%4 = "tfl.dequantize"(%3) : (tensor<32x12x!quant.uniform<u8<1:255>:f32, 0.021826678373682216:151>>) -> tensor<32x12xf32>
  %5 = "tfl.fully_connected"(%2, %4, %cst) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<1x224x224x3xf32>, tensor<32x12xf32>, tensor<32xf32>) -> tensor<1x112x112x32xf32>

    ```mlir
      %0 = "tf.Const"() {value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
      %1 = "tf.Const"() {device = "", value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
      %2 = "tf.Const"() {device = "baz", value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
    ```

    then running this pass with 'default-device=foobar', we get:

    ```mlir

  // CHECK-NEXT: [[HANDLE:%.*]] = tf_mlrt.executeop
  %handle = "tf.VarHandleOp"() {__op_key = 3: i32, container = "x", shared_name = "y"} : () -> tensor<!tf_type.resource<tensor<3x1xf32>>>
  // CHECK-NEXT: "tf_mlrt.ifrt_restore_variable"([[PREFIX]], [[NAME]], [[SLICE]], [[HANDLE]]) <{restored_dtypes = [f32], truncate_in_cast = array<i1: true>}>

    "tfl.yield"(%1#0, %1#1, %1#2) : (tensor<1x?x16xf32>, tensor<1x!tf_type.string>, tensor<1xi32>) -> ()
  }) {_tfl_quant_trait = "fully_quantizable"} : (tensor<1x!tf_type.string>) -> (tensor<1x?x16xf32>, tensor<1x!tf_type.string>, tensor<1xi32>)

func.func private @gelu_decomp_2(%arg0: tensor<5x10xf32>) -> tensor<5x10xf32>
func.func @gelu_aten_approximate(%arg0: tensor<5x10xf32>) -> (tensor<*xf32>) {
  %0 = mhlo.composite "aten.gelu.default" %arg0 {composite_attributes = {approximate = "tanh"}, decomposition = @gelu_decomp_2} : (tensor<5x10xf32>) -> tensor<5x10xf32>

  // expected-error @+1 {{expect 'output' num_elements % 40 == 0, got 'tensor<1x41xf32>'}}
  %0 = "tfl.fully_connected"(%arg0, %arg1, %arg2) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<1x37xf32>, tensor<40x37xf32>, tensor<40xf32>) -> tensor<1x41xf32>
  func.return %0 : tensor<1x41xf32>
}

// -----

  // CHECK: return %0
}

// CHECK-LABEL: testAddOfNegRight
func.func @testAddOfNegRight(%arg0: tensor<8x16xf32>, %arg1: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.Neg"(%arg1) : (tensor<8x16xf32>) -> tensor<8x16xf32>
  %1 = "tf.Add"(%arg0, %0) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<8x16xf32>, tensor<8x16xf32>) -> tensor<8x16xf32>

func.func @QuantizedCatsAddRequantsTest(%arg0: tensor<1x1xf32>, %arg1: tensor<1x1xf32>, %arg2: tensor<1x1xf32>, %arg3: tensor<1x1xf32>) -> (tensor<1x4xf32>, tensor<1x3xf32>) {
  %0 = "quantfork.stats"(%arg0) {layerStats = dense<[-0.440728068, 0.189515018]> : tensor<2xf32>} : (tensor<1x1xf32>) -> tensor<1x1xf32>
  %1 = "quantfork.stats"(%arg1) {layerStats = dense<[-0.154693216, 0.26483655]> : tensor<2xf32>} : (tensor<1x1xf32>) -> tensor<1x1xf32>

// CHECK:           %[[VAL_2:.*]] = "tf.Const"() <{value = dense<1> : tensor<i64>}> : () -> tensor<i64>
// CHECK:           %[[VAL_3:.*]] = "tf.ConcatV2"(%[[VAL_0]], %[[VAL_1]], %[[VAL_2]]) : (tensor<3x3xf32>, tensor<3x3xf32>, tensor<i64>) -> tensor<3x6xf32>
// CHECK:           return %[[VAL_3]] : tensor<3x6xf32>
// CHECK:         }
func.func @concat_v2_1d_axis(%arg0: tensor<3x3xf32>, %arg1: tensor<3x3xf32>) -> tensor<3x6xf32> {

    }) {is_stateless = false} : (tensor<i32>, tensor<!tf_type.variant<tensor<?x1xf32>>>) -> (tensor<i32>, tensor<!tf_type.variant<tensor<?x1xf32>>>)
    %elem_1 = "tf._SomeOtherOp"() : () -> tensor<8x1xf32>
    %tl_set_item = "tf.TensorListSetItem"(%while#1, %one, %elem_1) : (tensor<!tf_type.variant<tensor<?x1xf32>>>, tensor<i32>, tensor<8x1xf32>) -> tensor<!tf_type.variant<tensor<?x1xf32>>>
    func.return
  }