[Test] Fit Qwen2.5 online serving placement to its two-card allocation

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
2026-10-04 03:26:52 +08:00
4 changed files with 7 additions and 29 deletions
@@ -1,26 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
-
-import pytest
-
-
-@pytest.fixture
-def sana_model_parallel(tmp_path):
-    """Construct native parallel layers with real single-rank GPU groups."""
-    from vllm_omni.diffusion.distributed.parallel_state import (
-        destroy_distributed_env,
-        init_distributed_environment,
-        initialize_model_parallel,
-    )
-
-    init_distributed_environment(
-        world_size=1,
-        rank=0,
-        local_rank=0,
-        distributed_init_method=f"file://{tmp_path / 'sana-distributed'}",
-    )
-    try:
-        initialize_model_parallel()
-        yield
-    finally:
-        destroy_distributed_env()
@@ -202,7 +202,6 @@ def test_sana_video_transformer_matches_frozen_golden(
    variant: str,
    _hardware,
    accuracy_artifact_root: Path,
-    sana_model_parallel,
 ) -> None:
    del _hardware
    from vllm.config import LoadConfig
@@ -189,7 +189,6 @@ def _cfg_prediction(
 def test_sana_video_quick_pipeline_stages_match_diffusers(
    variant: str,
    _hardware,
-    sana_model_parallel,
 ) -> None:
    """Compare the first denoising step without replacing native attention.

@@ -21,7 +21,13 @@ models = ["Qwen/Qwen2.5-Omni-7B"]

 # Single CI deploy YAML; rocm/xpu deltas are picked automatically via the
 # platforms: section in vllm_omni/deploy/ci/qwen2_5_omni.yaml.
-stage_configs = [modify_stage_config(get_deploy_config_path("ci/qwen2_5_omni.yaml"))]
+stage_configs = [
+    modify_stage_config(
+        get_deploy_config_path("ci/qwen2_5_omni.yaml"),
+        # This suite requests two cards; Code2Wav shares the Talker's card.
+        updates={"stages": {2: {"devices": "1"}}},
+    )
+]

 # Create parameter combinations for model and stage config
 test_params = [