voice_audio_multi/model_specs/audio_flamingo.json

{
  "schema_version": 1,
  "family": "audio_flamingo",
  "display_name": "Audio Flamingo 3 / Next / Music Flamingo",
  "description": "Audio Flamingo 3, Audio Flamingo Next, and Music Flamingo for transcription, captioning, and audio questions, with optional PEFT LoRA and SpotSound temporal grounding.",
  "category": "asr",
  "status": "supported",
  "tasks": [
    "asr"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "en"
  ],
  "capabilities": {},
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "options": {
    "request": [
      {
        "name": "instruct",
        "type": "string",
        "default": "Transcribe the input speech.",
        "required": false,
        "description": "Question or instruction about the input audio."
      },
      {
        "name": "template_name",
        "type": "enum",
        "values": [
          "default",
          "temporal_grounding"
        ],
        "default": "default",
        "required": false,
        "description": "Prompt layout. temporal_grounding uses SpotSound timestamp-interleaved audio tokens and wraps instruct or text as an event query; requires Audio Flamingo 3 and at least one second of audio. Use with its matching adapter."
      },
      {
        "name": "max_tokens",
        "type": "int",
        "min": 1,
        "required": false,
        "description": "Maximum generated response tokens. Defaults to checkpoint generation configuration (2048 fallback); prompt plus output budget must fit model positions."
      },
      {
        "name": "do_sample",
        "type": "bool",
        "default": false,
        "required": false,
        "description": "Sample tokens instead of greedy decoding."
      },
      {
        "name": "temperature",
        "type": "float",
        "default": 1,
        "required": false,
        "description": "Sampling temperature. Used only when do_sample=true; must be positive when sampling."
      },
      {
        "name": "top_k",
        "type": "int",
        "default": 50,
        "required": false,
        "description": "Sampling candidate count; zero disables top-k filtering. Used only when sampling; nonpositive values disable top-k filtering."
      },
      {
        "name": "top_p",
        "type": "float",
        "default": 1,
        "required": false,
        "description": "Nucleus sampling probability. Used only when sampling; zero is accepted and values >=1 disable nucleus filtering."
      },
      {
        "name": "repetition_penalty",
        "type": "float",
        "default": 1,
        "required": false,
        "description": "Token repetition penalty. Must be positive."
      },
      {
        "name": "seed",
        "type": "int",
        "default": 42,
        "required": false,
        "description": "Sampling seed. Parsed as a signed integer and converted to uint32."
      }
    ],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "preset": "weight_type_full",
        "default": "native",
        "required": false,
        "description": "Weight storage type; native preserves the packaged tensor types."
      },
      {
        "name": "lora",
        "type": "path",
        "required": false,
        "description": "Matching PEFT LoRA adapter directory or checkpoint file; merged at session creation. Keep adapter_config.json beside the weights."
      },
      {
        "name": "lora_scale",
        "type": "float",
        "default": 1,
        "required": false,
        "description": "Multiplier for the adapter's alpha/r scaling."
      }
    ],
    "load": []
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/Audio-Flamingo-GGUF",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "audio_flamingo_next_q4_k",
      "display_name": "Audio Flamingo Next GGUF Q4_K",
      "default": true,
      "format": "gguf",
      "precision": "q4_k",
      "target_directory": "Audio-Flamingo-GGUF",
      "files": [
        "audio-flamingo-next-q4_k.gguf"
      ]
    },
    {
      "id": "audio_flamingo_next_q8_0",
      "display_name": "Audio Flamingo Next GGUF Q8_0",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Audio-Flamingo-GGUF",
      "files": [
        "audio-flamingo-next-q8_0.gguf"
      ]
    },
    {
      "id": "audio_flamingo_next_bf16",
      "display_name": "Audio Flamingo Next GGUF BF16",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "Audio-Flamingo-GGUF",
      "files": [
        "audio-flamingo-next-bf16.gguf"
      ]
    },
    {
      "id": "audio_flamingo_3_q4_k",
      "display_name": "Audio Flamingo 3 GGUF Q4_K",
      "format": "gguf",
      "precision": "q4_k",
      "target_directory": "Audio-Flamingo-GGUF",
      "files": [
        "audio-flamingo-3-q4_k.gguf"
      ]
    },
    {
      "id": "audio_flamingo_3_q8_0",
      "display_name": "Audio Flamingo 3 GGUF Q8_0",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Audio-Flamingo-GGUF",
      "files": [
        "audio-flamingo-3-q8_0.gguf"
      ]
    },
    {
      "id": "audio_flamingo_3_bf16",
      "display_name": "Audio Flamingo 3 GGUF BF16",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "Audio-Flamingo-GGUF",
      "files": [
        "audio-flamingo-3-bf16.gguf"
      ]
    },
    {
      "id": "music_flamingo_2601_q8_0",
      "display_name": "Music Flamingo 2601 GGUF Q8_0",
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Audio-Flamingo-GGUF",
      "files": [
        "music-flamingo-2601-q8_0.gguf"
      ]
    }
  ],
  "dependencies": [],
  "ui": {
    "tags": [
      "GGUF"
    ],
    "docs": [
      "docs/models/audio_flamingo.md"
    ],
    "recommended_package": "audio_flamingo_next_q4_k"
  },
  "sources": [
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "generation_config": "model:generation_config.json",
        "processor_config": "model:processor_config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_json": "model:tokenizer.json",
        "chat_template": "model:chat_template.jinja"
      },
      "tensors": {
        "weights": {
          "source": "model:model.safetensors"
        }
      }
    },
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "generation_config": "model:generation_config.json",
        "processor_config": "model:processor_config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_json": "model:tokenizer.json",
        "chat_template": "model:chat_template.jinja"
      },
      "tensors": {
        "weights": {
          "source": "weights:",
          "prefix": "weights"
        }
      }
    }
  ]
}