quantize_config.json

{
  "quant_method": "fp8_mixed",
  "bits": 8,
  "protected_layers": [],
  "desc": "FP8 mixed precision quantization for Gemma 3 12B (Heretic-X Edition)"
}