Type targetElementType = getElementType(parameterizedTargetType, 0);
                    return convertCollectionInternal(rawClass, targetElementType, (Iterable<?>) sourceObject, decoration, graphDetails);
                }
                if (Map.class.isAssignableFrom(rawClass)) {
                    Type targetKeyType = getElementType(parameterizedTargetType, 0);

    if (auto dq_op = dyn_cast_or_null<quantfork::DequantizeCastOp>(
            op->getOperand(i).getDefiningOp())) {
      auto type =
          mlir::cast<TensorType>(dq_op.getArg().getType()).getElementType();
      if (auto per_axis_qtype =
              mlir::dyn_cast_or_null<quant::UniformQuantizedPerAxisType>(
                  QuantizedType::getQuantizedElementType(type))) {
        return true;
      }
    }

  bool need_to_set_input_nodes_quantization_params = false;
  for (const BlockArgument arg : func.getArguments()) {
    auto shaped = mlir::dyn_cast<ShapedType>(arg.getType());
    if (shaped && mlir::isa<FloatType>(shaped.getElementType()) &&
        !has_quantize_op(arg)) {
      need_to_set_input_nodes_quantization_params = true;
      break;
    }
  }

  if (!need_to_set_input_nodes_quantization_params) {
    return false;
  }

                    CollectionSchema<P, ?> propertyCollectionsSchema = (CollectionSchema<P, ?>) propertySchema;
                    ModelType<?> elementType = propertyCollectionsSchema.getElementType();
                    nodeInitializerRegistry.ensureHasInitializer(forProperty(elementType, property, bindings.getPublicSchema().getType()));
                }
                if (property.isDeclaredAsHavingUnmanagedType()) {

    // TODO(hinsu): Handle constants that are not scalar booleans.
    auto cond_type = mlir::dyn_cast<RankedTensorType>(cond.getType());
    if (!cond_type || !cond_type.getShape().equals({}) ||
        !cond_type.getElementType().isInteger(/*width=*/1))
      return failure();

    // Identify the branch to inline.
    bool cond_value = (*cond.value_begin<APInt>()).getSExtValue();
    func::FuncOp func = cond_value ? then_func : else_func;

  SmallVector<int64_t, 4> permutedShape = applyPermutation(
      type.getShape(), isInvert ? invertPermutationVector(perm) : perm);
  return RankedTensorType::get(permutedShape, type.getElementType());
}

static RankedTensorType GetInvertPermutedTensorType(RankedTensorType type,
                                                    ArrayRef<int64_t> perm) {

    if (!const_op) return failure();

    // Only support float tensors.
    auto tensor_type = mlir::dyn_cast_or_null<TensorType>(const_op.getType());
    if (!tensor_type || !tensor_type.getElementType().isF32()) {
      return failure();
    }

    return success(
        mlir::isa_and_nonnull<DenseFPElementsAttr>(const_op.getValue()));
  }

  void rewrite(mlir::stablehlo::TransposeOp op,

    auto tensor =
        llvm::dyn_cast<TensorType>(var_handle_op.getResource().getType());
    if (!tensor) return nullptr;
    TF::ResourceType resource =
        llvm::dyn_cast<TF::ResourceType>(tensor.getElementType());
    if (!resource || resource.getSubtypes().size() != 1) return nullptr;
    return resource.getSubtypes().front();
  } else if (auto global_tensor_op =

  bool need_to_set_input_nodes_quantization_params = false;
  for (const BlockArgument arg : func.getArguments()) {
    auto shaped = mlir::dyn_cast<ShapedType>(arg.getType());
    if (shaped && mlir::isa<FloatType>(shaped.getElementType()) &&
        !has_quantize_op(arg)) {
      need_to_set_input_nodes_quantization_params = true;
      break;
    }
  }

  if (!need_to_set_input_nodes_quantization_params) {
    return false;
  }

                         QuantizationUnits& quantizable_ops) const {
    // Non-float tensors do not need quantization.
    auto type = mlir::dyn_cast<ShapedType>(op.getType());
    if (!type || !type.getElementType().isF32()) return false;

    Value value = op.getResult();

    // Check whether dynamic range quantization can be applied.
    for (auto& use : value.getUses()) {
      Operation* user = use.getOwner();