[V0 Deprecation] Remove VLLM_USE_V1 from tests (#26341)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

tests/compile/piecewise/test_full_cudagraph.py

@@ -66,7 +66,6 @@ def llm_pair(request):
             pytest.skip("Only Blackwell GPUs support Cutlass MLA")
 
     env_vars = {
-        "VLLM_USE_V1": "1",
         # Force native sampler to avoid potential nondeterminism in FlashInfer
         # when per-request generators are not used in V1.
         "VLLM_USE_FLASHINFER_SAMPLER": "0",
@@ -161,7 +160,6 @@ def test_full_cudagraph_with_invalid_backend():
     with (
         temporary_environ(
             {
-                "VLLM_USE_V1": "1",
                 "VLLM_ATTENTION_BACKEND": "FLEX_ATTENTION",
                 # Flex_Attention is not supported with full cuda graph
             }

tests/compile/piecewise/test_simple.py

@@ -18,7 +18,6 @@ from vllm.config import (
     VllmConfig,
     set_current_vllm_config,
 )
-from vllm.envs import VLLM_USE_V1
 from vllm.forward_context import BatchDescriptor, set_forward_context
 from vllm.utils import is_torch_equal_or_newer
 
@@ -127,7 +126,6 @@ def _run_simple_model(
 @pytest.mark.parametrize("use_inductor", [True, False])
 @torch.inference_mode()
 def test_simple_piecewise_compile(use_inductor):
-    assert VLLM_USE_V1
     _run_simple_model(
         splitting_ops=["silly.attention"],
         use_inductor_graph_partition=False,
@@ -146,7 +144,6 @@ def test_simple_piecewise_compile(use_inductor):
 @torch.inference_mode()
 @pytest.mark.parametrize("splitting_ops", [["silly.attention"], []])
 def test_simple_inductor_graph_partition(splitting_ops):
-    assert VLLM_USE_V1
     if not is_torch_equal_or_newer("2.9.0.dev"):
         pytest.skip("inductor graph partition is only available in PyTorch 2.9+")