{
  "schema_version": 1,
  "family": "sortformer_diar",
  "display_name": "Sortformer Diarization",
  "description": "NVIDIA Transformer-based end-to-end speaker diarization model trained primarily on English speech, predicting speaker labels directly from audio and resolving speaker ordering by arrival time for up to four speakers.",
  "category": "speech_analysis",
  "status": "supported",
  "tasks": [
    "diar"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "en"
  ],
  "capabilities": {
    "diar": [
      "speaker_turns"
    ]
  },
  "dependencies": [],
  "options": {
    "request": [
      {
        "name": "speaker_threshold",
        "type": "float",
        "description": "Speaker activity probability threshold used when decoding speaker turns; default 0.5.",
        "required": false,
        "min": 0.0,
        "max": 1.0,
        "default": 0.5
      },
      {
        "name": "speaker_min_frames",
        "type": "int",
        "description": "Minimum decoded segment length in model output frames; default 0 disables short-segment filtering.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "speaker_pad_frames",
        "type": "int",
        "description": "Pad each decoded speaker segment by this many model output frames before filtering and merging; default 0.",
        "required": false,
        "min": 0,
        "default": 0
      }
    ],
    "session": [
      {
        "name": "graph_arena_mb",
        "type": "int",
        "description": "Inference graph arena size in MiB; default 512.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "weight_context_mb",
        "type": "int",
        "description": "Weight context size in MiB; default 128.",
        "required": false,
        "min": 1,
        "default": 128
      },
      {
        "name": "session_len_sec",
        "type": "float",
        "description": "Base offline graph context length in seconds; must be positive, default 20.",
        "required": false,
        "min": 0.0,
        "default": 20.0
      },
      {
        "name": "graph_capacity_mode",
        "type": "enum",
        "description": "Offline graph capacity policy; defaults to tiered on host-graph backends and fixed otherwise.",
        "values": [
          "fixed",
          "tiered",
          "grow",
          "double"
        ],
        "required": false
      },
      {
        "name": "speaker_threshold",
        "type": "float",
        "description": "Default speaker activity probability threshold for decoded speaker turns; default 0.5.",
        "required": false,
        "min": 0.0,
        "max": 1.0,
        "default": 0.5
      },
      {
        "name": "speaker_min_frames",
        "type": "int",
        "description": "Default minimum decoded segment length in model output frames; default 0 disables short-segment filtering.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "speaker_pad_frames",
        "type": "int",
        "description": "Default decoded segment padding in model output frames; default 0.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "weight_type",
        "type": "enum",
        "description": "All weight storage type; default f32.",
        "preset": "weight_type_full",
        "required": false,
        "default": "f32"
      },
      {
        "name": "matmul_weight_type",
        "type": "enum",
        "description": "Matmul weight storage type; defaults to weight_type when set, otherwise f32.",
        "preset": "weight_type_full",
        "required": false
      },
      {
        "name": "conv_weight_type",
        "type": "enum",
        "description": "Convolution weight storage type; defaults to weight_type when set, otherwise f32.",
        "preset": "weight_type_full",
        "required": false
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "ui": {
    "recommended_package": "sortformer_diar_4spk_v1_q8_0",
    "tags": [
      "Diar",
      "GGUF"
    ],
    "docs": [
      "docs/audio_tools.md",
      "docs/gguf.md"
    ]
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "sortformer_diar_4spk_v1_q8_0",
      "display_name": "Sortformer Diar 4spk v1 Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Sortformer-Diar-4spk-v1-GGUF",
      "files": [
        "Sortformer-Diar-4spk-v1-GGUF/sortformer-diar-4spk-v1-q8_0.gguf"
      ],
      "strip_prefix": "Sortformer-Diar-4spk-v1-GGUF"
    },
    {
      "id": "sortformer_diar_4spk_v1_f16",
      "display_name": "Sortformer Diar 4spk v1 F16 GGUF",
      "format": "gguf",
      "precision": "f16",
      "target_directory": "Sortformer-Diar-4spk-v1-GGUF",
      "files": [
        "Sortformer-Diar-4spk-v1-GGUF/sortformer-diar-4spk-v1-f16.gguf"
      ],
      "strip_prefix": "Sortformer-Diar-4spk-v1-GGUF"
    },
    {
      "id": "sortformer_diar_4spk_v1_safetensors",
      "display_name": "Sortformer Diar 4spk v1 Safetensors",
      "format": "safetensors",
      "precision": "native",
      "target_directory": "diar_sortformer_4spk-v1",
      "files": [
        "config.json",
        "model.safetensors",
        "processor_config.json"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "nvidia/diar_sortformer_4spk-v1"
      }
    }
  ],
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "processor": "model:processor_config.json"
      },
      "tensors": {
        "weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "processor": "model:processor_config.json"
      },
      "tensors": {
        "weights": "model:model.safetensors"
      }
    }
  ]
}