[Refactor][CI] Reorganize test tree, and move slow e2e to weekly (#4951)

Signed-off-by: wangyu <410167048@qq.com>

tests/diffusion/attention/test_attention_config.py

@@ -288,6 +288,13 @@ class TestBuildAttentionConfig:
 class TestOmniDiffusionConfigAttentionParsing:
     """Test OmniDiffusionConfig attention shorthand and structured config."""
 
+    @pytest.fixture(autouse=True)
+    def _clear_diffusion_attention_backend_env(self, monkeypatch):
+        # OmniDiffusionConfig.__post_init__ applies DIFFUSION_ATTENTION_BACKEND via
+        # build_attention_config(); clear it so these tests assert config defaults,
+        # not whatever the process inherited from CI / sibling tests.
+        monkeypatch.delenv("DIFFUSION_ATTENTION_BACKEND", raising=False)
+
     def test_diffusion_attention_backend_sets_default(self):
         config = OmniDiffusionConfig.from_kwargs(diffusion_attention_backend="SAGE_ATTN")
         assert isinstance(config.diffusion_attention_config, AttentionConfig)

tests/diffusion/attention/test_attention_sp.py

@@ -49,6 +49,8 @@ from vllm_omni.diffusion.distributed.parallel_state import (
 from vllm_omni.diffusion.forward_context import set_forward_context
 from vllm_omni.platforms import current_omni_platform
 
+pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
+
 
 def update_environment_variables(envs_dict: dict[str, str]):
     """Update multiple environment variables with logging."""

tests/diffusion/attention/test_flash_attn.py

@@ -18,6 +18,9 @@ from vllm_omni.diffusion.attention.backends.sdpa import SDPAImpl
 from vllm_omni.diffusion.attention.backends.utils import fa  # noqa: E402
 from vllm_omni.platforms import current_omni_platform
 
+pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
+
+
 is_gpu = current_omni_platform.is_cuda_alike() or current_omni_platform.is_xpu()
 HAS_FLASH_ATTN = fa.HAS_FLASH_ATTN
 flash_attn_func = fa.flash_attn_func  # noqa: N813

tests/diffusion/attention/test_piecewise_attn.py

@@ -27,6 +27,9 @@ from vllm_omni.diffusion.attention.backends.utils.piecewise_attn import (
     piecewise_attn,
 )
 
+pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
+
+
 DEVICE = torch.device("cpu")
 
 

tests/diffusion/attention/test_sage_attn3.py

@@ -9,6 +9,9 @@ import pytest
 import torch
 from vllm.platforms import current_platform
 
+pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
+
+
 SAGE_ATTN3_MODULE = "vllm_omni.diffusion.attention.backends.sage_attn3"
 
 

tests/diffusion/attention/test_ulysses_uaa.py

@@ -22,6 +22,8 @@ from vllm_omni.diffusion.distributed.parallel_state import (
 from vllm_omni.diffusion.forward_context import get_forward_context, set_forward_context
 from vllm_omni.platforms import current_omni_platform
 
+pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
+
 
 def _find_free_port() -> int:
     with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s: