%2 = "quantfork.stats"(%1) {layerStats = dense<[5.00000000e-6, 7.00000000e-1]> : tensor<2xf32>} : (tensor<1x3xf32>) -> tensor<1x3xf32>
    return %2 : tensor<1x3xf32>
  }
// CHECK: func.func private @quantize_dot_general_with_bias_same_shape_fn(%[[ARG_0:.+]]: tensor<1x2xf32>) -> tensor<1x3xf32> attributes {tf._original_func_name = "main_0"}

^bb0(%arg0: tensor<1x2xf32>, %arg1: tensor<f32>, %arg2: tensor<f32>):
  %cst_0 = arith.constant dense<[2, -1]> : tensor<2xi64>
  %0 = "tf.FakeQuantWithMinMaxVars"(%arg0, %arg1, %arg2) {num_bits = 5, narrow_range = false} : (tensor<1x2xf32>, tensor<f32>, tensor<f32>) -> tensor<1x2xf32>
  %1 = "tf.Reshape"(%0, %cst_0) : (tensor<1x2xf32>, tensor<2xi64>) -> tensor<2x1xf32>

  %res = "tf.Relu6"(%biasadd) : (tensor<*xf32>) -> tensor<*xf32>
  %q_res = "quantfork.qcast"(%res) : (tensor<*xf32>) -> tensor<*x!quant.uniform<i8:f32, 0.023529411764705882:-128>>
  %dq_res = "quantfork.dcast"(%q_res) : (tensor<*x!quant.uniform<i8:f32, 0.023529411764705882:-128>>) -> tensor<*xf32>

  func.return %dq_res : tensor<*xf32>
}

^bb0(%arg0: tensor<1x2xf32>, %arg1: tensor<f32>, %arg2: tensor<f32>):
  %cst_0 = arith.constant dense<[2, -1]> : tensor<2xi64>
  %0 = "tf.FakeQuantWithMinMaxVars"(%arg0, %arg1, %arg2) {num_bits = 3, narrow_range = false} : (tensor<1x2xf32>, tensor<f32>, tensor<f32>) -> tensor<1x2xf32>
  %1 = "tf.Reshape"(%0, %cst_0) : (tensor<1x2xf32>, tensor<2xi64>) -> tensor<2x1xf32>

  %6 = "tfl.quantize"(%5) {qtype = tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>} : (tensor<1x112x112x32xf32>) -> tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>
  func.return %6 : tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>

    %1 = "tfl.abs"(%arg2) {fused_activation_function = "NONE"} : (tensor<5xf32>) -> tensor<5xf32>
    %2 = stablehlo.add %1, %arg2 : tensor<5xf32>
    %3 = "tfl.abs"(%2) {fused_activation_function = "NONE"} : (tensor<5xf32>) -> tensor<5xf32>
    "stablehlo.return"(%3) : (tensor<5xf32>) -> ()
  }) {dimensions = array<i64: 0>} : (tensor<7x5xf32>, tensor<5xf32>) -> tensor<5xf32>
  func.return %0: tensor<5xf32>
}

      time_major = false} : (
        tensor<1x2x3xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        none, none, none,
        tensor<3xf32>, tensor<3xf32>, tensor<3xf32>, tensor<3xf32>,
        none, none,
        tensor<1x3xf32>, tensor<1x3xf32>,
        none, none, none, none) -> tensor<1x2x3xf32>

// QDQ-NEXT:  return %[[split]]#0 : tensor<2xf32>
}

// CHECK-LABEL: FoldTranspose
func.func @FoldTranspose(%arg0: tensor<1x10x20x3xf32>) -> tensor<1x20x40x16xf32> {
  %cst = arith.constant dense<[1, 20, 40, 16]> : tensor<4xi32>
  %cst_0 = arith.constant dense<[2, 0, 1, 3]> : tensor<4xi32>

module {
  func.func @wrap_composite_func(%arg0: tensor<*xf32>, %arg1: tensor<*xf32>) -> tensor<*xf32> {
    %0 = "tf.PartitionedCall"(%arg0, %arg1) <{f = @composite_conv2d_with_relu6_fn}> {_tfl_quant_trait = "fully_quantizable"}
          : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
    func.return %0 : tensor<*xf32>
  }

  func.func @no_composite_func(%arg0: tensor<*xf32>, %arg1: tensor<*xf32>) -> tensor<*xf32> {

    %0 = "quantfork.stats"(%arg1) {layerStats = dense<[-3.54062747, 0.54742622]> : tensor<2xf32>} : (tensor<2x3x3x2xf32>) -> tensor<2x3x3x2xf32>
    %1 = "quantfork.stats"(%arg0) {layerStats = dense<[1.27501142, 2.824783]> : tensor<2xf32>} : (tensor<1x3x2x3xf32>) -> tensor<1x3x2x3xf32>