oobabooga/textgen · error · ValueError

`tfs` has to be a float >= 0 and <= 1, but is {tfs}

Error message

`tfs` has to be a float >= 0 and <= 1, but is {tfs}

What it means

Error "`tfs` has to be a float >= 0 and <= 1, but is {tfs}" thrown in oobabooga/textgen.

Source

Thrown at modules/sampler_hijack.py:136

        diff = scores - max_logit
        k = (3 - self.smoothing_curve) / 2
        s = (self.smoothing_curve - 1) / 2

        # Apply transformation to non-negative infinity values
        transformed_logits = torch.where(
            scores != float('-inf'),
            -(k * self.smoothing_factor * diff**2) + (s * self.smoothing_factor * diff**3) + max_logit,
            scores
        )

        return transformed_logits


class TailFreeLogitsWarper(LogitsProcessor):
    def __init__(self, tfs: float, filter_value: float = -float("Inf"), min_tokens_to_keep: int = 1):
        tfs = float(tfs)
        if tfs < 0 or tfs > 1.0:
            raise ValueError(f"`tfs` has to be a float >= 0 and <= 1, but is {tfs}")
        self.tfs = tfs
        self.filter_value = filter_value
        self.min_tokens_to_keep = min_tokens_to_keep

    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor:
        sorted_logits, sorted_indices = torch.sort(scores, descending=True)
        probs = sorted_logits.softmax(dim=-1)

        # Compute second derivative normalized CDF
        d2 = probs.diff().diff().abs()
        normalized_d2 = d2 / d2.sum(dim=-1, keepdim=True)
        normalized_d2_cdf = normalized_d2.cumsum(dim=-1)

        # Remove tokens with CDF value above the threshold (token with 0 are kept)
        sorted_indices_to_remove = normalized_d2_cdf > self.tfs

        # Centre the distribution around the cutoff as in the original implementation of the algorithm
        sorted_indices_to_remove = torch.cat(

View on GitHub (pinned to ed888c71f2)

Solutions

1. Set tfs (tail free sampling) to a float between 0 and 1 inclusive; use 1.0 to disable it.

When it happens

Trigger: Raised during sampler validation when the `tfs` (tail-free sampling) parameter is outside the range [0, 1]. Triggers when tfs is set negative or greater than 1 in generation settings.

Common situations: See trigger scenarios.

AI-assisted analysis of oobabooga/textgen@ed888c71f2 (2026-08-15). Data as JSON: /api/errors/2a979f33d469c116. Report an issue: GitHub.