CVE-2025-62372

MEDIUM

vLLM 0.5.5-0.11.1 - Denial of Service via Multimodal Embedding Input Shape Mismatch

Title source: llm

Description

vLLM is an inference and serving engine for large language models (LLMs). From version 0.5.5 to before 0.11.1, users can crash the vLLM engine serving multimodal models by passing multimodal embedding inputs with correct ndim but incorrect shape (e.g. hidden dimension is wrong), regardless of whether the model is intended to support such inputs (as defined in the Supported Models page). This issue has been patched in version 0.11.1.

References (4)

Core 4

Core References

Mitigation, Vendor Advisory x_refsource_confirm

https://github.com/vllm-project/vllm/security/advisories/GHSA-pmqf-x6x8-p7qw

Issue Tracking, Patch, Vendor Advisory x_refsource_misc

https://github.com/vllm-project/vllm/pull/27204

Issue Tracking x_refsource_misc

https://github.com/vllm-project/vllm/pull/6613

Patch x_refsource_misc

https://github.com/vllm-project/vllm/commit/58fab50d82838d5014f4a14d991fdb9352c9c84b

View Patch ZIP pw:eip

Scores

CVSS v3 6.5

EPSS 0.0032

EPSS Percentile 23.4%

Attack Vector NETWORK

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H

CISA SSVC

Vulnrichment

Exploitation none

Automatable no

Technical Impact partial

Details

CWE

CWE-129

Status published

Products (3)

pypi/vllm 0.5.5 - 0.11.1PyPI

vllm/vllm 0.11.1 rc0 (2 CPE variants)

vllm/vllm 0.5.5 - 0.11.1

Published Nov 21, 2025

Tracked Since Feb 18, 2026