tensorflow::quantization::QuantizationComponentSpec& weight_spec) {
  QuantizedType quant_type =
      CalculateUniformQuantParams(rewriter, op, weight_spec);
  if (!quant_type) return nullptr;

  std::optional<Value> quantized_val =
      AddUniformQuantizeOps(rewriter, op, quant_type);
  if (!quantized_val.has_value()) return std::nullopt;

  std::optional<TF::PartitionedCallOp> dequantized_val =

  }
  // Check quantized types.
  if (auto quant_type = element_type.dyn_cast<mlir::quant::QuantizedType>()) {
    // TFLite supports QI16, QI32, QI8, and QUI8
    if ((quant_type.getStorageTypeIntegralWidth() == 16 &&
         quant_type.isSigned()) ||
        quant_type.getStorageTypeIntegralWidth() == 8 ||
        (quant_type.getStorageTypeIntegralWidth() == 32 &&
         quant_type.isSigned()))
      return true;
  }

            /*isSigned=*/true);
      }
      if (quant_specs_.legacy_float_scale) {
        quant_type = quant::DownCastScale(quant_type, min, max, op.getLoc());
      }
    }
    rewriter.setInsertionPointAfter(stats_op);
    Type result_type = quant_type.castFromExpressedType(stats_op.getType());
    auto q =
        rewriter.create<Q>(stats_op.getLoc(), result_type, stats_op.getArg());

    Block& block = func.front();
    Operation* terminator = block.getTerminator();

    // Replace input_arg(tensor<quant_type>) -> tfr.cast
    // with input_arg(tensor<storage_type>) -> tfr.cast
    for (BlockArgument arg : block.getArguments()) {
      Type arg_type = arg.getType();
      if (auto quant_type = arg_type.cast<TensorType>()
                                .getElementType()

              is_narrow_range, is_legacy_float));
    }
    return insertQDQ(rewriter, op, quant_type, quant_op);
  }

  // Insert Quantize and Dequantize ops.
  bool insertQDQ(PatternRewriter& rewriter, arith::ConstantOp op,
                 QuantizedType quant_type, QuantizationUnit quant_op) const {
    if (!quant_type) return false;

    Operation* quantize_op = quant_op.first;

              is_narrow_range, is_legacy_float));
    }
    return insertQDQ(rewriter, op, quant_type, quant_op);
  }

  // Insert Quantize and Dequantize ops.
  bool insertQDQ(PatternRewriter& rewriter, arith::ConstantOp op,
                 QuantizedType quant_type,
                 std::pair<Operation*, int> quant_op) const {
    if (!quant_type) return false;

    Operation* quantize_op = quant_op.first;

    }

    auto quant_type = dyn_cast<quant::QuantizedType>(weight_type);
    if (!quant_type) {
      op->emitError(
          "Failed to get weight quantization parameters for weight-only "
          "quantization.");
      return;
    }

    const Type expressed_type = op->getResult(0).getType();
    const Type quantized_type =
        quant_type.castFromExpressedType(expressed_type);

        mins.push_back(rmin);
        maxs.push_back(rmax);
      }
      quant_type = quantfork::fakeQuantAttrsToType(
          op.getLoc(), num_bits, *op.getAxis(), mins, maxs, narrow_range,
          expressed, is_signed);
      if (legacy_float_scale) {
        quant_type = DownCastScale(quant_type, mins, maxs, op->getLoc());
      }
    } else if (auto stats =

TfLiteStatus QuantizeWeights(flatbuffers::FlatBufferBuilder* builder,
                             const tflite::Model* input_model,
                             BufferType quant_type,
                             bool use_updated_hybrid_scheme) {
  tflite::TensorType inference_type;
  switch (quant_type) {
    case BufferType::QUANTIZED_FLOAT16:
      inference_type = tflite::TensorType_FLOAT16;
      break;
    default:

TfLiteStatus QuantizeWeights(flatbuffers::FlatBufferBuilder* builder,
                             const tflite::Model* input_model,
                             BufferType quant_type = BufferType::QUANTIZED_INT8,
                             bool use_updated_hybrid_scheme = true);

TfLiteStatus QuantizeWeights(flatbuffers::FlatBufferBuilder* builder,