zylon-ai/private-gpt · error · ValueError

Could not load tokenizer from '{model_id}': {e}

Error message

Could not load tokenizer from '{model_id}': {e}

What it means

With downloads allowed, any OSError from AutoProcessor.from_pretrained is wrapped as ValueError("Could not load tokenizer from '{model_id}': {e}"). OSError from the Hub typically means the repo id is invalid/not found, the connection failed, or the repo is gated.

Source

Thrown at private_gpt/components/llm/tokenizers/huggingface.py:104

            # Extract tokenizer from multimodal processor if needed
            tokenizer: PreTrainedTokenizerBase
            if hasattr(loaded, "tokenizer"):
                processor = cast(ProcessorMixin, loaded)
                tokenizer = cast(PreTrainedTokenizerBase, loaded.tokenizer)
                is_multimodal = True
            else:
                tokenizer = cast(PreTrainedTokenizerBase, loaded)

            return cls(tokenizer, is_multimodal=is_multimodal, processor=processor)

        except OSError as e:
            if local_files_only:
                raise FileNotFoundError(
                    f"Local model files not found at '{model_id}'. "
                    f"Ensure the model is downloaded locally."
                ) from e
            raise ValueError(f"Could not load tokenizer from '{model_id}': {e}") from e
        except Exception as e:
            raise ValueError(f"Failed to load tokenizer: {e}") from e

    @classmethod
    def is_available(cls, model_id: str | Path | None, **kwargs: Any) -> bool:
        return bool(model_id)

    @property
    def all_special_tokens(self) -> list[str]:
        tokens: list[str] = self._tokenizer.all_special_tokens
        return tokens

    @property
    def all_special_ids(self) -> list[int]:
        ids: list[int] = self._tokenizer.all_special_ids
        return ids

    @property

View on GitHub (pinned to 4a030776a3)

Solutions

1. Check the underlying message chained in {e}: fix the repo id or network access accordingly.
2. For gated models run `hf auth login` and set HF_TOKEN so the download is authorized.
3. For offline-capable setups, pre-download the model and pass local_files_only=True (which then gives the more specific FileNotFoundError if still missing).

Example fix

# before
tok = HuggingFaceTokenizer.from_pretrained('mistralai/mistral-7b-typo')

# after
hf auth login  # if gated
tok = HuggingFaceTokenizer.from_pretrained('mistralai/Mistral-7B-Instruct-v0.3')

from huggingface_hub import HfApi

def repo_exists(model_id: str) -> bool:
    try:
        HfApi().repo_info(model_id)
        return True
    except Exception:
        return False

try:
    tok = HuggingFaceTokenizer.from_pretrained(model_id)
except ValueError as e:
    if 'Could not load tokenizer' in str(e):
        cause = e.__cause__
        logger.error('load failed for %s: %s', model_id, cause)
        raise ModelLoadError(model_id, cause) from e
    raise

When it happens

Trigger: HuggingFaceTokenizer.from_pretrained('nonexistent/model') with local_files_only=False; network outage or blocked HF endpoint in sandboxes; gated repo without prior `hf auth login`; malformed repo id (more than one slash, spaces).

Common situations: Typo'd or retired model ids; egress-restricted corporate networks; missing HF token for Llama-style gated models; proxies/SSL interception breaking huggingface_hub requests.

Related errors

• Transformers dependencies are not installed.
• Local model files not found at '{model_id}'. Ensure the mode
• Failed to load tokenizer: {e}
• HuggingFaceTokenizer is not available with the given configu
• Remote tokenizer response did not contain a valid 'tokens' f

AI-assisted analysis of zylon-ai/private-gpt@4a030776a3 (2026-08-15). Data as JSON: /api/errors/3f674801fbdc6d25. Report an issue: GitHub.