// CHECK-SAME: quant.uniform<i8:f32, 0.0040316890267764818:127>
  %cst = stablehlo.constant dense<[[-0.960978984, -0.390246302], [-0.790828585, -0.601039409], [-1.0280807, -1.02731466]]> : tensor<3x2xf32>
  // CHECK: %[[q2:.*]] = "quantfork.qcast"(%[[ARG_0]])
  // CHECK-SAME: quant.uniform<i4:f32, 0.13329366842905679:-1>
  // CHECK: %[[dq2:.*]] = "quantfork.dcast"(%[[q2]])
  // CHECK-SAME: quant.uniform<i4:f32, 0.13329366842905679:-1>

include "mlir/IR/OpBase.td"
include "mlir/Dialect/Quant/QuantOpsBase.td"

//===----------------------------------------------------------------------===//
// QuantizedType definitions.
//===----------------------------------------------------------------------===//

// The base class of a quantized type.
class QuantizedType<string n, list<int> params, bit signed>
  : Type<And<[CPred<"$_self.isa<mlir::quant::QuantizedType>()">,

tensor<2048x640x!quant.uniform<i8<-127:127>:f32, 0.032276000827550888>>, tensor<2048x640x!quant.uniform<i8<-127:127>:f32, 0.035427000373601913>>, none, none, none, tensor<2048x!quant.uniform<i32:f32, 4.2675782196965883E-7>>, tensor<2048x!quant.uniform<i32:f32, 1.0742187583900886E-7>>, tensor<2048x!quant.uniform<i32:f32, 1.6406249869760359E-7>>, tensor<2048x!quant.uniform<i32:f32, 1.523437447303877E-7>>, tensor<640x2048x!quant.uniform<i8<-127:127>:f32, 0.021174000576138496>>, tensor<640x!quant.uniform<i32:f32,...

  // CHECK-NEXT:}

  %0 = "tfl.pseudo_qconst"() { qtype = tensor<3x!quant.uniform<i8:f32, 0.1>>, value = dense<2> : tensor<3xi8>} : () -> tensor<3x!quant.uniform<i8:f32, 0.1>>
  %1 = "tfl.mul"(%arg0, %0) {fused_activation_function = "NONE"} : (tensor<3x!quant.uniform<i8:f32, 0.1>>, tensor<3x!quant.uniform<i8:f32, 0.1>>) -> tensor<3x!quant.uniform<i8:f32, 0.1>> loc("mul")
  func.return %1 : tensor<3x!quant.uniform<i8:f32, 0.1>>

  // CHECK-NEXT:}

  %0 = "tfl.pseudo_qconst"() { qtype = tensor<3x!quant.uniform<i8:f32, 1.0>>, value = dense<2> : tensor<3xi8>} : () -> tensor<3x!quant.uniform<i8:f32, 1.0>>
  %1 = "tfl.mul"(%arg0, %0) {fused_activation_function = "NONE"} : (tensor<3x!quant.uniform<i8:f32, 1.0>>, tensor<3x!quant.uniform<i8:f32, 1.0>>) -> tensor<3x!quant.uniform<i8:f32, 1.0>> loc("mul")
  func.return %1 : tensor<3x!quant.uniform<i8:f32, 1.0>>

  }
  pm.addNestedPass<mlir::func::FuncOp>(
      mlir::quant::CreateConvertTfXlaOpToTfOpPass());
  pm.addNestedPass<mlir::func::FuncOp>(
      mlir::quant::CreatePrepareLiftingPass(quantization_options.op_set()));

  pm.addPass(mlir::quant::CreateLiftQuantizableSpotsAsFunctionsPass(
      quantization_options));
  pm.addPass(mlir::quant::CreateInsertQuantizedFunctionsPass(

// 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]]

  %5 = "tfl.reshape"(%4, %0) : (tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>, tensor<2xi32>) -> tensor<1x401408x!quant.uniform<u8:f32, 0.023528476789885875>>
  %6 = "tfl.softmax"(%5) {beta = 1.000000e+00 : f32} : (tensor<1x401408x!quant.uniform<u8:f32, 0.023528476789885875>>) -> tensor<1x401408x!quant.uniform<u8:f32, 3.906250e-03>>

// CHECK-DAG: %[[input_7:.*]] = "tfl.dequantize"({{.*}}) : (tensor<2x4x!quant.uniform<i8<-127:127>:f32, 0.010094007169167826>>) -> tensor<2x4xf32>
// CHECK-DAG: %[[input_8:.*]] = "tfl.dequantize"({{.*}}) : (tensor<2x4x!quant.uniform<i8<-127:127>:f32, 0.018637238525030179>>) -> tensor<2x4xf32>

// CustomOp-DAG: %[[q_b:.*]] = "tfl.quantize"(%[[b]]) <{qtype = tensor<2048x1x1x1x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>}> : (tensor<2048x1x1x1xf32>) -> tensor<2048x1x1x1x!quant.uniform<i8<-127:127>:f32, 1.000000e+00>>