Eq(BuiltinOperator_SPLIT));
  EXPECT_THAT(GetBuiltinCode(model_.operator_codes[add->opcode_index].get()),
              Eq(BuiltinOperator_ADD));

  // There should be 5 tensors: input, output, split, split/split_dim, split:1.
  // Tensor indices could be different between original and quantized.
  EXPECT_THAT(subgraph->tensors, SizeIs(5));
  const int input_idx = 0;
  EXPECT_THAT(subgraph->tensors[input_idx]->type, Eq(TensorType_INT8));

//
// For example, given the following IR:
//
// %split_sizes = "tf.Const"() {value = dense<[1, -1, 3]> : tensor<3xi32>}
// %split_dim = "tf.Const"() {value = dense<1> : tensor<i32>}
// %0:3 = "tf.SplitV"(%input, %split_sizes, %split_dim) :
//                   (tensor<4x6xf32>, tensor<3xi32>, tensor<i32>) ->
//                   (tensor<4x1xf32>, tensor<4x2xf32>, tensor<4x3xf32>)
//

`[-rank(value), rank(value))`.}]>:$split_dim,
    Arg<TF_Tensor, [{The tensor to split.}]>:$value
  );

  let results = (outs
    Res<Variadic<TF_Tensor>, [{They are identically shaped tensors, whose shape matches that of `value`
except along `axis`, where their sizes are
`values.shape[split_dim] / num_split`.}]>:$output
  );

`seq_dim` becomes `seq_axis` * `tf.sparse_concat`: `concat_dim` becomes `axis` *
`tf.sparse_reduce_sum`: `reduction_axes` becomes `axis` *
`tf.sparse_reduce_sum_sparse`: `reduction_axes` becomes `axis` *
`tf.sparse_split`: `split_dim` becomes `axis` * `tf.listdiff` has been renamed
to `tf.setdiff1d` to match NumPy naming. * `tf.inv` has been renamed to be
`tf.reciprocal` (component-wise reciprocal) to avoid confusion with `np.inv`