vllm-project/vllm · error · LogprobsError

requested {parameter} of {requested}, which is greater than

Error message

requested {parameter} of {requested}, which is greater than max allowed: {max_allowed}

What it means

The requested logprobs count (after `-1` expansion to vocab size) exceeds the server's maximum allowed count (`sampling_limits.max_logprobs`, itself normalized and validated in rust/src/text/src/lower/logprobs.rs:48-73). Message reports the parameter name (`logprobs` or `prompt_logprobs`), the requested count, and the cap.

Source

Thrown at rust/src/text/src/lower/logprobs.rs:17

// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: Copyright contributors to the vLLM project

//! Python-compatible validation for logprobs sampling params.
//!
//! `-1` is expanded only for bounds checks. The original request values are
//! passed through to engine-core.

use thiserror::Error;

use crate::backend::SamplingLimits;

#[derive(Debug, Error)]
pub enum LogprobsError {
    #[error("{parameter} must be non-negative or -1, got {value}")]
    InvalidCount { parameter: &'static str, value: i32 },
    #[error(
        "requested {parameter} of {requested}, which is greater than max allowed: {max_allowed}"
    )]
    TooManyCount {
        parameter: &'static str,
        requested: usize,
        max_allowed: usize,
    },
    #[error(
        "requested logprob_token_ids of length {requested}, \
         which is greater than max allowed: {max_allowed}"
    )]
    TooManyTokenIds {
        requested: usize,
        max_allowed: usize,
    },
    #[error(
        "when both logprobs and logprob_token_ids are set, logprobs must equal \
         len(logprob_token_ids). Got logprobs={logprobs}, len(logprob_token_ids)={num_token_ids}."

View on GitHub (pinned to c794754062)

Solutions

1. Lower the request's logprobs/prompt_logprobs to <= the reported max_allowed
2. If you own the server, raise the max_logprobs limit at startup

Example fix

# before
# server started with max_logprobs=20, request asks for all
request.logprobs = -1

# after
request.logprobs = 20

let cap = server_max_logprobs; // from /v1/models or server config
let requested = if logprobs == Some(-1) { vocab_size } else { logprobs.unwrap_or(0) };
assert!(requested <= cap, "requested {requested} > max_logprobs {cap}");

fn is_too_many_count(e: &Error) -> bool {
    matches!(e, Error::Logprobs(LogprobsError::TooManyCount { .. }))
}

match err {
    Error::Logprobs(LogprobsError::TooManyCount { parameter, requested, max_allowed }) =>
        bad_request(format!("{parameter}={requested} exceeds cap {max_allowed}")),
    _ => /* ... */
}

When it happens

Trigger: Requesting `logprobs: N` where N > max_logprobs, or `logprobs: -1` (expands to vocab_size) on a server whose max_logprobs cap is below the vocabulary size; same for prompt_logprobs.

Common situations: Using -1 'give me everything' against a server started with a low --max-logprobs; raising logprobs after a model swap to one with a larger vocab while the cap stayed fixed; client defaults copied from a server with a higher cap.

Related errors

• max_logprobs must be non-negative or -1, got {}
• max_logprobs must be non-negative or -1
• {parameter} must be non-negative or -1, got {value}
• requested logprob_token_ids of length {requested}, which is
• when both logprobs and logprob_token_ids are set, logprobs m

AI-assisted analysis of vllm-project/vllm@c794754062 (2026-08-14). Data as JSON: /api/errors/7dfcd535d436fd38. Report an issue: GitHub.