Live › vulnerability › CVE-2026-115913
Go to dashboardvLLM's Phi-4 multimodal audio encoder uses eager attention by default, so one long audio clip from a remote client makes it store full quadratic attention matrices in GPU memory. The out-of-memory condition terminates EngineCore and interrupts service for all users.
Summary written from the sources below. Check the originals before acting.
First seen by InfraStack 2026-10-26 17:31 UTC. All times on this page are UTC.
Teams self-hosting vLLM below 0.32.0 that serve the Phi-4 multimodal model and accept audio input from clients they do not fully trust. Deployments that serve only text models, or that do not accept audio from remote clients, are not exposed to this path.
No product ranges from the sources yet.
pip show vllm and treat anything below 0.32.0 as affected.0.32.0 or later.Each row is the first time that source carried this signal. The analyst’s rows are its own findings.
vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the Phi-4 multimodal audio encoder in phi4mm_audio.py and phi4mm_utils.py uses eager attention by default, allowing a remote client to submit one legal long audio clip that causes MultiHeadedAttention.forward_attention to materialize and retain full attention matrices in self.attn. The scaled dot-product attention branch does not materialize these matrices. The resulting quadratic GPU-memory use can cause an out-of-memory condition and terminate EngineCore for all users. This issue is fixed in version 0.32.0.
This page as .md · .json · in your agent: get_signal("CVE-2026-115913")
NVD data is provided by NIST and is public domain. Posts quoted belong to their authors.