unpacked_buffer.reserve(num_elements);

  for (uint8_t value : src_buffer) {
    // Cast to signed before right-shifting to ensure correct sign extension
    unpacked_buffer.push_back(static_cast<int8_t>(value << 4) >> 4);
    unpacked_buffer.push_back(static_cast<int8_t>(value) >> 4);
  }

  // The last element might be a padded zero, so check and pop if needed
  if (unpacked_buffer.size() > num_elements) {

    inputs.push_back(preceding_const_op.getValue());
  }

  SmallVector<Attribute> result_values;
  if (failed(TF::EvaluateOperation(op, inputs, result_values))) {
    return failure();
  }

  results.clear();
  builder.setInsertionPointAfter(op);
  for (const auto& result_value : result_values) {
    results.push_back(builder.create<TF::ConstOp>(op->getLoc(), result_value));
  }

    for (const mlir::Type &t : types) {
      if (mlir::isa<mhlo::TokenType>(t)) continue;
      auto layout = GetTPUInfeedLayout({t}, rewriter);
      if (failed(layout)) return failure();
      v.push_back(layout.value());
    }
    ArrayRef<Attribute> shape(v);
    return rewriter.getArrayAttr(shape);
  } else if (mlir::isa<TupleType>(types[0])) {
    auto tuple_type = mlir::dyn_cast<TupleType>(types[0]);

        input_tensor->quantization->scale.clear();
        input_tensor->quantization->zero_point.clear();
        input_tensor->quantization->min.push_back(-1.0);
        input_tensor->quantization->max.push_back(1.0);

        auto& output_tensor = sub_graph->tensors[op->outputs[0]];
        auto shape = output_tensor->shape;
        output_tensor->quantization->scale.clear();

    for (auto result : llvm::enumerate(results)) {
      TF::AssignVariableOp assign_var =
          GetSingleUseResourceWrite(parallel_execute, result.value());
      if (!assign_var) {
        results_to_remap.push_back(result.value());
        continue;
      }

      // Move AssignVariableOp and update the value to be written to the
      // resource variable to be the non forwarded value from within the

  EXPECT_THAT(pre_calibration_result, IsOk());

  SmallVector<func::FuncOp> func_ops;
  for (auto func_op : pre_calibration_result->getOps<func::FuncOp>()) {
    func_ops.push_back(func_op);
  }
  ASSERT_THAT(func_ops, SizeIs(2));
  EXPECT_THAT(func_ops, Contains(HasSymName("main")));
  EXPECT_THAT(func_ops, Contains(HasSymName("composite_dot_general_fn_1")));

SmallVector<Value> GetAccessedResources(Operation& op) {
  SmallVector<Value, 4> accessed_resources;
  for (auto operand : op.getOperands()) {
    if (!IsResourceType(operand.getType())) continue;
    accessed_resources.push_back(operand);
  }
  return std::move(accessed_resources);
}

// Lifts the tail writes outside of tf_device.replicate. The written value is
// added to the values returned by tf_device.replicate op. Modify the assign

        t.flat<float>().setZero();
        outputs->push_back(t);
      } else if (output_name == "dense/kernel") {
        Tensor t =
            Tensor(tensorflow::DT_FLOAT, tensorflow::TensorShape({100, 50}));
        t.flat<float>().setZero();
        outputs->push_back(t);
      } else {
        // Create a scalar float tensor.

        auto dequantize_op = builder->create<TFL::DequantizeOp>(
            op->getLoc(), dequantized_input_type, input.get());
        dequantized_inputs.push_back(dequantize_op);
      } else {
        dequantized_inputs.push_back(input.get());
      }
    }

    // Result types.
    SmallVector<Type, 4> result_types;
    for (auto result_type : op->getResultTypes()) {

    worklist.pop();
    auto result = u.walk([&](SymbolUserOpInterface op) {
      if (llvm::isa<T, Types...>(op)) {
        if (!predicate || predicate(op)) {
          hits.push_back(op);
        } else {
          first_misses.push_back(op);
          return WalkResult::advance();
        }
      }
      llvm::SmallVector<func::FuncOp> callees;
      if (GetCallees(op, symtab, callees).failed()) {