default_stage:
  default_modifiers:
    SmoothQuantModifier:
      smoothing_strength: 0.8
      mappings:
      - !!python/tuple
        - ['re:.*self_attn.q_proj', 're:.*self_attn.k_proj', 're:.*self_attn.v_proj']
        - re:.*self_attn.(q_norm|k_norm)
      - !!python/tuple
        - ['re:.*mlp.gate_proj', 're:.*mlp.up_proj']
        - re:.*post_attention_layernorm
      ignore: []
    GPTQModifier:
      targets: [Linear]
      ignore: [lm_head]
      scheme: W8A8
      block_size: 128
      dampening_frac: 0.01
      actorder: static
      offload_hessians: false