CVE-2026-115916
mediumvLLM is an inference and serving engine for large language models.
Vendor / product: Vllm — vllm
vLLM is an inference and serving engine for large language models. Prior to 0.32.0, vllm/tokenizers/hf.py computes max_token_id with an off-by-one upper bound, allowing a pre-tokenized prompt ID equal to the vocabulary size to pass validation. For a tensor-parallel size of one and an exact multiple-of-64 vocabulary, VocabParallelEmbedding.forward does not mask that value before F.embedding uses it, causing a CUDA device-side assertion and terminating the engine. This issue is fixed in version 0.32.0.
CVSS
6.5
6.5 (CVSS v3.1)
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H Severity source: NVD CVSS v3.1 (security-advisories@github.com)
Timeline (UTC)
Published: 2026-10-26 18:17
Modified: 2026-10-28 15:36
NVD status: Analyzed
Last synced 1 day ago
Verified against the source API 2026-10-28 16:01 UTC
Exploitation
Not listed in CISA KEV (as of last catalog sync)
Absence from KEV does not mean a CVE is not exploited. Exploitation status: unknown (no authoritative evidence in our sources).
Affected products
vllm / vllm <0.32.0
Weakness classification (CWE)
Source: NVD / CISA. “NVD-CWE-Other” and “NVD-CWE-noinfo” mean NVD could not assign a specific weakness.
Mitigation & advisories
Patch / mitigation: https://github.com/vllm-project/vllm/commit/37cecc8141e2dcba9dd534897a164d4baaf27229
Patch / mitigation: https://github.com/vllm-project/vllm/pull/59491
Patch / mitigation: https://github.com/vllm-project/vllm/security/advisories/GHSA-q97c-p368-75c4
References
- https://github.com/vllm-project/vllm/commit/37cecc8141e2dcba9dd534897a164d4baaf27229 security-advisories@github.com
- https://github.com/vllm-project/vllm/pull/59491 security-advisories@github.com
- https://github.com/vllm-project/vllm/releases/tag/v0.32.0 security-advisories@github.com
- https://github.com/vllm-project/vllm/security/advisories/GHSA-q97c-p368-75c4 security-advisories@github.com
Data sources
NVD CVE API 2.0 — fetched 2026-10-28 16:01 UTC · record ↗