What causes "Unknown Qwen3 size: {model_size}" in jamiepine/voicebox?

jamiepine/voicebox · error · ValueError

Unknown Qwen3 size: {model_size}

Error message

Unknown Qwen3 size: {model_size}

What it means

Raised by the PyTorch Qwen LLM backend's _get_model_path when model_size is not a key of PYTORCH_HF_REPOS, which contains "0.6B" (Qwen/Qwen3-0.6B), "1.7B" (Qwen/Qwen3-1.7B), and "4B" (Qwen/Qwen3-4B). Note this is the Qwen3 LLM (text) model, not the TTS model — the valid set differs from the TTS backends.

Source

Thrown at backend/backends/qwen_llm_backend.py:77

class PyTorchQwenLLMBackend:
    """Qwen3 LLM backend using HuggingFace transformers."""

    def __init__(self, model_size: str = "0.6B"):
        self.model = None
        self.tokenizer = None
        self.model_size = model_size
        self._current_model_size: Optional[str] = None
        self.device = self._get_device()

    def _get_device(self) -> str:
        return get_torch_device(allow_xpu=True, allow_directml=True, allow_mps=True)

    def is_loaded(self) -> bool:
        return self.model is not None

    def _get_model_path(self, model_size: str) -> str:
        if model_size not in PYTORCH_HF_REPOS:
            raise ValueError(f"Unknown Qwen3 size: {model_size}")
        return PYTORCH_HF_REPOS[model_size]

    def _is_model_cached(self, model_size: str) -> bool:
        return is_model_cached(self._get_model_path(model_size))

    async def load_model(self, model_size: Optional[str] = None) -> None:
        if model_size is None:
            model_size = self.model_size

        if self.model is not None and self._current_model_size == model_size:
            return

        if self.model is not None and self._current_model_size != model_size:
            self.unload_model()

        await asyncio.to_thread(self._load_model_sync, model_size)

    def _load_model_sync(self, model_size: str) -> None:

View on GitHub (pinned to 51f49dea19)

Solutions

Pass "0.6B", "1.7B", or "4B".
Confirm you are calling the LLM backend, not a TTS backend, before assuming "4B" is valid.
Normalize/case-fold the value where it enters the system.

Example fix

// before
llm_backend.load_model(model_size="base")
// after
llm_backend.load_model(model_size="0.6B")

Defensive patterns

Strategy: validation

Validate before calling

from backend.backends.qwen_llm_backend import PYTORCH_HF_REPOS
if model_size not in PYTORCH_HF_REPOS:
    raise ValueError(f"size must be one of {sorted(PYTORCH_HF_REPOS)}")
await llm_backend.load_model(model_size=model_size)

Type guard

def is_qwen_llm_pytorch_size(value: str) -> bool:
    from backend.backends.qwen_llm_backend import PYTORCH_HF_REPOS
    return isinstance(value, str) and value in PYTORCH_HF_REPOS

Try / catch

try:
    await llm_backend.load_model(model_size=model_size)
except ValueError as exc:
    if "Unknown Qwen3 size" in str(exc):
        await llm_backend.load_model(model_size="0.6B")
    else:
        raise

Prevention

Remember this LLM backend also accepts "4B"; the TTS backends do not — keep them separate.
Validate against PYTORCH_HF_REPOS keys rather than a hardcoded list.
Normalize casing/whitespace before forwarding user input.

When it happens

Trigger: Calling load_model on the Qwen LLM backend with a string outside {"0.6B","1.7B","4B"} — e.g. "base", "1.7b", or a TTS-only token like "12Hz".

Common situations: Confusing the LLM size set with the TTS size set (TTS does not accept "4B"; this backend does); forwarding a whisper model name; case mismatch.

Related errors

AI-assisted analysis of jamiepine/voicebox@51f49dea19 (2026-08-12). Data as JSON: /api/errors/81286a1da9a33942. Report an issue: GitHub.