[V1] Make v1 more testable (#9888)

Signed-off-by: Joe Runde <Joseph.Runde@ibm.com>

tests/conftest.py

@@ -5,6 +5,7 @@ from collections import UserList
 from enum import Enum
 from typing import (Any, Callable, Dict, List, Optional, Tuple, Type,
                     TypedDict, TypeVar, Union)
+from unittest.mock import patch
 
 import numpy as np
 import pytest
@@ -108,6 +109,23 @@ VIDEO_ASSETS = _VideoAssets()
 """Singleton instance of :class:`_VideoAssets`."""
 
 
+@pytest.fixture(params=[True, False])
+def run_with_both_engines(request):
+    # Automatically runs tests twice, once with V1 and once without
+    use_v1 = request.param
+    # Tests decorated with `@skip_v1` are only run without v1
+    skip_v1 = request.node.get_closest_marker("skip_v1")
+
+    if use_v1:
+        if skip_v1:
+            pytest.skip("Skipping test on vllm V1")
+        with patch('vllm.envs.VLLM_USE_V1', True):
+            yield
+    else:
+        with patch('vllm.envs.VLLM_USE_V1', False):
+            yield
+
+
 @pytest.fixture(autouse=True)
 def init_test_http_connection():
     # pytest_asyncio may use a different event loop per test