expanded_fft_len.getResult());

    // Insert the squeeze op.
    auto squeeze_dim = rewriter.getI64ArrayAttr({-2});
    TF::SqueezeOp squeeze = rewriter.create<TF::SqueezeOp>(
        rfft_op.getLoc(), output_type, rfft2d.getResult(), squeeze_dim);

    rewriter.replaceOp(op, squeeze.getResult());

    return success();
  }
};

  %outputs_20 = "tf.ResizeBilinear"(%outputs_18, %outputs) {align_corners = false, device = "", half_pixel_centers = false} : (tensor<1x?x?x3xui8>, tensor<2xi32>) -> tensor<1x224x224x3xf32>
  %outputs_22 = "tf.Squeeze"(%outputs_20) {device = "", squeeze_dims = [0]} : (tensor<1x224x224x3xf32>) -> tensor<224x224x3xf32>
  %outputs_24 = "tf.Cast"(%outputs_22) {Truncate = false, device = ""} : (tensor<224x224x3xf32>) -> tensor<224x224x3xui8>

  auto float_graph = readonly_model_->subgraphs()->Get(0);
  // The original model reshape->custom->custom->squeeze.
  ASSERT_THAT(*float_graph->operators(), SizeIs(4));
  // The resulting model should be:
  // reshape->dequantize->custom->custom->quantize->squeeze.
  ASSERT_THAT(subgraph->operators, SizeIs(6));
  const std::vector<BuiltinOperator> op_codes = {

func.func @PadStridedSliceNewAxisMask2(%arg0: tensor<4x64x64x1xf32>) -> tensor<1x4x64x64xf32> {
  %cst = arith.constant dense<0> : tensor<3xi32>
  %cst_0 = arith.constant dense<1> : tensor<3xi32>
  %0 = "tf.Squeeze"(%arg0) {T = f32, _output_shapes = ["tfshape$dim { size: 4 } dim { size: 64 } dim { size: 64 }"], device = "", squeeze_dims = []} : (tensor<4x64x64x1xf32>) -> tensor<4x64x64xf32>

          {"MISC",
           // clang-format off
     {"ApproxTopK", "BroadcastTo", "ExpandDims", "Fill", "NoOp",
      "Range", "Rank", "Reshape", "Shape", "ShapeN", "Size", "Squeeze",
      "Transpose", "ZerosLike", "OnesLike", "BiasAdd" /*PW + Broadcast*/,
      "BroadcastArgs", "BroadcastGradientArgs", "OneHot", "Concat", "ConcatV2",