default_stage:
  default_modifiers:
    GPTQModifier:
      ignore: [lm_head, model.embed_tokens]
      scheme: W4A16_ASYM
      block_size: 32