recipe.yaml

quant_stage:
  quant_modifiers:
    GPTQModifier:
      sequential_update: true
      ignore: [lm_head, 're:.*mlp.gate$']
      config_groups:
        group_0:
          weights: {num_bits: 4, type: int, symmetric: true, strategy: channel, dynamic: false}
          targets: [Linear]