Evidence · DiggingBeagle record
vLLM Phi-4 audio encoder unbounded attention memory, CVE-2026-115913
Advisory showing a valid long audio input exhausting shared GPU memory in Phi-4 multimodal inference.
- Source role
- primary disclosure
Evidence record
Advisory showing a valid long audio input exhausting shared GPU memory in Phi-4 multimodal inference.
Claim-level citations (2)
- supportsvLLM Phi-4 audio attention allocations exhausted GPU memory on valid input: The vLLM Phi-4 audio advisory reproduces GPU memory exhaustion from a permitted long WAV input because eager attention tensors are retained across the encoder’s layers.
GHSA-8397-8qf6-f2gm: Phi-4 eager attention, ten-minute waveform and GPU allocation trace
- supportsvLLM Phi-4 audio attention allocations exhausted GPU memory on valid input: Phi-4 multimodal audio attention materialized full layer-by-layer attention matrices for long waveforms in vLLM instead of using a bounded streaming form. The issue requires the affected Phi-4 audio encoder and a long permitted audio input on an inadequately provisioned worker. A legal ten-minute WAV was enough in the reproduction to exhaust roughly 24 GiB of GPU memory and terminate the engine.
GHSA-8397-8qf6-f2gm: Phi-4 eager attention, ten-minute waveform and GPU allocation trace