vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine

影響: pip:vllm < 0.32.0 → 0.32.0

The Phi-4 multimodal audio encoder (phi4mm_audio.py / phi4mm_utils.py) runs its conformer attention in eager mode by default: the official HF config of microsoft/Phi-4-multimodal-instruct does not set use_pt_scaled_dot_product_attention, so the default False applies (phi4mm_audio.py:166, :876; phi4mm_utils.py:1669). The eager path fully materializes the attention probability matrix of shape [n_chunks, 16, 500, 500] per layer — and then stores it on the module: self.attn = attn (phi4mm_utils.py:…

GitHub Advisory Database에서 읽기

2026. 10. 29. 오전 06:15, GitHub Advisory Database에서 "vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine"을(를) 게시했습니다 (유형: 취약점 권고). 심각도는 "보통"(medium)입니다. 관련 기술: Python. CVE: CVE-2026-115913.