[Security] Gate per-request multimodal processor kwargs (#58830)
Signed-off-by: Juan Pérez de Algaba <jperezde@redhat.com> Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
co-authored by
Cursor
parent
2f41e00235
commit
31f2e70cd3
@@ -46,6 +46,7 @@ def server():
|
||||
json.dumps({"image": MAXIMUM_IMAGES}),
|
||||
"--chat-template",
|
||||
str(vlm2vec_jinja_path),
|
||||
"--trust-request-mm-kwargs",
|
||||
]
|
||||
|
||||
with RemoteOpenAIServer(MODEL_NAME, args) as remote_server:
|
||||
|
||||
@@ -26,6 +26,36 @@ TEXTS_2 = [
|
||||
]
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_flash_late_interaction_rejects_mm_processor_kwargs_by_default():
|
||||
from unittest.mock import MagicMock
|
||||
|
||||
from vllm.config.multimodal import MultiModalConfig
|
||||
from vllm.entrypoints.pooling.base.serving import PoolingBaseServing
|
||||
from vllm.entrypoints.pooling.scoring.protocol import ScoreQueriesDocumentsRequest
|
||||
from vllm.entrypoints.pooling.scoring.serving import ServingScores
|
||||
from vllm.exceptions import VLLMValidationError
|
||||
|
||||
serving = MagicMock()
|
||||
serving.model_config = MagicMock(multimodal_config=MultiModalConfig())
|
||||
serving.trust_request_mm_kwargs = False
|
||||
serving.io_processor = MagicMock()
|
||||
serving._init_ctx = PoolingBaseServing._init_ctx.__get__(serving)
|
||||
request = ScoreQueriesDocumentsRequest(
|
||||
model=MODEL_NAME,
|
||||
queries=TEXTS_1[0],
|
||||
documents=TEXTS_2[0],
|
||||
mm_processor_kwargs={"vision_min_num_patches": 3_000_000_000},
|
||||
)
|
||||
|
||||
with pytest.raises(
|
||||
VLLMValidationError,
|
||||
match="Per-request mm_processor_kwargs are disabled",
|
||||
):
|
||||
await ServingScores.flash_late_interaction(serving, request)
|
||||
serving._preprocessing.assert_not_called()
|
||||
|
||||
|
||||
@pytest.fixture(scope="module", params=[True, False])
|
||||
def server(request):
|
||||
args = [
|
||||
|
||||
Reference in New Issue
Block a user