recipe.yaml

default_stage:
  default_modifiers:
    SmoothQuantModifier: {smoothing_strength: 0.6}
    GPTQModifier:
      block_size: 64
      scheme: W8A8
      targets: Linear
      ignore: [lm_head]