np.random.uniform(size=bias_shape), dtype=dtypes.float32
          )

        self._kernel = np.random.uniform(size=y_shape).astype('f4')
        self._min = (-0.8, -0.8, -0.9)
        self._max = (0.9, 0.9, 1.0)

      @def_function.function(
          input_signature=[
              tensor_spec.TensorSpec(
                  name='x', shape=x_signature, dtype=dtypes.float32
              )
          ]

                throw new IllegalArgumentException(String.format("Cannot set the %s of a property of type %s with key type %s and value type %s " +
                        "using a provider with key type %s and value type %s.", valueKind, Map.class.getName(), keyType.getName(), valueType.getName(),
                    providerKeyType.getName(), providerValueType.getName()));
            }

                                     zero_points[0], q_type.getStorageTypeMin(),
                                     q_type.getStorageTypeMax());
  } else if (auto q_type = dyn_cast<quant::UniformQuantizedPerAxisType>(type)) {
    return quant::UniformQuantizedPerAxisType::get(
        q_type.getFlags(), q_type.getStorageType(), q_type.getExpressedType(),
        scales, zero_points, q_type.getQuantizedDimension(),

        QType::getDefaultMinimumForInteger(/*isSigned=*/true, num_bits);

    auto flags = quant::QuantizationFlags::Signed;
    QType new_qtype;
    if (auto uqtype = qtype.template dyn_cast<quant::UniformQuantizedType>()) {
      new_qtype = quant::UniformQuantizedType::getChecked(
          op.getLoc(), flags, qtype.getStorageType(), qtype.getExpressedType(),
          uqtype.getScale(), uqtype.getZeroPoint() - offset,

        private final Class<? extends T> type;
        protected final Collector<T> delegate;
        private final ValueCollector<T> valueCollector;

        public TypedCollector(@Nullable Class<? extends T> type, Collector<T> delegate) {
            this.type = type;
            this.delegate = delegate;
            this.valueCollector = ValueSanitizers.collectorFor(type);
        }

        @Nullable

	//         dates:
	//           type: array
	//           items:
	//             type: string
	//             format: date-time
	//      metadata:
	//        type: object
	//        additionalProperties:
	//          type: object
	//          properties:
	//            key:
	//              type: string
	//            values:
	//              type: array
	//              items: string

// RUN: odml-to-stablehlo-opt --compose-uniform-quantized-type \
// RUN:     --split-input-file --verify-diagnostics %s | FileCheck %s

module {
// CHECK-LABEL: quantized_conv_op
// CHECK-SAME: %[[ARG:.*]]: tensor<1x3x3x4xf32>
  func.func @quantized_conv_op(%arg0: tensor<1x3x3x4xf32>) -> tensor<1x3x3x4xf32> {
    %1 = stablehlo.constant dense<1.000000e+03> : tensor<1x1x1x1xf32>  // Input inverse scale.

  ConvertToMlirShape(input_shape, &shape);
  auto type = RankedTensorType::get(shape, elt_type);

#define CONVERT_FLAT(DTYPE, CTYPE) \
  case DTYPE:                      \
    return ConvertFlatTensor<CTYPE>(input_tensor, type);

  // TODO(fengliuai): customize the conversions for quantized types.
  switch (input_dtype) {
    CONVERT_FLAT(DT_BOOL, bool)
    CONVERT_FLAT(DT_FLOAT, float)

                throw new IllegalArgumentException(String.format("Cannot set the value of a property of type %s with element type %s using a provider with element type %s.", collectionType.getName(), elementType.getName(), collectionProp.getElementType().getName()));
            }
        }
        setSupplier(new CollectingSupplier(new ElementsFromCollectionProvider<>(p)));
    }

        "//tensorflow/compiler/mlir/quantization/stablehlo/cc:saved_model_export",
        "//tensorflow/compiler/mlir/quantization/stablehlo/cc:saved_model_import",
        "//tensorflow/compiler/mlir/quantization/stablehlo/cc:types",
        "//tensorflow/compiler/mlir/quantization/stablehlo/cc:weight_only_ptq",
        "//tensorflow/compiler/mlir/quantization/stablehlo/cc/calibration:component",