2noise/ChatTTS · error · ValueError

repetition_penalty must be in (0, 2], got {self.repetition_p

Error message

repetition_penalty must be in (0, 2], got {self.repetition_penalty}.

What it means

SamplingParams._verify_args enforces repetition_penalty in the range (0, 2] (open at zero). A repetition_penalty of exactly 0 or negative is meaningless — it would zero all logits — so it is rejected.

Source

Thrown at ChatTTS/model/velocity/sampling_params.py:198

    def _verify_args(self) -> None:
        if self.n < 1:
            raise ValueError(f"n must be at least 1, got {self.n}.")
        if self.best_of < self.n:
            raise ValueError(
                f"best_of must be greater than or equal to n, "
                f"got n={self.n} and best_of={self.best_of}."
            )
        if not -2.0 <= self.presence_penalty <= 2.0:
            raise ValueError(
                "presence_penalty must be in [-2, 2], got " f"{self.presence_penalty}."
            )
        if not -2.0 <= self.frequency_penalty <= 2.0:
            raise ValueError(
                "frequency_penalty must be in [-2, 2], got "
                f"{self.frequency_penalty}."
            )
        if not 0.0 < self.repetition_penalty <= 2.0:
            raise ValueError(
                "repetition_penalty must be in (0, 2], got "
                f"{self.repetition_penalty}."
            )
        # if self.temperature < 0.0:
        #     raise ValueError(
        #         f"temperature must be non-negative, got {self.temperature}.")
        if not 0.0 < self.top_p <= 1.0:
            raise ValueError(f"top_p must be in (0, 1], got {self.top_p}.")
        if self.top_k < -1 or self.top_k == 0:
            raise ValueError(
                f"top_k must be -1 (disable), or at least 1, " f"got {self.top_k}."
            )
        if not 0.0 <= self.min_p <= 1.0:
            raise ValueError("min_p must be in [0, 1], got " f"{self.min_p}.")
        if self.max_tokens < 1:
            raise ValueError(f"max_tokens must be at least 1, got {self.max_tokens}.")
        if self.logprobs is not None and self.logprobs < 0:
            raise ValueError(f"logprobs must be non-negative, got {self.logprobs}.")

View on GitHub (pinned to 77b89ee281)

Solutions

Use repetition_penalty=1.0 to disable the penalty (not 0)
Keep the value within (0, 2]
Clamp/validate user-supplied values at your boundary

Example fix

# before
params = SamplingParams(repetition_penalty=0)  # trying to disable

# after
params = SamplingParams(repetition_penalty=1.0)  # no penalty

Defensive patterns

Strategy: validation

Validate before calling

def normalize_repetition(v) -> float:
    v = float(v)
    if v == 0.0:
        return 1.0  # caller meant "disabled"
    return min(max(v, 0.000001), 2.0)

Type guard

def is_valid_repetition_penalty(v) -> bool:
    return isinstance(v, (int, float)) and 0.0 < v <= 2.0

Prevention

Use 1.0 to disable repetition penalty, never 0
Cap HuggingFace-generation-config penalties above 2 before passing them in

When it happens

Trigger: Constructing SamplingParams with repetition_penalty=0, a negative value, or anything above 2.0 (e.g. the common-looking 2.5).

Common situations: Disabling repetition penalty by setting it to 0 instead of 1.0 (1.0 = no effect); copying a penalty of 2.5 or 3.0 from a HuggingFace generation_config; passing values through from users unvalidated.

Related errors

AI-assisted analysis of 2noise/ChatTTS@77b89ee281 (2026-08-26). Data as JSON: /api/errors/396b798c5e16c141. Report an issue: GitHub.