default_stage:
  default_modifiers:
    AWQModifier:
      targets: [Linear]
      ignore: [lm_head, 're:.*visual.*', 're:.*linear_attn\.in_proj_a', 're:.*linear_attn\.in_proj_b',
        're:.*mlp\.gate$', 're:.*shared_expert_gate', 're:.*linear_attn\.in_proj_qkv', 're:.*linear_attn\.in_proj_z',
        're:.*linear_attn\.out_proj', 're:.*mlp\.shared_expert\.gate_proj', 're:.*mlp\.shared_expert\.up_proj',
        're:.*mlp\.shared_expert\.down_proj', 're:.*self_attn\.q_proj', 're:.*self_attn\.k_proj',
        're:.*self_attn\.v_proj', 're:.*self_attn\.o_proj']
      scheme: NVFP4
      bypass_divisibility_checks: false
      mappings:
      - smooth_layer: re:.*post_attention_layernorm$
        balance_layers: ['re:.*mlp\.shared_expert\.gate_proj$', 're:.*mlp\.shared_expert\.up_proj$']
        activation_hook_target: null
      - smooth_layer: re:.*shared_expert\.up_proj$
        balance_layers: ['re:.*shared_expert\.down_proj$']
        activation_hook_target: null
      offload_device: !!python/object/apply:torch.device [cpu]
      duo_scaling: both
      n_grid: 20
      grid_batch_size: 16