voice_audio_multi/model_specs/audio8_asr.json

{
  "family": "audio8_asr",
  "display_name": "Audio8-ASR-0.1B",
  "description": "Audio8-ASR-0.1B compact multilingual ASR (en/zh/yue/ja/ko/fr/de): Qwen3-ASR audio encoder with an MLP-tower adapter and an 8-layer Qwen2-style decoder. 30-second offline transcription windows. Checkpoint license: CC-BY-NC-4.0; GGUF conversion is local-only, no public redistribution is approved.",
  "category": "asr",
  "status": "community",
  "tasks": [
    "asr"
  ],
  "modes": [
    "offline"
  ],
  "languages": [
    "en",
    "zh",
    "yue",
    "ja",
    "ko",
    "fr",
    "de"
  ],
  "capabilities": {
    "asr": []
  },
  "options": {
    "request": [],
    "session": [
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Decoder and adapter weight storage type.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "audio_encoder_weight_type",
        "type": "enum",
        "description": "Audio encoder weight storage type (quantized storage is rejected for the encoder; already-quantized GGUF tensors load as native).",
        "values": [
          "native",
          "f32",
          "f16"
        ],
        "required": false,
        "default": "native"
      },
      {
        "name": "encoder_graph_arena_mb",
        "type": "int",
        "description": "Audio encoder graph arena size in MB. Must be positive; sufficient capacity depends on the graph.",
        "required": false,
        "min": 1,
        "default": 128
      },
      {
        "name": "projector_graph_arena_mb",
        "type": "int",
        "description": "Adapter (MLP tower + pooling + projector) graph arena size in MB. Must be positive; sufficient capacity depends on the graph.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "prefill_graph_arena_mb",
        "type": "int",
        "description": "Decoder prefill graph arena size in MB. Must be positive; sufficient capacity depends on the graph.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "decode_graph_arena_mb",
        "type": "int",
        "description": "Decoder step decode graph arena size in MB. Must be positive; sufficient capacity depends on the graph.",
        "required": false,
        "min": 1,
        "default": 256
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf",
      "server",
      "cuda",
      "metal",
      "cpu"
    ]
  },
  "ui": {
    "recommended_package": "audio8_asr_0_1b_q8_0",
    "tags": [
      "ASR",
      "GGUF"
    ],
    "docs": [
      "docs/asr.md",
      "docs/models/audio8_asr.md"
    ],
    "summary": "Audio8-ASR-0.1B compact multilingual offline transcription."
  },
  "package_defaults": {
    "download": {
      "kind": "unsupported",
      "reason": "CC-BY-NC-4.0 checkpoint: convert locally with audiocpp_gguf; no public audio.cpp GGUF distribution is approved."
    }
  },
  "packages": [
    {
      "id": "audio8_asr_0_1b_q8_0",
      "display_name": "Audio8-ASR-0.1B Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "Audio8-ASR-0.1B-GGUF",
      "files": [
        "Audio8-ASR-0.1B-GGUF/audio8-asr-0.1b-q8_0.gguf"
      ],
      "strip_prefix": "Audio8-ASR-0.1B-GGUF"
    },
    {
      "id": "audio8_asr_0_1b_f16",
      "display_name": "Audio8-ASR-0.1B F16 GGUF",
      "format": "gguf",
      "precision": "f16",
      "target_directory": "Audio8-ASR-0.1B-GGUF",
      "files": [
        "Audio8-ASR-0.1B-GGUF/audio8-asr-0.1b-f16.gguf"
      ],
      "strip_prefix": "Audio8-ASR-0.1B-GGUF"
    },
    {
      "id": "audio8_asr_0_1b_safetensors",
      "display_name": "Audio8-ASR-0.1B HF Safetensors",
      "format": "safetensors",
      "precision": "bfloat16",
      "target_directory": "Audio8-ASR-0.1B-hf",
      "files": [
        "config.json",
        "generation_config.json",
        "preprocessor_config.json",
        "tokenizer.json",
        "tokenizer_config.json",
        "vocab.json",
        "merges.txt",
        "model.safetensors"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "Audio8/Audio8-ASR-0.1B",
        "revision": "main",
        "gated": false
      }
    }
  ],
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "generation_config": "model:generation_config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "preprocessor_config": "model:preprocessor_config.json"
      },
      "optional_files": {
        "vocab": "model:vocab.json",
        "merges": "model:merges.txt",
        "tokenizer_json": "model:tokenizer.json"
      },
      "tensors": {
        "weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "generation_config": "model:generation_config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "preprocessor_config": "model:preprocessor_config.json"
      },
      "optional_files": {
        "vocab": "model:vocab.json",
        "merges": "model:merges.txt",
        "tokenizer_json": "model:tokenizer.json"
      },
      "tensors": {
        "weights": "model:model.safetensors"
      }
    }
  ]
}