huggingface/transformers · error · ValueError

`input_starting_length` has to be a non-negative integer, bu

Error message

`input_starting_length` has to be a non-negative integer, but is {input_start_len}

What it means

Error "`input_starting_length` has to be a non-negative integer, but is {input_start_len}" thrown in huggingface/transformers.

Source

Thrown at src/transformers/generation/logits_process.py:2202

    This logits processor is exclusively compatible with
    [Bark](https://huggingface.co/docs/transformers/en/model_doc/bark)'s fine submodel. See the model documentation
    for examples.

    </Tip>

    Args:
        input_start_len (`int`):
            The length of the initial input sequence.
        semantic_vocab_size (`int`):
            Vocabulary size of the semantic part, i.e number of tokens associated to the semantic vocabulary.
        codebook_size (`int`):
            Number of tokens associated to the codebook.
    """

    def __init__(self, input_start_len: int, semantic_vocab_size: int, codebook_size: int):
        if not isinstance(input_start_len, int) or input_start_len < 0:
            raise ValueError(f"`input_starting_length` has to be a non-negative integer, but is {input_start_len}")

        self.input_start_len = input_start_len
        self.semantic_vocab_size = semantic_vocab_size
        self.codebook_size = codebook_size

    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor:
        curr_len = input_ids.shape[-1]

        # even -> first codebook, odd -> second codebook
        is_first_codebook = ((curr_len - self.input_start_len) % 2) == 0

        scores_processed = scores.clone()
        if is_first_codebook:
            scores_processed[:, : self.semantic_vocab_size] = -float("inf")
            scores_processed[:, self.semantic_vocab_size + self.codebook_size :] = -float("inf")
        else:
            scores_processed[:, : self.semantic_vocab_size + self.codebook_size] = -float("inf")

View on GitHub (pinned to a597f97485)

Solutions

1. Pass a non-negative integer for `input_starting_length`.
2. Use `input_ids.shape[-1]` for the actual prompt length.

When it happens

Trigger: Raised in ClassifierFreeGuidanceLogitsProcessor.__init__ when input_starting_length (or input_start_len) is negative or not an integer.

Common situations: Misconfigured CFG processor construction with a negative input_starting_length value.

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/00461c811aa6c5f1. Report an issue: GitHub.