[V0 Deprecation] Enable the remaining multimodal tests in V1 (#25307)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-09-20 17:50:58 +00:00
committed by GitHub
parent d88918e4c2
commit bef180f009
8 changed files with 195 additions and 214 deletions
+11 -14
View File
@@ -5,7 +5,6 @@ import pytest
import torch
from tests.conftest import VllmRunner
from vllm.utils import set_default_torch_num_threads
@pytest.mark.parametrize(
@@ -25,19 +24,17 @@ def test_inference(
prompt = dict(prompt_token_ids=[1],
multi_modal_data=dict(pixel_values=pixel_values,
location_coords=location_coords))
with (
set_default_torch_num_threads(1),
vllm_runner(
model,
runner="pooling",
dtype=torch.float16,
enforce_eager=True,
skip_tokenizer_init=True,
# Limit the maximum number of sequences to avoid the
# test going OOM during the warmup run
max_num_seqs=32,
) as vllm_model,
):
with vllm_runner(
model,
runner="pooling",
dtype="half",
enforce_eager=True,
skip_tokenizer_init=True,
# Limit the maximum number of sequences to avoid the
# test going OOM during the warmup run
max_num_seqs=32,
default_torch_num_threads=1,
) as vllm_model:
vllm_output = vllm_model.llm.encode(prompt)
assert torch.equal(