{
  "family": "pocket_tts_small",
  "display_name": "PocketTTS Small",
  "description": "English PocketTTS Small 18m, 7m, and 2m models, with regular and ReFL variants, for text-to-speech and voice cloning.",
  "category": "tts",
  "status": "experimental",
  "tasks": [
    "tts",
    "clone"
  ],
  "modes": [
    "offline",
    "streaming"
  ],
  "languages": [
    "en"
  ],
  "options": {
    "request": [
      {
        "name": "frames_after_eos",
        "type": "int",
        "description": "Extra frames after EOS; -1 uses the model-derived value.",
        "required": false,
        "min": -1,
        "default": -1
      },
      {
        "name": "max_frames",
        "type": "int",
        "description": "Maximum generated audio frames; zero estimates the cap from text length. Captured during session preparation.",
        "required": false,
        "min": 0,
        "default": 0
      },
      {
        "name": "text_chunk_tokens",
        "type": "int",
        "description": "Tokenizer tokens per text chunk; must be positive, default 50. Set during preparation and overridable per request.",
        "required": false,
        "min": 1,
        "default": 50
      },
      {
        "name": "temperature",
        "type": "float",
        "description": "Positive flow noise temperature; defaults to the model configuration. Captured during session preparation.",
        "required": false
      },
      {
        "name": "seed",
        "type": "int",
        "description": "Noise seed; omitted chooses a random seed. Captured during session preparation.",
        "required": false,
        "min": 0,
        "max": 4294967295
      },
      {
        "name": "noise_clamp",
        "type": "float",
        "description": "Noise clamp; a negative value disables clamping. Captured during session preparation.",
        "required": false,
        "default": -1
      },
      {
        "name": "eos_threshold",
        "type": "float",
        "description": "EOS detection threshold; defaults to the checkpoint recommendation.",
        "required": false,
        "default": -1
      },
      {
        "name": "noise_file",
        "type": "path",
        "description": "Precomputed noise schedule file. Captured during session preparation.",
        "required": false
      },
      {
        "name": "voice_embedding_file",
        "type": "path",
        "description": "Precomputed voice-state embedding path. Captured during session preparation.",
        "required": false
      },
      {
        "name": "truncate_clone_audio",
        "type": "bool",
        "description": "Truncate voice clone audio. Captured during session preparation.",
        "required": false,
        "default": false
      },
      {
        "name": "text_chunk_size",
        "type": "int",
        "description": "Optional per-request character chunk size; otherwise use sentence/token budgeting.",
        "required": false
      },
      {
        "name": "export_voice_state_file",
        "type": "path",
        "description": "Export the prepared voice state during session preparation.",
        "required": false
      }
    ],
    "session": [
      {
        "name": "attention",
        "type": "enum",
        "description": "GGML acoustic attention implementation.",
        "required": false,
        "values": [
          "auto",
          "eager",
          "flash"
        ],
        "default": "auto"
      },
      {
        "name": "kv_cache_type",
        "type": "enum",
        "description": "GGML acoustic KV cache storage. Native CPU flow requires f32.",
        "required": false,
        "values": [
          "f32",
          "f16"
        ],
        "default": "f32"
      },
      {
        "name": "voice_state_cache_slots",
        "type": "int",
        "description": "Cached prepared voice states; zero disables reuse.",
        "required": false,
        "min": 0,
        "default": 4
      },
      {
        "name": "offline_graph_capacity_mode",
        "type": "enum",
        "description": "Offline graph capacity policy; CPU/Metal default tiered, other backends double.",
        "required": false,
        "values": [
          "fixed",
          "double",
          "tiered",
          "grow"
        ]
      },
      {
        "name": "prompt_graph_capacity",
        "type": "int",
        "description": "Optional reserved prompt capacity.",
        "required": false,
        "min": 1
      },
      {
        "name": "generation_graph_capacity",
        "type": "int",
        "description": "Optional reserved generation capacity.",
        "required": false,
        "min": 1
      },
      {
        "name": "weight_type",
        "type": "enum",
        "description": "Shared matmul weight storage; component matmul settings inherit it. Convolution storage remains independently native unless overridden.",
        "required": false,
        "preset": "weight_type_full"
      },
      {
        "name": "conv_weight_type",
        "type": "enum",
        "description": "Convolution weight storage.",
        "required": false,
        "preset": "weight_type_conv",
        "default": "native"
      },
      {
        "name": "weight_context_mb",
        "type": "int",
        "description": "Context or graph arena size in MiB.",
        "required": false,
        "min": 1,
        "default": 64
      },
      {
        "name": "flow_weight_context_mb",
        "type": "int",
        "description": "Flow weight context in MiB; inherits pocket_tts_small.weight_context_mb when omitted.",
        "required": false,
        "min": 1
      },
      {
        "name": "mimi_encoder_weight_context_mb",
        "type": "int",
        "description": "Mimi encoder weight context in MiB; inherits pocket_tts_small.weight_context_mb when omitted.",
        "required": false,
        "min": 1
      },
      {
        "name": "mimi_decoder_weight_context_mb",
        "type": "int",
        "description": "Mimi decoder weight context in MiB; inherits pocket_tts_small.weight_context_mb when omitted.",
        "required": false,
        "min": 1
      },
      {
        "name": "flow_weights_view_context_mb",
        "type": "int",
        "description": "Context or graph arena size in MiB.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "flow_step_graph_arena_mb",
        "type": "int",
        "description": "Flow-step host graph-metadata arena size in MiB, not device workspace.",
        "required": false,
        "min": 1,
        "default": 256
      },
      {
        "name": "mimi_encoder_graph_arena_mb",
        "type": "int",
        "description": "Mimi encoder host graph-metadata arena size in MiB, not device workspace.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "mimi_conv_graph_arena_mb",
        "type": "int",
        "description": "Mimi convolution host graph-metadata arena size in MiB, not device workspace.",
        "required": false,
        "min": 1,
        "default": 32
      },
      {
        "name": "mimi_transformer_graph_arena_mb",
        "type": "int",
        "description": "Mimi transformer host graph-metadata arena size in MiB, not device workspace.",
        "required": false,
        "min": 1,
        "default": 96
      },
      {
        "name": "mimi_tail_graph_arena_mb",
        "type": "int",
        "description": "Mimi tail host graph-metadata arena size in MiB, not device workspace.",
        "required": false,
        "min": 1,
        "default": 512
      },
      {
        "name": "mimi_chunk_frames",
        "type": "int",
        "description": "Latent frames per bounded causal waveform-decoder block.",
        "required": false,
        "min": 1,
        "default": 4
      },
      {
        "name": "mimi_stage2_chunk_frames",
        "type": "int",
        "description": "Second-stage decoder chunk frames.",
        "required": false,
        "min": 1,
        "default": 900
      },
      {
        "name": "use_full_mimi",
        "type": "bool",
        "description": "Use a full-utterance waveform graph instead of bounded causal blocks. This can use more device memory.",
        "required": false,
        "default": false
      },
      {
        "name": "exec_plan",
        "type": "string",
        "description": "Component execution plan. Flow defaults to ggml; native is an experimental CPU F32 implementation with an optional threads parameter.",
        "required": false
      }
    ],
    "load": []
  },
  "capabilities": {
    "clone": [
      "speaker_reference"
    ]
  },
  "runtime": {
    "tags": [
      "gguf"
    ]
  },
  "ui": {
    "recommended_package": "pocket_tts_small_english_18m_refl_f32",
    "default_voice": "alba",
    "builtin_voices": [
      "alba"
    ],
    "tags": [
      "TTS",
      "Clone",
      "GGUF"
    ],
    "docs": [
      "docs/models/pocket_tts_small.md",
      "docs/gguf.md"
    ]
  },
  "package_defaults": {
    "download": {
      "kind": "huggingface_snapshot",
      "repo": "audio-cpp/audio.cpp-gguf",
      "revision": "main",
      "gated": false
    }
  },
  "packages": [
    {
      "id": "pocket_tts_small_english_18m_refl_f32",
      "display_name": "PocketTTS Small English 18m ReFL F32 GGUF",
      "format": "gguf",
      "precision": "f32",
      "target_directory": "PocketTTS-Small-GGUF/english-18m-refl",
      "files": [
        "PocketTTS-Small-GGUF/english-18m-refl/english-18m-refl-f32.gguf",
        "PocketTTS-Small-GGUF/english-18m-refl/embeddings/alba.safetensors"
      ],
      "default": true,
      "strip_prefix": "PocketTTS-Small-GGUF/english-18m-refl"
    },
    {
      "id": "pocket_tts_small_english_18m_f32",
      "display_name": "PocketTTS Small English 18m F32 GGUF",
      "format": "gguf",
      "precision": "f32",
      "target_directory": "PocketTTS-Small-GGUF/english-18m",
      "files": [
        "PocketTTS-Small-GGUF/english-18m/english-18m-f32.gguf",
        "PocketTTS-Small-GGUF/english-18m/embeddings/alba.safetensors"
      ],
      "strip_prefix": "PocketTTS-Small-GGUF/english-18m"
    },
    {
      "id": "pocket_tts_small_english_7m_refl_f32",
      "display_name": "PocketTTS Small English 7m ReFL F32 GGUF",
      "format": "gguf",
      "precision": "f32",
      "target_directory": "PocketTTS-Small-GGUF/english-7m-refl",
      "files": [
        "PocketTTS-Small-GGUF/english-7m-refl/english-7m-refl-f32.gguf",
        "PocketTTS-Small-GGUF/english-7m-refl/embeddings/alba.safetensors"
      ],
      "strip_prefix": "PocketTTS-Small-GGUF/english-7m-refl"
    },
    {
      "id": "pocket_tts_small_english_7m_f32",
      "display_name": "PocketTTS Small English 7m F32 GGUF",
      "format": "gguf",
      "precision": "f32",
      "target_directory": "PocketTTS-Small-GGUF/english-7m",
      "files": [
        "PocketTTS-Small-GGUF/english-7m/english-7m-f32.gguf",
        "PocketTTS-Small-GGUF/english-7m/embeddings/alba.safetensors"
      ],
      "strip_prefix": "PocketTTS-Small-GGUF/english-7m"
    },
    {
      "id": "pocket_tts_small_english_2m_refl_f32",
      "display_name": "PocketTTS Small English 2m ReFL F32 GGUF",
      "format": "gguf",
      "precision": "f32",
      "target_directory": "PocketTTS-Small-GGUF/english-2m-refl",
      "files": [
        "PocketTTS-Small-GGUF/english-2m-refl/english-2m-refl-f32.gguf",
        "PocketTTS-Small-GGUF/english-2m-refl/embeddings/alba.safetensors"
      ],
      "strip_prefix": "PocketTTS-Small-GGUF/english-2m-refl"
    },
    {
      "id": "pocket_tts_small_english_2m_f32",
      "display_name": "PocketTTS Small English 2m F32 GGUF",
      "format": "gguf",
      "precision": "f32",
      "target_directory": "PocketTTS-Small-GGUF/english-2m",
      "files": [
        "PocketTTS-Small-GGUF/english-2m/english-2m-f32.gguf",
        "PocketTTS-Small-GGUF/english-2m/embeddings/alba.safetensors"
      ],
      "strip_prefix": "PocketTTS-Small-GGUF/english-2m"
    }
  ],
  "sources": [
    {
      "format": "gguf",
      "roots": {
        "model": ".",
        "weights": "$gguf"
      },
      "files": {
        "tokenizer": "model:tokenizer.model",
        "config": "model:config.yaml"
      },
      "tensors": {
        "weights": "weights:"
      }
    },
    {
      "format": "safetensors",
      "roots": {
        "model": "."
      },
      "files": {
        "tokenizer": "model:tokenizer.model",
        "config": "model:config.yaml"
      },
      "tensors": {
        "weights": "model:model.safetensors"
      }
    }
  ],
  "schema_version": 1,
  "dependencies": []
}