model/recipe.yaml

default_stage:
  default_modifiers:
    SmoothQuantModifier:
      smoothing_strength: 0.8
      mappings:
      - - - .*norm.*
          - [.*(q|k|v)_proj]
      ignore: []
    QuantizationModifier:
      targets: [Linear]
      ignore: [embed_tokens, lm_head]
      scheme: W8A8
      kv_cache_scheme:
        num_bits: 8
        type: int
        symmetric: true
        group_size: null
        strategy: tensor
        block_structure: null
        dynamic: false
        actorder: null
        scale_dtype: null
        zp_dtype: null
        observer: memoryless_minmax
        observer_kwargs: {}
      observer: {weights: minmax, input: minmax}
      bypass_divisibility_checks: false