quant_stage:
  quant_modifiers:
    QuantizationModifier:
      ignore: [lm_head, 're:.*block_sparse_moe.gate']
      scheme: FP8
      targets: Linear