[Bugfix] Make MM batching more robust (#33817)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-02-05 20:40:58 +00:00
committed by GitHub
parent 4145e50d85
commit 116880a5a0
13 changed files with 625 additions and 428 deletions
+9 -3
View File
@@ -5,8 +5,10 @@ import pytest
import torch
from tests.conftest import VllmRunner
from tests.utils import create_new_process_for_each_test
@create_new_process_for_each_test() # Memory is not cleaned up properly otherwise
@pytest.mark.parametrize(
"model",
[
@@ -22,10 +24,14 @@ def test_inference(
location_coords = torch.full((1, 2), 1.0, dtype=torch.float16)
prompt = dict(
prompt_token_ids=[1],
multi_modal_data=dict(
pixel_values=pixel_values, location_coords=location_coords
),
multi_modal_data={
"image": {
"pixel_values": pixel_values,
"location_coords": location_coords,
}
},
)
with vllm_runner(
model,
runner="pooling",