%0 = "tfl.pseudo_const"() {value = dense<[1, 2]> : tensor<2xi32>} : () -> tensor<2xi32>
   %1 = "tfl.mean"(%arg0, %0) {keep_dims = false} : (tensor<?x7x7x2048x!quant.uniform<i8:f32, 0.6:-128>>, tensor<2xi32>) -> tensor<?x2048x!quant.uniform<i8:f32, 0.6:-128>>
  func.return %1 : tensor<?x2048x!quant.uniform<i8:f32, 0.6:-128>>
}

// CHECK-SAME: {_quantization_method = "static_range_ptq { }"}
// CHECK-SAME: : (tensor<1x4x!quant.uniform<i8:f32, 6.000000e-03:-128>>, tensor<4x3x!quant.uniform<i8<-127:127>:f32:1, {5.000000e-03,5.000000e-03,5.000000e-03}>>) -> tensor<1x3x!quant.uniform<i8:f32, 1.000000e-03:-3>>
// CHECK: %[[DCAST_0:.+]] = "quantfork.dcast"(%[[CALL_0]]) :  (tensor<1x3x!quant.uniform<i8:f32, 1.000000e-03:-3>>) -> tensor<1x3xf32>
// CHECK: return

  // CHECK-SAME: quant.uniform<i8:f32, 0.023529411764705882:-128>
  // CHECK: %[[dq2:.*]] = "quantfork.dcast"(%[[q2]])
  // CHECK-SAME: quant.uniform<i8:f32, 0.023529411764705882:-128>
  // CHECK: %[[q3:.*]] = "quantfork.qcast"(%[[cst3]])
  // CHECK-SAME: quant.uniform<i8:f32, 3.9215686274509805E-9>
  // CHECK: %[[dq3:.*]] = "quantfork.dcast"(%[[q3]])
  // CHECK-SAME: quant.uniform<i8:f32, 3.9215686274509805E-9>

  %2 = "tfl.reshape"(%arg0, %0) : (tensor<4x384x32x!quant.uniform<i8:f32, 0.2:-3>>, tensor<4xi32>) -> tensor<1x4x384x32x!quant.uniform<i8:f32, 0.2:-3>>
  // CHECK-NOT: tac.device tac.inference_type
  %3 = "tfl.quantize"(%2) {qtype = tensor<1x4x384x32x!quant.uniform<i8:f32, 0.19:1>>} : (tensor<1x4x384x32x!quant.uniform<i8:f32, 0.2:-3>>) -> tensor<1x4x384x32x!quant.uniform<i8:f32, 0.19:1>>
  // CHECK: tac.device = "CPU", tac.inference_type = "QUANTIZED_INT8"

  %arg0: tensor<1x10x!quant.uniform<i8:f32, 0.1:-128>>,
  %arg1: tensor<1x5xf32>) -> (tensor<1x10x!quant.uniform<i8:f32, 0.2:42>>, tensor<1x5xf32>) {
  %0 = "tf.MyRequantize"(%arg0) : (tensor<1x10x!quant.uniform<i8:f32, 0.1:-128>>) -> tensor<1x10x!quant.uniform<i8:f32, 0.2:42>>
  %1 = "tf.Intermediate"(%arg1) : (tensor<1x5xf32>) -> tensor<1x5xf32>
  func.return %0, %1 : tensor<1x10x!quant.uniform<i8:f32, 0.2:42>>, tensor<1x5xf32>

  %q2 = "quantfork.qcast"(%q1) {volatile} : (tensor<3x2x!quant.uniform<i8:f32, 6.000000e-03:-128>>) -> tensor<3x2x!quant.uniform<i8:f32, 0.013075299590241675:-64>>
  %dq1 = "quantfork.dcast"(%q2) : (tensor<3x2x!quant.uniform<i8:f32, 0.013075299590241675:-64>>) -> tensor<3x2xf32>

// CHECK: %[[CST:.+]] = stablehlo.constant() <{value = dense<127> : tensor<2x3xi8>}> : () -> tensor<2x3x!quant.uniform<i8<-127:127>:f32, 0.0023622048182750312>>
// CHECK: %[[CALL:.+]] = call @quantized_dot_general_fn(%[[ARG0]], %[[CST]]) {_quantization_method = "weight_only_ptq { }"} : (tensor<1x2xf32>, tensor<2x3x!quant.uniform<i8<-127:127>:f32, 0.0023622048182750312>>) -> tensor<1x3xf32>
// CHECK: return %[[CALL]]

// CHECK: quantized_dot_general_fn

               ->( tensor<i8>  {tf_saved_model.index_path = [0]},
                   tensor<i16> {tf_saved_model.index_path = [0]})
                    attributes {tf_saved_model.exported_names = ["a", "b", "c"]} {
    %0 = "tf.Const"() {value = dense<42> : tensor<i8>} : () -> tensor<i8>
    %1 = "tf.Const"() {value = dense<4242> : tensor<i16>} : () -> tensor<i16>
    return %0, %1: tensor<i8>, tensor<i16>
  }
}

// -----

  %w = "tfl.pseudo_qconst"() {qtype = tensor<12x2x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>, value = dense<127> : tensor<12x2xi8>} : () -> tensor<12x2x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>
  %dq_w = "tfl.dequantize"(%w) : (tensor<12x2x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>) -> tensor<12x2xf32>

  %dq_input= "quantfork.dcast"(%q_input) : (tensor<1x3x4x3x!quant.uniform<i8:f32, 0.58810077742034317:-128>>) -> tensor<1x3x4x3xf32>
  %q_weight = "quantfork.qcast"(%weight) : (tensor<2x3x3x2xf32>) -> tensor<2x3x3x2x!quant.uniform<i8:f32, 0.074855112561992565:-1>>
  %dq_weight = "quantfork.dcast"(%q_weight) : (tensor<2x3x3x2x!quant.uniform<i8:f32, 0.074855112561992565:-1>>) -> tensor<2x3x3x2xf32>