi_align:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "mrcnn_roi_align": {
    "pool_size": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_4:out0'])"
    ]
   }
  },
  "blob_map": {
   "mrcnn_roi_align": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "crop_and_resize",
  "src_ops_alias": [
   "CropAndResize",
   "C",
   "C_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "CropAndResize:in1"
   ],
   [
    "C_1:out0",
    "CropAndResize:in2"
   ],
   [
    "C_2:out0",
    "CropAndResize:in3"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "CropAndResize:in0"
   ]
  ],
  "src_out_tensor": [
   "CropAndResize:out0"
  ],
  "acu_lys_alias": [
   "cropandresize"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "cropandresize:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "CropAndResize:out0",
    "cropandresize:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "cropandresize": {
    "num_crop_boxes": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['C:out0'])[0]"
    ],
    "crop_size": [
     "INTS",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "resize_method": [
     "STRING",
     "CODE",
     "self.attr_pick(node['CropAndResize'], 'method', 'bilinear')"
    ]
   }
  },
  "blob_map": {
   "cropandresize": {
    "boxes": [
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "box_ind": [
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "lstm_broadcom",
  "src_ops_alias": [
   "TensorArrayGatherV3",
   "TensorArrayV3",
   "Range",
   "Exit",
   "C",
   "C_1",
   "TensorArraySizeV3",
   "C_2",
   "Switch",
   "Merge",
   "LoopCond",
   "Enter",
   "NextIteration",
   "LogicalAnd",
   "TensorArrayWriteV3",
   "Less",
   "Less_1",
   "Enter_1",
   "Identity",
   "Mul",
   "Identity_1",
   "Merge_1",
   "Enter_2",
   "Merge_2",
   "Enter_3",
   "Switch_1",
   "Sigmoid",
   "Tanh",
   "Enter_4",
   "NextIteration_1",
   "Enter_5",
   "NextIteration_2",
   "C_3",
   "Split",
   "AddV2",
   "C_4",
   "AddV2_1",
   "C_5",
   "AddV2_2",
   "C_6",
   "BiasAdd",
   "Mul_1",
   "Mul_2",
   "Identity_2",
   "C_7",
   "C_8",
   "MatMul",
   "Enter_6",
   "Sigmoid_1",
   "Identity_3",
   "Sigmoid_2",
   "Tanh_1",
   "Switch_2",
   "ConcatV2",
   "Enter_7",
   "C_9",
   "AddV2_3",
   "Switch_3",
   "TensorArrayReadV3",
   "Identity_4",
   "C_10",
   "C_11",
   "C_12",
   "Merge_3",
   "Enter_8",
   "Enter_9",
   "Switch_4",
   "Enter_10",
   "NextIteration_3",
   "TensorArrayV3_1",
   "TensorArrayScatterV3",
   "Merge_4",
   "C_13",
   "C_14",
   "Enter_11",
   "NextIteration_4",
   "C_15"
  ],
  "src_inter_flow": [
   [
    "TensorArrayV3:out0",
    "TensorArrayGatherV3:in0"
   ],
   [
    "Range:out0",
    "TensorArrayGatherV3:in1"
   ],
   [
    "Exit:out0",
    "TensorArrayGatherV3:in2"
   ],
   [
    "C:out0",
    "TensorArrayV3:in0"
   ],
   [
    "C_1:out0",
    "Range:in0"
   ],
   [
    "TensorArraySizeV3:out0",
    "Range:in1"
   ],
   [
    "C_2:out0",
    "Range:in2"
   ],
   [
    "Switch:out0",
    "Exit:in0"
   ],
   [
    "TensorArrayV3:out0",
    "TensorArraySizeV3:in0"
   ],
   [
    "Exit:out0",
    "TensorArraySizeV3:in1"
   ],
   [
    "Merge:out0",
    "Switch:in0"
   ],
   [
    "LoopCond:out0",
    "Switch:in1"
   ],
   [
    "Enter:out0",
    "Merge:in0"
   ],
   [
    "NextIteration:out0",
    "Merge:in1"
   ],
   [
    "TensorArrayV3:out1",
    "Enter:in0"
   ],
   [
    "LogicalAnd:out0",
    "LoopCond:in0"
   ],
   [
    "TensorArrayWriteV3:out0",
    "NextIteration:in0"
   ],
   [
    "Less:out0",
    "LogicalAnd:in0"
   ],
   [
    "Less_1:out0",
    "LogicalAnd:in1"
   ],
   [
    "Enter_1:out0",
    "TensorArrayWriteV3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayWriteV3:in1"
   ],
   [
    "Mul:out0",
    "TensorArrayWriteV3:in2"
   ],
   [
    "Identity_1:out0",
    "TensorArrayWriteV3:in3"
   ],
   [
    "Merge_1:out0",
    "Less:in0"
   ],
   [
    "TensorArrayV3:out0",
    "Enter_1:in0"
   ],
   [
    "Enter_2:out0",
    "Less:in1"
   ],
   [
    "Merge_2:out0",
    "Less_1:in0"
   ],
   [
    "Enter_3:out0",
    "Less_1:in1"
   ],
   [
    "Switch_1:out1",
    "Identity:in0"
   ],
   [
    "Switch:out1",
    "Identity_1:in0"
   ],
   [
    "Sigmoid:out0",
    "Mul:in0"
   ],
   [
    "Tanh:out0",
    "Mul:in1"
   ],
   [
    "C:out0",
    "Enter_2:in0"
   ],
   [
    "Enter_4:out0",
    "Merge_1:in0"
   ],
   [
    "NextIteration_1:out0",
    "Merge_1:in1"
   ],
   [
    "LoopCond:out0",
    "Switch_1:in1"
   ],
   [
    "Enter_5:out0",
    "Merge_2:in0"
   ],
   [
    "NextIteration_2:out0",
    "Merge_2:in1"
   ],
   [
    "C_3:out0",
    "Enter_3:in0"
   ],
   [
    "Merge_2:out0",
    "Switch_1:in0"
   ],
   [
    "Split:out3",
    "Sigmoid:in0"
   ],
   [
    "AddV2:out0",
    "Tanh:in0"
   ],
   [
    "C_4:out0",
    "Enter_4:in0"
   ],
   [
    "AddV2_1:out0",
    "NextIteration_1:in0"
   ],
   [
    "C_5:out0",
    "Enter_5:in0"
   ],
   [
    "AddV2_2:out0",
    "NextIteration_2:in0"
   ],
   [
    "C_6:out0",
    "Split:in0"
   ],
   [
    "BiasAdd:out0",
    "Split:in1"
   ],
   [
    "Mul_1:out0",
    "AddV2:in0"
   ],
   [
    "Mul_2:out0",
    "AddV2:in1"
   ],
   [
    "Identity:out0",
    "AddV2_2:in0"
   ],
   [
    "Identity_2:out0",
    "AddV2_1:in0"
   ],
   [
    "C_7:out0",
    "AddV2_1:in1"
   ],
   [
    "C_8:out0",
    "AddV2_2:in1"
   ],
   [
    "Identity_2:out4096",
    "C_6:in0"
   ],
   [
    "MatMul:out0",
    "BiasAdd:in0"
   ],
   [
    "Enter_6:out0",
    "BiasAdd:in1"
   ],
   [
    "Sigmoid_1:out0",
    "Mul_1:in0"
   ],
   [
    "Identity_3:out0",
    "Mul_1:in1"
   ],
   [
    "Sigmoid_2:out0",
    "Mul_2:in0"
   ],
   [
    "Tanh_1:out0",
    "Mul_2:in1"
   ],
   [
    "Switch_2:out1",
    "Identity_2:in0"
   ],
   [
    "Identity_2:out4096",
    "C_7:in0"
   ],
   [
    "Identity_2:out4096",
    "C_8:in0"
   ],
   [
    "ConcatV2:out0",
    "MatMul:in0"
   ],
   [
    "Enter_7:out0",
    "MatMul:in1"
   ],
   [
    "C_9:out0",
    "Enter_6:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_2:in1"
   ],
   [
    "Split:out0",
    "Sigmoid_2:in0"
   ],
   [
    "AddV2_3:out0",
    "Sigmoid_1:in0"
   ],
   [
    "Merge_1:out0",
    "Switch_2:in0"
   ],
   [
    "Split:out1",
    "Tanh_1:in0"
   ],
   [
    "Switch_3:out1",
    "Identity_3:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_3:in1"
   ],
   [
    "TensorArrayReadV3:out0",
    "ConcatV2:in0"
   ],
   [
    "Identity_4:out0",
    "ConcatV2:in1"
   ],
   [
    "C_10:out0",
    "ConcatV2:in2"
   ],
   [
    "Split:out2",
    "AddV2_3:in0"
   ],
   [
    "Identity:out0",
    "TensorArrayReadV3:in1"
   ],
   [
    "C_11:out0",
    "Enter_7:in0"
   ],
   [
    "C_12:out0",
    "AddV2_3:in1"
   ],
   [
    "Merge_3:out0",
    "Switch_3:in0"
   ],
   [
    "Identity_2:out4096",
    "C_10:in0"
   ],
   [
    "Enter_8:out0",
    "TensorArrayReadV3:in0"
   ],
   [
    "Enter_9:out0",
    "TensorArrayReadV3:in2"
   ],
   [
    "Switch_4:out1",
    "Identity_4:in0"
   ],
   [
    "Identity_2:out4096",
    "C_12:in0"
   ],
   [
    "LoopCond:out0",
    "Switch_4:in1"
   ],
   [
    "C:out0",
    "TensorArrayV3_1:in0"
   ],
   [
    "Enter_10:out0",
    "Merge_3:in0"
   ],
   [
    "NextIteration_3:out0",
    "Merge_3:in1"
   ],
   [
    "TensorArrayV3_1:out0",
    "Enter_8:in0"
   ],
   [
    "AddV2:out0",
    "NextIteration_3:in0"
   ],
   [
    "TensorArrayScatterV3:out0",
    "Enter_9:in0"
   ],
   [
    "Merge_4:out0",
    "Switch_4:in0"
   ],
   [
    "C_13:out0",
    "Enter_10:in0"
   ],
   [
    "TensorArrayV3_1:out0",
    "TensorArrayScatterV3:in0"
   ],
   [
    "TensorArrayV3_1:out1",
    "TensorArrayScatterV3:in3"
   ],
   [
    "Mul:out0",
    "NextIteration_4:in0"
   ],
   [
    "C_14:out0",
    "TensorArrayScatterV3:in1"
   ],
   [
    "Enter_11:out0",
    "Merge_4:in0"
   ],
   [
    "NextIteration_4:out0",
    "Merge_4:in1"
   ],
   [
    "C_15:out0",
    "Enter_11:in0"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "TensorArrayScatterV3:in2"
   ]
  ],
  "src_out_tensor": [
   "TensorArrayGatherV3:out0"
  ],
  "acu_lys_alias": [
   "lstm"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "lstm:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "TensorArrayGatherV3:out0",
    "lstm:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "lstm": {
    "time_major": [
     "BOOL",
     "VALUE",
     true
    ],
    "forget_bias": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_12:out0'])"
    ],
    "weights": [
     "INT",
     "CODE",
     "self.shape_pick(tensor['Enter_7:out0'])[1] / 4"
    ]
   }
  },
  "blob_map": {
   "lstm": {
    "wi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "wo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_w(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    input_weight = np.split(weight, [input_shape[-1]], axis=0)[0]",
      "    splited_w = np.split(input_weight, 4, 1)[index]",
      "    return splited_w",
      ""
     ]
    ],
    "hi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "hf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "ho": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"weight_name\": \"C_11:out0\", \"input_name\": \"I:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_weight_h(self, node, tensor, weight_name, input_name, index):",
      "    import numpy as np",
      "    weight = self.tensor_to_numpy(tensor[weight_name])",
      "    input_shape = self.shape_pick(tensor[input_name])",
      "    output_weight = np.split(weight, [input_shape[-1]], axis=0)[1]",
      "    splited_h = np.split(output_weight, 4, 1)[index]",
      "    return splited_h",
      ""
     ]
    ],
    "bi": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_9:out0\", \"index\": 0}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bc": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_9:out0\", \"index\": 1}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bf": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_9:out0\", \"index\": 2}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ],
    "bo": [
     "PYFUNC",
     [
      "__rule_func_additional_args = {\"bias_name\": \"C_9:out0\", \"index\": 3}",
      "def r_lstmunit_rule_get_bias(self, node, tensor, bias_name, index):",
      "    import numpy as np",
      "    bias = self.tensor_to_numpy(tensor[bias_name])",
      "    splited_bias = np.split(bias, 4, 0)[index]",
      "    return splited_bias",
      ""
     ]
    ]
   }
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "round",
  "src_ops_alias": [
   "Round"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Round:in0"
   ]
  ],
  "src_out_tensor": [
   "Round:out0"
  ],
  "acu_lys_alias": [
   "round"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "round:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Round:out0",
    "round:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "round": {}
  },
  "blob_map": {
   "round": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "sequencemask_with_maxlen",
  "src_ops_alias": [
   "Less",
   "C",
   "Cast",
   "ExpandDims",
   "C_1"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "Less:in0"
   ],
   [
    "Cast:out0",
    "Less:in1"
   ],
   [
    "ExpandDims:out0",
    "Cast:in0"
   ],
   [
    "C_1:out0",
    "ExpandDims:in1"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "ExpandDims:in0"
   ]
  ],
  "src_out_tensor": [
   "Less:out0"
  ],
  "acu_lys_alias": [
   "sequence_mask"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "sequence_mask:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Less:out0",
    "sequence_mask:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "sequence_mask": {
    "maxlen": [
     "INT",
     "CODE",
     "len(list(self.tensor_to_numpy(tensor['C:out0'])))"
    ]
   }
  },
  "blob_map": {
   "sequence_mask": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_erf",
  "src_ops_alias": [
   "Erf"
  ],
  "src_inter_flow": [],
  "src_in_anchor": [
   [
    "I:out0",
    "Erf:in0"
   ]
  ],
  "src_out_tensor": [
   "Erf:out0"
  ],
  "acu_lys_alias": [
   "erf"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "erf:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "Erf:out0",
    "erf:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "erf": {}
  },
  "blob_map": {
   "erf": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 },
 {
  "ruler_name": "r_one_hot",
  "src_ops_alias": [
   "OneHot",
   "C",
   "C_1",
   "C_2"
  ],
  "src_inter_flow": [
   [
    "C:out0",
    "OneHot:in1"
   ],
   [
    "C_1:out0",
    "OneHot:in2"
   ],
   [
    "C_2:out0",
    "OneHot:in3"
   ]
  ],
  "src_in_anchor": [
   [
    "I:out0",
    "OneHot:in0"
   ]
  ],
  "src_out_tensor": [
   "OneHot:out0"
  ],
  "acu_lys_alias": [
   "one_hot"
  ],
  "src_acu_in_tensor_map": [
   [
    "I:out0",
    "one_hot:in0"
   ]
  ],
  "src_acu_out_tensor_map": [
   [
    "OneHot:out0",
    "one_hot:out0"
   ]
  ],
  "acu_inter_flow": [],
  "param_map": {
   "one_hot": {
    "depth": [
     "INT",
     "CODE",
     "self.tensor_to_numpy(tensor['C:out0'])"
    ],
    "dtype": [
     "ORIGIN",
     "CODE",
     "self.tf_type_enum_to_ac_type(self.attr_pick(node['OneHot'], 'T', 1))"
    ],
    "on_value": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_1:out0'])"
    ],
    "off_value": [
     "FLOAT",
     "CODE",
     "self.tensor_to_numpy(tensor['C_2:out0'])"
    ],
    "axis": [
     "INT",
     "CODE",
     "self.attr_pick(node['OneHot'], 'axis', -1)"
    ]
   }
  },
  "blob_map": {
   "one_hot": {}
  },
  "priority_tip": 0,
  "pre_condition": null
 }
]