because the tensors are loaded asynchronously.
  }];

  let arguments = (ins
    Arg<TF_StrTensor, [{Must have a single element.  The prefix of a V2 checkpoint.}]>:$prefix,
    Arg<TF_StrTensor, [{shape {N}.  The names of the tensors to be restored. This is tensor name in the model.}]>:$tensor_names,
    Arg<TF_StrTensor, [{shape {N}.  The slice specs of the tensors to be restored.

  let summary =
    "Transforms a vector of tf.Example protos (as strings) into typed tensors.";

  let arguments = (ins
    TF_StrTensor:$serialized,
    TF_StrTensor:$names,
    TF_StrTensor:$sparse_keys,
    TF_StrTensor:$dense_keys,
    TF_StrTensor:$ragged_keys,
    Variadic<TensorOf<[TF_Float32, TF_Int64, TF_Str]>>:$dense_defaults,

    ConfinedAttr<I64Attr, [IntMinValue<0>]>:$num_sparse,

    Arg<TF_VariantTensor, [{A variant tensor representing the input dataset.}]>:$input_dataset,
    Arg<TF_StrTensor, [{A `tf.string` vector `tf.Tensor` identifying user enabled optimizations.}]>:$optimizations_enabled,
    Arg<TF_StrTensor, [{A `tf.string` vector `tf.Tensor` identifying user disabled optimizations.}]>:$optimizations_disabled,
    Arg<TF_StrTensor, [{A `tf.string` vector `tf.Tensor` identifying optimizations by default.}]>:$optimizations_default,

// String/variant/resource types (including corresponding reference types)

def TF_Str : AnyTypeOf<
  [TF_TensorFlowType<"String", "str">, TF_StrRef], "string">;
def TF_StrTensor : TensorOf<[TF_Str]>;

def TF_Variant : AnyTypeOf<
  [TF_TensorFlowType<"Variant", "var">, TF_VariantRef], "variant">;
def TF_VariantTensor : TensorOf<[TF_Variant]>;

def TF_Resource : AnyTypeOf<