default_stage:
  default_modifiers:
    QuantizationModifier:
      config_groups:
        group_fp8:
          targets: ['re:.*\.linear_attn\.in_proj_qkv$', 're:.*\.linear_attn\.in_proj_z$',
            're:.*\.linear_attn\.out_proj$', 're:.*\.self_attn\.q_proj$', 're:.*\.self_attn\.k_proj$',
            're:.*\.self_attn\.v_proj$', 're:.*\.mlp\.down_proj$', 're:.*\.mlp\.experts\.down_proj$',
            're:.*\.mlp\.shared_expert\.down_proj$']
          weights:
            num_bits: 8
            type: float
            symmetric: true
            group_size: null
            strategy: channel
            block_structure: null
            dynamic: false
            actorder: null
            scale_dtype: null
            zp_dtype: null
            observer: memoryless_minmax
            observer_kwargs: {}
          input_activations:
            num_bits: 8
            type: float
            symmetric: true
            group_size: null
            strategy: token
            block_structure: null
            dynamic: true
            actorder: null
            scale_dtype: null
            zp_dtype: null
            observer: null
            observer_kwargs: {}
          output_activations: null
          format: null
        group_nvfp4:
          targets: ['re:.*\.self_attn\.o_proj$', 're:.*\.mlp\.gate_proj$', 're:.*\.mlp\.up_proj$',
            're:.*\.mlp\.experts\.gate_up_proj$', 're:.*\.mlp\.shared_expert\.gate_proj$',
            're:.*\.mlp\.shared_expert\.up_proj$']
          weights:
            num_bits: 4
            type: float
            symmetric: true
            group_size: 16
            strategy: tensor_group
            block_structure: null
            dynamic: false
            actorder: null
            scale_dtype: torch.float8_e4m3fn
            zp_dtype: null
            observer: memoryless_minmax
            observer_kwargs: {}
          input_activations:
            num_bits: 4
            type: float
            symmetric: true
            group_size: 16
            strategy: tensor_group
            block_structure: null
            dynamic: local
            actorder: null
            scale_dtype: torch.float8_e4m3fn
            zp_dtype: null
            observer: minmax
            observer_kwargs: {}
          output_activations: null
          format: null
      targets: [Linear]
      ignore: [lm_head, 're:model\.embed_tokens$', 're:visual.*', 're:model\.visual.*', 're:.*\.mlp\.gate$',
        're:.*\.mlp\.shared_expert_gate$', 're:.*\.linear_attn\.in_proj_b$', 're:.*\.linear_attn\.in_proj_a$']
      bypass_divisibility_checks: false