mlx_config.json

{
  "architecture": "EuroBertForTokenClassification",
  "format": "mlx",
  "implementation": "modeling_eurobert_mlx.py",
  "source_model": "feyninc/pulpie-orange-small",
  "variants": {
    "model-4bit.safetensors": {
      "description": "MLX affine weight quantized 4-bit variant.",
      "dtype": "quantized",
      "quantization": {
        "bits": 4,
        "group_size": 64,
        "mode": "affine"
      }
    },
    "model-8bit.safetensors": {
      "description": "MLX affine weight quantized 8-bit variant.",
      "dtype": "quantized",
      "quantization": {
        "bits": 8,
        "group_size": 64,
        "mode": "affine"
      }
    },
    "model-bf16.safetensors": {
      "description": "Native 16-bit BF16 MLX weights converted from the source safetensors.",
      "dtype": "bfloat16",
      "quantization": null
    }
  }
}