quantization_stage:
  sparsity_modifiers:
    SparseGPTModifier:
      sparsity: 0.5
      mask_structure: '2:4'
      sequential_update: false
      sequential_targets: [LlamaDecoderLayer]
      targets: [Linear]
      ignore: ['re:.*lm_head']
      block_size: 128
      dampening_frac: 0.01
      preserve_sparsity_mask: false
      offload_hessians: false
  quantization_modifiers:
    QuantizationModifier:
      targets: [Linear]
      ignore: [lm_head]
      scheme: FP8_DYNAMIC