// CHECK-SAME: %[[input_0]],
// CHECK-SAME: %[[input_1]], %[[input_2]], %[[input_3]], %[[input_4]],
// CHECK-SAME: %[[input_5]], %[[input_6]], %[[input_7]], %[[input_8]],
// CHECK-SAME: %[[input_9]], %[[input_9]], %[[input_9]],
// CHECK-SAME: %[[input_10]], %[[input_11]], %[[input_12]], %[[input_13]],
// CHECK-SAME: %[[input_9]], %[[input_9]],
// CHECK-SAME: %[[input_14]], %[[input_15]],

%9 = "tf.opJ"() : () -> tensor<i1>
%output:8 = tf_device.replicate([%0, %1] as %input_0: tensor<i32>,
                                [%2, %3] as %input_1: tensor<f32>,
                                [%4, %5] as %input_2: tensor<!tf_type.resource>,
                                [%6, %7] as %input_3: tensor<!tf_type.string>,
                                %8 as %input_4: tensor<!tf_type.variant>)
                {n = 2 : i32,

                @InputFiles FileCollection inputs1
                @InputFiles FileCollection inputs2

                @OutputDirectory File output = project.buildDir

                @TaskAction void action() {}
            }
        """

        buildFile << """
            task test(type: TaskWithTwoFileCollectionInputs) {
                inputs1 = files("input1.txt", "input2.txt")

  tensorflow::Scope s = tensorflow::Scope::NewRootScope();
  Output input_a = ops::Placeholder(s.WithOpName("input_a"), DT_INT32,
                                    ops::Placeholder::Attrs{});
  Output input_b = ops::Placeholder(s.WithOpName("shape_b"), DT_FLOAT,
                                    ops::Placeholder::Attrs{});
  Output mul = ops::Mul(s.WithOpName("mul"), input_b, input_b);
  Output shape_of_mul = ops::Shape(s.WithOpName("shape_of_mul"), mul);

      : inputs_(inputs.begin(), inputs.end()) {}

  InputList(
      const std::initializer_list<Output>& out) {  // NOLINT(runtime/explicit)
    for (auto const& x : out) {
      inputs_.push_back(x);
    }
  }

  typename std::vector<Input>::iterator begin() { return inputs_.begin(); }
  typename std::vector<Input>::iterator end() { return inputs_.end(); }

  std::vector<TFE_TensorHandle*> compute_fn_inputs;
  TFE_TensorHandle* input_a = TestScalarTensorHandle(ctx, 2.0f);
  TFE_TensorHandle* input_b = TestScalarTensorHandle(ctx, 1.0f);
  compute_fn_inputs.push_back(input_a);
  compute_fn_inputs.push_back(input_b);

  TFE_Op* compute_fn_op = TF_ConcreteFunctionMakeCallOp(

  }

  void Run(const std::vector<TF_Output>& run_outputs,
           std::initializer_list<int> input_values) {
    DCHECK_EQ(inputs_.size(), input_values.size());
    std::vector<std::pair<TF_Operation*, TF_Tensor*>> inputs(inputs_.size());
    int i = 0;
    for (int v : input_values) {
      inputs[i] = {inputs_[i].oper, Int32Tensor(v)};
      ++i;
    }
    // TODO(skyewm): use std::make_unique or absl::make_unique when possible.

    TF_AddInput(op_.get(), t->output_);
    return absl::OkStatus();
  }
  Status AddInputList(absl::Span<AbstractTensorHandle* const> inputs) override {
    std::vector<TF_Output> tf_outputs(inputs.size());
    for (int i = 0; i < inputs.size(); i++) {
      GraphTensor* t = dyn_cast<GraphTensor>(inputs[i]);
      if (!t) {
        return tensorflow::errors::InvalidArgument(
            "Unable to cast input to GraphTensor");
      }

  std::vector<const Tensor*> inputs;
  inputs.push_back(a);
  inputs.push_back(b);
  TF_ASSERT_OK_AND_ASSIGN(auto execute_outputs,
                          RunExecutable(inputs, {}, result, executable));

  TF_EXPECT_OK(PopulateCtxOutputsFromPjRtExecutableOutputs(
      /*num_missing_prefix_ctx_inputs=*/0, inputs, {}, *result,

inputs-report.png...