recipe.yaml

default_stage:
  default_modifiers:
    IMatrixGatherer:
      targets: ['re:.*layers\.(?:0|1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27)\.mlp\.(?:gate|up|down)_proj$']
      ignore: ['re:.*mtp.*', 're:.*visual.*', 're:.*vision.*', 're:.*linear_attn\.(in_proj_a|in_proj_b)$']
      weight_observer: imatrix_mse
    QuantizationModifier:
      config_groups:
        fp8_group:
          targets: ['re:.*self_attn\.(q|k|v|o)_proj$', 're:.*linear_attn\.(in_proj_qkv|in_proj_z|out_proj)$',
            're:.*lm_head$', 're:.*layers\.(?:28|29|30|31)\.mlp\.(?:gate|up|down)_proj$']
          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
      targets: [Linear]
      ignore: ['re:.*mtp.*', 're:.*visual.*', 're:.*vision.*', 're:.*linear_attn\.(in_proj_a|in_proj_b)$']
      kv_cache_scheme:
        num_bits: 8
        type: float
        symmetric: true
        group_size: null
        strategy: tensor
        block_structure: null
        dynamic: false
        actorder: null
        scale_dtype: null
        zp_dtype: null
        observer: static_minmax
        observer_kwargs: {}
      bypass_divisibility_checks: false
    GPTQModifier:
      config_groups:
        nvfp4_group:
          targets: ['re:.*layers\.(?:0|1|2|3|4|5|6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27)\.mlp\.(?:gate|up|down)_proj$']
          weights:
            num_bits: 4
            type: float
            symmetric: true
            group_size: 16
            strategy: tensor_group
            block_structure: null
            dynamic: false
            actorder: static
            scale_dtype: torch.float8_e4m3fn
            zp_dtype: null
            observer: imatrix_mse
            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: static_minmax
            observer_kwargs: {}
          output_activations: null
          format: null
      targets: [Linear]
      ignore: ['re:.*mtp.*', 're:.*visual.*', 're:.*vision.*', 're:.*linear_attn\.(in_proj_a|in_proj_b)$']
      bypass_divisibility_checks: false
      block_size: 128
      dampening_frac: 0.01
      actorder: static
      offload_hessians: false