jamiepine/voicebox · error · ValueError

Unknown model size: {model_size}

Error message

Unknown model size: {model_size}

What it means

Raised by MlxTTSBackend._get_model_path when model_size is not a key of mlx_model_map, whose only entries are "1.7B" (mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16) and "0.6B" (mlx-community/Qwen3-TTS-12Hz-0.6B-Base-bf16). The map resolves the Apple-Silicon MLX weight repo on HuggingFace Hub, so an unknown size cannot be turned into a download path.

Source

Thrown at backend/backends/mlx_backend.py:53

        return self.model is not None

    def _get_model_path(self, model_size: str) -> str:
        """
        Get the MLX model path.

        Args:
            model_size: Model size (1.7B or 0.6B)

        Returns:
            HuggingFace Hub model ID for MLX
        """
        mlx_model_map = {
            "1.7B": "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16",
            "0.6B": "mlx-community/Qwen3-TTS-12Hz-0.6B-Base-bf16",
        }

        if model_size not in mlx_model_map:
            raise ValueError(f"Unknown model size: {model_size}")

        hf_model_id = mlx_model_map[model_size]
        logger.info("Will download MLX model from HuggingFace Hub: %s", hf_model_id)

        return hf_model_id

    def _is_model_cached(self, model_size: str) -> bool:
        return is_model_cached(
            self._get_model_path(model_size),
            weight_extensions=(".safetensors", ".bin", ".npz"),
        )

    async def load_model_async(self, model_size: Optional[str] = None):
        """
        Lazy load the MLX TTS model.

        Args:
            model_size: Model size to load (1.7B or 0.6B)

View on GitHub (pinned to 51f49dea19)

Solutions

1. Pass one of the supported literals: "1.7B" (default, higher quality) or "0.6B" (faster, lighter).
2. Normalize caller input before forwarding: value.strip() and uppercase the size token.
3. If "4B" is required it is not supported by the MLX TTS engine — pick a different engine instead of retrying.

Example fix

// before
await mlx_backend.load_model_async(model_size="4B")
// after
await mlx_backend.load_model_async(model_size="1.7B")

MLX_TTS_SIZES = {"1.7B", "0.6B"}
if model_size not in MLX_TTS_SIZES:
    raise ValueError(f"model_size must be one of {sorted(MLX_TTS_SIZES)}; got {model_size!r}")
await mlx_backend.load_model_async(model_size=model_size)

def is_mlx_tts_size(value: str) -> bool:
    return isinstance(value, str) and value in {"1.7B", "0.6B"}

try:
    path = mlx_backend._get_model_path(model_size)
except ValueError as exc:
    if "Unknown model size" in str(exc):
        # fall back to a known-good size or surface to the user
        model_size = "1.7B"
        path = mlx_backend._get_model_path(model_size)
    else:
        raise

When it happens

Trigger: Calling load_model_async(model_size=...) or _get_model_path(model_size) with anything other than the literal strings "1.7B" or "0.6B" — e.g. "4B", "1.7b", "base", "1.7B\n".

Common situations: Reusing a size string valid for the Qwen LLM backend (which also accepts "4B") against the TTS backend; case mismatch from user input; trailing whitespace from config; copying the engine default of a different engine.

Related errors

• Unknown model size: {model_size}
• Unknown model size: {model_size}
• Unknown Qwen3 size: {model_size}
• Invalid STT model '{model_size}'. Must be one of: {', '.join
• RPC ${method}: ${json.error.message}

AI-assisted analysis of jamiepine/voicebox@51f49dea19 (2026-08-12). Data as JSON: /api/errors/abbc06fd3492caef. Report an issue: GitHub.