custom_tfjs_config.json

{
  "_comment": "see https://www.tensorflow.org/js/tutorials/deployment/size_optimized_bundles to recreate the customized TF file if models change",
  "kernels": [
    "BatchMatMul",
    "Concat",
    "Sigmoid",
    "RealDiv",
    "Imag",
    "GatherV2",
    "Pack",
    "Real",
    "Less",
    "Reshape",
    "Add",
    "Unpack",
    "Sub",
    "Cast",
    "FFT",
    "Conv2D",
    "Relu",
    "ExpandDims",
    "_FusedMatMul",
    "StridedSlice",
    "Log",
    "fusedMatMul__op",
    "Sum",
    "SplitV",
    "Exp",
    "Fill",
    "Max",
    "ComplexAbs",
    "Multiply",
    "Identity",
    "Transpose",
    "PadV2",
    "Complex",
    "ZerosLike",
    "Maximum",
    "Tanh",
    "ArgMax"
  ],
  "backends": ["cpu", "webgl"],
  "_comment": "this model path is invalid, so it will produce an error, but we want the custom distro generator to export the loadGraphModel fn from tfjs-converter",
  "models": ["./models/filter.json"],
  "outputPath": "./custom_tfjs",
  "forwardModeOnly": true
}