quantization_config.json

{
  "bits": 4,
  "group_size": 32,
  "sym": true,
  "data_type": "mx_fp4e2m1",
  "batch_size": 4,
  "iters": 0,
  "nsamples": 512,
  "autoround_version": "0.5.1",
  "quant_method": "auto-round",
  "packing_format": "mx_fp",
  "scale_format": [
    "e8m0"
  ],
  "scale_calculation_mode": [
    "even"
  ]
}