"inference type specification";
      signalPassFailure();
      return;
    }
    quant_specs_.post_training_quantization = post_training_quantize_;
    quant_specs_.legacy_float_scale = legacy_float_scale_;
    quant_specs_.disable_set_input_nodes_quantization_params =
        disable_set_input_nodes_quantization_params_;
  }

      Option<"activation_number_of_bits_", "activation-number-of-bits", "int", "8",
             "number of bits for inference type. Only used in tests">,
      Option<"post_training_quantize_", "post-training-quantize", "bool", "false",
             "enable post training quantization. Only used in tests">,
      Option<"legacy_float_scale_", "legacy-float-scale", "bool", "false",

  }

  // Some extra flag related to post training quantization. If post-training
  // quantization is enabled, `inference_type` and `inference_input_type` are
  // not used by MLIR passes.
  if (toco_flags.post_training_quantize()) {
    quant_specs->weight_quantization = true;
    quant_specs->disable_per_channel =
        toco_flags.disable_per_channel_quantization();
    if (toco_flags.quantize_to_float16()) {

  // when toco flag post training quantization is specified with converting to
  // stablehlo, we automatically enable dynamic range quantization

  if (toco_flags.post_training_quantize()) {
    const auto status = quantization::PreprocessAndFreezeGraph(
        module, module.getContext(), session);
    if (!status.ok()) {
      return status_handler.Combine(