[Tests] Select V2 for diffusion scheduler unit tests (#58272)

Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
Andreas Karatzas
2026-09-23 02:04:22 -05:00
committed by GitHub
co-authored by OpenAI Codex
parent eb2e91d3ba
commit 0f2a15c927
+12
View File
@@ -6896,6 +6896,13 @@ def _diffusion_request(req_id: str, extra_args: dict) -> Request:
return request
@pytest.fixture
def diffusion_model_runner(monkeypatch):
# These CPU tests only exercise scheduling, not Triton kernels.
monkeypatch.setattr("vllm.config.vllm.HAS_TRITON", True)
monkeypatch.setenv("VLLM_USE_V2_MODEL_RUNNER", "1")
def _diffusion_scheduler(**kwargs) -> DiffusionAsyncScheduler:
scheduler = create_scheduler(
async_scheduling=True,
@@ -6908,6 +6915,7 @@ def _diffusion_scheduler(**kwargs) -> DiffusionAsyncScheduler:
@pytest.mark.parametrize("async_scheduling", [False, True])
@pytest.mark.usefixtures("diffusion_model_runner")
def test_diffusion_scheduler_is_selected_by_default(async_scheduling):
config = create_scheduler(
async_scheduling=async_scheduling, diffusion_canvas_length=8
@@ -6917,6 +6925,7 @@ def test_diffusion_scheduler_is_selected_by_default(async_scheduling):
)
@pytest.mark.usefixtures("diffusion_model_runner")
def test_diffusion_scheduler_narrows_the_canvas_per_request():
scheduler = _diffusion_scheduler()
wide = _diffusion_request("wide", {})
@@ -6934,6 +6943,7 @@ def test_diffusion_scheduler_narrows_the_canvas_per_request():
@pytest.mark.parametrize("structured", [False, True])
@pytest.mark.usefixtures("diffusion_model_runner")
def test_diffusion_scheduler_trims_full_width_worker_drafts(structured):
"""Padded worker drafts must be narrowed before scheduling or grammar validation."""
scheduler = _diffusion_scheduler()
@@ -6964,6 +6974,7 @@ def test_diffusion_scheduler_trims_full_width_worker_drafts(structured):
assert narrow.structured_output_request.grammar.seen == [tokens[:4]]
@pytest.mark.usefixtures("diffusion_model_runner")
def test_diffusion_scheduler_defers_a_read_with_every_step_in_flight():
scheduler = _diffusion_scheduler()
one = _diffusion_request(
@@ -6986,6 +6997,7 @@ def test_diffusion_scheduler_defers_a_read_with_every_step_in_flight():
assert set(scheduler.schedule().num_scheduled_tokens) == {"gen"}
@pytest.mark.usefixtures("diffusion_model_runner")
def test_diffusion_read_deferral_keeps_a_longer_pp_wait():
scheduler = _diffusion_scheduler(pipeline_parallel_size=3, use_v2_model_runner=True)
read = _diffusion_request(