sgl-project/sglang · error · ValueError

presence_penalty must be in [-2, 2], got {self.presence_pena

Error message

presence_penalty must be in [-2, 2], got {self.presence_penalty}.

What it means

SamplingParams.verify() requires presence_penalty to lie in [-2, 2] (OpenAI-compatible range). presence_penalty adds a flat penalty to tokens that have already appeared at least once; values outside the range raise this ValueError during normalize()/verify().

Source

Thrown at python/sglang/srt/sampling/sampling_params.py:179

        if not 0.0 < self.top_p <= 1.0:
            raise ValueError(f"top_p must be in (0, 1], got {self.top_p}.")
        if not 0.0 <= self.min_p <= 1.0:
            raise ValueError(f"min_p must be in [0, 1], got {self.min_p}.")
        if self.top_k < 1 or self.top_k == -1:
            raise ValueError(
                f"top_k must be -1 (disable) or at least 1, got {self.top_k}."
            )
        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 -2.0 <= self.presence_penalty <= 2.0:
            raise ValueError(
                "presence_penalty must be in [-2, 2], got " f"{self.presence_penalty}."
            )
        if not 0.0 < self.repetition_penalty <= 2.0:
            raise ValueError(
                "repetition_penalty must be in (0, 2] (1.0 = no penalty), "
                f"got {self.repetition_penalty}."
            )
        if not 0 <= self.min_new_tokens:
            raise ValueError(
                f"min_new_tokens must be in [0, max_new_tokens], got "
                f"{self.min_new_tokens}."
            )
        if self.max_new_tokens is not None:
            if self.max_new_tokens < 0:
                raise ValueError(
                    f"max_new_tokens must be at least 0, got {self.max_new_tokens}."
                )
            if not self.min_new_tokens <= self.max_new_tokens:
                raise ValueError(
                    f"min_new_tokens must be in [0, max_new_tokens({self.max_new_tokens})], got "
                    f"{self.min_new_tokens}."
                )

View on GitHub (pinned to 0132848349)

Solutions

  1. Clamp presence_penalty to [-2, 2]
  2. Add request validation middleware for penalty fields
  3. Use 0.0 to disable

Example fix

# before
params = SamplingParams(presence_penalty=penalty)
# after
params = SamplingParams(presence_penalty=max(-2.0, min(2.0, penalty)))
Defensive patterns

Strategy: validation

Validate before calling

presence_penalty = max(-2.0, min(2.0, presence_penalty))
params = SamplingParams(presence_penalty=presence_penalty)

Type guard

def valid_presence_penalty(p):
    return isinstance(p,(int,float)) and -2.0 <= p <= 2.0

Try / catch

try:
    llm.generate(prompts, SamplingParams(presence_penalty=p))
except ValueError as e:
    if 'presence_penalty' in str(e):
        p = 0.0
    else:
        raise

Prevention

When it happens

Trigger: Passing SamplingParams(presence_penalty=2.5) or below -2; OpenAI-style API requests with out-of-range presence_penalty forwarded to SGLang.

Common situations: Proxy servers forwarding unvalidated OpenAI request fields; experiment scripts sweeping penalties outside the legal range; users assuming 0-10 scale.

Related errors

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/a19946a42bf8e6a3. Report an issue: GitHub.