fix(security): enforce audio decode duration limit in NanoNemotronVL (#50221)

Signed-off-by: jperezde <jperezde@redhat.com>
This commit is contained in:
Juan Pérez de Algaba
2026-08-13 03:31:55 -07:00
committed by GitHub
parent b8baa31a28
commit 37c3bdf5a7
2 changed files with 63 additions and 2 deletions
@@ -18,6 +18,7 @@ import torch
import torch.nn as nn
from transformers import BatchFeature, PretrainedConfig
from vllm import envs
from vllm.config import VllmConfig
from vllm.config.multimodal import BaseDummyOptions, VideoDummyOptions
from vllm.inputs import MultiModalDataDict, MultiModalInput
@@ -627,7 +628,12 @@ class NanoNemotronVLMultiModalProcessor(
"the chat API with a model that sets use_audio_in_video)."
)
try:
audio_items.append(load_audio_pyav(BytesIO(video_bytes)))
audio_items.append(
load_audio_pyav(
BytesIO(video_bytes),
max_duration_s=envs.VLLM_MAX_AUDIO_DECODE_DURATION_S,
)
)
has_audio.append(True)
except Exception:
logger.debug(