default_stage:
  default_modifiers:
    SmoothQuantModifier: {smoothing_strength: 0.5}
    QuantizationModifier:
      targets: [Linear]
      ignore: [lm_head, model.embed_tokens]
      scheme: W8A8