return %6 : tensor<?x2xf32>
  }

// WholeModel-LABEL: func @matmul2
// WholeModel-DAG: %[[b0:.*]] = stablehlo.constant dense<[-0.211145893
// WholeModel-DAG: %[[w0:.*]] = stablehlo.constant dense<{{\[\[}}-0.630731344, 0.54962182], [0.180364341, -0.764542698]]> : tensor<2x2xf32>

  Option<DebuggerType> debugger_type_{
      *this, "debugger_type",
      llvm::cl::init(DebuggerConfig::DEBUGGER_TYPE_UNSPECIFIED),
      llvm::cl::values(
          clEnumValN(DebuggerConfig::DEBUGGER_TYPE_WHOLE_MODEL, "whole_model",
                     "Whole model verify"),
          clEnumValN(DebuggerConfig::DEBUGGER_TYPE_INT_PER_LAYER,
                     "int_per_layer", "Int Per-layer verify"),

    DEBUGGER_TYPE_FLOAT_PER_LAYER = 3;
  }

  DebuggerType debugger_type = 1;

  // Path to save unquantized model with dump tensor ops attached.
  // Used when debugger_type is WHOLE_MODEL.
  string unquantized_dump_model_path = 2;

  // Path to save debugger related logs. Defaults to '/tmp/dumps'.
  string log_dir_path = 3;
}

// Defines various calibration options.
// Next ID: 6

    return %2 : tensor<?x2x2x2xf32>
  }

// WholeModel-LABEL: func @multiple_conv2d
// WholeModel-DAG: %[[b0:.*]] = "tf.Const"() <{value = dense<0.000000e+00>
// WholeModel-DAG: %[[b1:.*]] = "tf.Const"() <{value = dense<1.000000e+00>
// WholeModel-DAG: %[[w0:.*]] = "tf.Const"() <{value = dense<{{\[\[\[\[}}0.193340182, 0.285152316
// WholeModel-DAG: %[[w1:.*]] = "tf.Const"() <{value = dense<{{\[\[\[\[}}-0.174680978, -0.367524445

// XLA_PerChannel-DAG: %[[out_scale:.*]] = "tf.Const"() <{value = dense<7.24974147E-4> : tensor<f32>}> : () -> tensor<f32>
}

// TODO(b/308773062): Add whole_model unit-test

// -----

module {
  func.func @matmul2_with_int_per_layer(%arg0: tensor<2x2xf32> {tf_saved_model.index_path = ["input_tensor"]}) -> tensor<2x2xf32> {

          outputs={output_key: output_tensor},
      )

    return in_placeholder

  def _create_while_model(self, input_shape: Sequence[int] = (1, 32, 32, 512)):
    class WhileModel(module.Module):
      """A model with a while op."""

      def __init__(self):
        w_shape = [3, 3] + [input_shape[-1], input_shape[-1]]