{
  "family": "voxcpm2",
  "display_name": "VoxCPM2",
  "description": "OpenBMB tokenizer-free TTS model supporting 30 languages and 9 Chinese dialects, with 48 kHz output, natural-language voice design, controllable short-reference voice cloning, and expressive style guidance.",
  "category": "tts",
  "status": "supported",
  "tasks": [
    "tts",
    "clone",
    "design"
  ],
  "modes": [
    "offline",
    "streaming"
  ],
  "languages": [
    "ar",
    "my",
    "zh",
    "zh dialects",
    "da",
    "nl",
    "en",
    "fi",
    "fr",
    "de",
    "el",
    "he",
    "hi",
    "id",
    "it",
    "ja",
    "km",
    "ko",
    "lo",
    "ms",
    "no",
    "pl",
    "pt",
    "ru",
    "es",
    "sw",
    "sv",
    "tl",
    "th",
    "tr",
    "vi"
  ],
  "capabilities": {
    "clone": [
      "speaker_reference"
    ],
    "design": [
      "voice_design"
    ]
  },
  "options": {
    "request": [
      {
        "name": "reference_text",
        "type": "string",
        "description": "Transcript for prompt/reference audio in ultimate-cloning and continuation-style prompts.",
        "required": false
      },
      {
        "name": "min_tokens",
        "type": "int",
        "description": "Minimum generated acoustic patches before EOS can stop generation; default 2.",
        "required": false,
        "min": 0,
        "default": 2
      },
      {
        "name": "max_tokens",
        "type": "int",
        "description": "Maximum generated acoustic patches; default 4096, and 0 requests the model config max_length.",
        "required": false,
        "min": 0,
        "default": 4096
      },
      {
        "name": "num_inference_steps",
        "type": "int",
        "description": "Local DiT/CFM inference steps per generated patch; default 10.",
        "required": false,
        "min": 1,
        "default": 10
      },
      {
        "name": "guidance_scale",
        "type": "float",
        "description": "Classifier-free guidance scale for local DiT/CFM generation; default 2.0.",
        "required": false,
        "min": 0.0,
        "default": 2.0
      },
      {
        "name": "retry_badcase",
        "type": "bool",
        "description": "Retry over-long offline generations detected by the patch/text ratio; default true, but streaming requests must set false.",
        "required": false,
        "default": true
      },
      {
        "name": "retry_badcase_max_times",
        "type": "int",
        "description": "Maximum retry attempts for bad-case detection; default 3.",
        "required": false,
        "min": 1,
        "default": 3
      },
      {
        "name": "retry_badcase_ratio_threshold",
        "type": "float",
        "description": "Generated-patch to target-text-token ratio threshold for bad-case detection; must be positive, default 6.0.",
        "required": false,
        "default": 6.0
      },
      {
        "name": "seed",
        "type": "int",
        "description": "CFM noise seed; default 1234.",
        "required": false,
        "min": 0,
        "default": 1234
      },
      {
        "name": "noise_path",
        "type": "path",
        "description": "F32 CFM noise file for deterministic acoustic patch replay.",
        "required": false
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Maximum UTF-8 codepoints per offline text chunk; default 2048.",
        "required": false,
        "min": 1,
        "default": 2048
      },
      {
        "name": "text_chunk_mode",
        "type": "enum",
        "description": "Text chunking mode; default tag_aware.",
        "preset": "text_chunk_mode_full",
        "required": false,
        "default": "tag_aware"
      }
    ],
    "session": [
      {
        "name": "mem_saver",
        "type": "bool",
        "description": "Use tighter graph workspaces and release request runtime graphs; default false.",
        "required": false,
        "default": false
      },
      {
        "name": "prompt_cache_slots",
        "type": "int",
        "description": "Prompt and prompt-audio embedding cache slots; default 1.",
        "required": false,
        "min": 0,
        "default": 1
      },
      {
        "name": "weight_type",
        "type": "enum",
        "description": "MiniCPM, local encoder, projection, and DiT weight storage type; default native.",
        "preset": "weight_type_full",
        "required": false,
        "default": "native"
      },
      {
        "name": "audiovae_weight_type",
        "type": "enum",
        "description": "AudioVAE weight storage type; default f32.",
        "preset": "weight_type_full",
        "required": false,
        "default": "f32"
      },
      {
        "name": "weight_context_mb",
        "type": "int",
        "description": "Shared generator weight context size in MiB; default 3072.",
        "required": false,
        "min": 1,
        "default": 3072
      },
      {
        "name": "text_embedding_graph_arena_mb",
        "type": "int",
        "description": "Text embedding graph arena size in MiB; default 64.",
        "required": false,
        "min": 1,
        "default": 64
      },
      {
        "name": "lm_step_graph_arena_mb",
        "type": "int",
        "description": "MiniCPM LM step graph arena size in MiB; default 1024.",
        "required": false,
        "min": 1,
        "default": 1024
      },
      {
        "name": "projection_graph_arena_mb",
        "type": "int",
        "description": "Step projection graph arena size in MiB; default 256.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "local_encoder_graph_arena_mb",
        "type": "int",
        "description": "Local encoder graph arena size in MiB; default 512.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "dit_graph_arena_mb",
        "type": "int",
        "description": "DiT estimator graph arena size in MiB; default 1024.",
        "required": false,
        "min": 1,
        "default": 1024
      },
      {
        "name": "audiovae_weight_context_mb",
        "type": "int",
        "description": "AudioVAE weight context size in MiB; default 768.",
        "required": false,
        "min": 1,
        "default": 768
      },
      {
        "name": "audiovae_graph_arena_mb",
        "type": "int",
        "description": "AudioVAE decoder graph arena size in MiB; default 1024.",
        "required": false,
        "min": 1,
        "default": 1024
      },
      {
        "name": "audiovae_encoder_graph_arena_mb",
        "type": "int",
        "description": "AudioVAE prompt encoder graph arena size in MiB; default 1024.",
        "required": false,
        "min": 1,
        "default": 1024
      },
      {
        "name": "audiovae_latent_capacity",
        "type": "int",
        "description": "AudioVAE decoder latent-frame graph capacity; default 0 fits each request.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "audiovae_encoder_sample_capacity",
        "type": "int",
        "description": "AudioVAE prompt encoder sample capacity; must be positive and divisible by the encoder stride, default 240000.",
        "required": false,
        "min": 1,
        "default": 240000
      }
    ],
    "load": []
  },
  "runtime": {
    "tags": [
      "gguf",
      "stream"
    ]
  },
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_json": "model:tokenizer.json",
        "special_tokens_map": "model:special_tokens_map.json"
      },
      "tensors": {
        "weights": {
          "source": "weights:",
          "prefix": "weights"
        },
        "audiovae_weights": {
          "source": "weights:",
          "prefix": "audiovae_weights"
        }
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "config": "model:config.json",
        "tokenizer_config": "model:tokenizer_config.json",
        "tokenizer_json": "model:tokenizer.json",
        "special_tokens_map": "model:special_tokens_map.json"
      },
      "tensors": {
        "weights": "model:model.safetensors",
        "audiovae_weights": "model:audiovae.safetensors"
      }
    }
  ],
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "voxcpm2_q8_0",
      "display_name": "VoxCPM2 Q8_0 GGUF",
      "default": true,
      "format": "gguf",
      "precision": "q8_0",
      "target_directory": "VoxCPM2-GGUF",
      "files": [
        "VoxCPM2-GGUF/voxcpm2-q8_0.gguf"
      ],
      "strip_prefix": "VoxCPM2-GGUF"
    },
    {
      "id": "voxcpm2_bf16",
      "display_name": "VoxCPM2 BF16 GGUF",
      "format": "gguf",
      "precision": "bf16",
      "target_directory": "VoxCPM2-GGUF",
      "files": [
        "VoxCPM2-GGUF/voxcpm2-bf16.gguf"
      ],
      "strip_prefix": "VoxCPM2-GGUF"
    },
    {
      "id": "voxcpm2_orig",
      "display_name": "VoxCPM2 Original-Dtype GGUF",
      "format": "gguf",
      "precision": "orig",
      "target_directory": "VoxCPM2-GGUF",
      "files": [
        "VoxCPM2-GGUF/voxcpm2-orig.gguf"
      ],
      "strip_prefix": "VoxCPM2-GGUF"
    },
    {
      "id": "voxcpm2_safetensors",
      "display_name": "VoxCPM2 Safetensors",
      "format": "safetensors",
      "precision": "native",
      "target_directory": "VoxCPM2",
      "files": [
        "config.json",
        "model.safetensors",
        "tokenizer.json",
        "tokenizer_config.json"
      ],
      "download": {
        "kind": "huggingface_snapshot",
        "repo": "OpenBMB/VoxCPM2"
      }
    }
  ]
}