p PatchFirst

CVE-2026-44223

CVE-2026-44223 · vLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parameters

Esta página presenta datos de fuentes públicas de vulnerabilidades. Consulta el aviso original antes de actualizar.

vLLM is an inference and serving engine for large language models (LLMs). From 0.18.0 to before 0.20.0, the extract_hidden_states speculative decoding proposer in vLLM returns a tensor with an incorrect shape after the first decode step, causing a RuntimeError that crashes the EngineCore process. The crash is triggered when any request in the batch uses sampling penalty parameters (repetition_penalty, frequency_penalty, or presence_penalty). A single request with a penalty parameter (e.g., "repetition_penalty": 1.1) is sufficient to crash the server. This vulnerability is fixed in 0.20.0.

CVSS 6.5 · Media · EPSS 0,4% · Corrección disponible

Publicación
Actualización

Pila tecnológica: python

Productos y versiones afectados

  • vllm

    Versiones
    • >= 0.18.0, < 0.20.0: Afectada

    Sin corrección publicada

  • vllm PyPI

    Versiones
    • 0.18.0
    • 0.18.1
    • 0.19.0
    • 0.19.1
    Intervalos de versiones afectados
    • ECOSYSTEM: ≥ 0.18.0, < 0.20.0

    Actualiza a 0.20.0

Fuentes originales

Comprueba tu servidor →