[Bugfix] Isolate weekly CPU tests from inherited attention env and hf_api mocks (#7565)

Signed-off-by: wangyu <410167048@qq.com>
Signed-off-by: [Your Name] <your.email@example.com>
Signed-off-by: [Your Name] [Your Email]

tests/diffusion/attention/test_attention_config.py

@@ -42,6 +42,14 @@ from vllm_omni.diffusion.data import (
 pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
 
 
+@pytest.fixture(autouse=True)
+def _clear_inherited_attention_backend(monkeypatch):
+    # Weekly CPU exports DIFFUSION_ATTENTION_BACKEND; OmniDiffusionConfig
+    # treats that as an explicit default so auto-backend cases break.
+    # Tests that setenv the same variable still work: this fixture runs first.
+    monkeypatch.delenv("DIFFUSION_ATTENTION_BACKEND", raising=False)
+
+
 class TestAttentionSpec:
     def test_construct_no_skip_softmax(self):
         spec = AttentionSpec(backend="FLASH_ATTN")
@@ -416,13 +424,6 @@ 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/model_loader/test_hub_prefetch.py

@@ -3,21 +3,21 @@
 
 import contextlib
 
-import huggingface_hub
 import pytest
 
+from tests.helpers.mock import patch_hf_snapshot_download
 from vllm_omni.diffusion.model_loader import hub_prefetch
 
 pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
 
 
-def test_prefetch_subfolders_propagates_revision(monkeypatch):
+def test_prefetch_subfolders_propagates_revision(monkeypatch, tmp_path):
     calls = []
 
-    def fake_snapshot_download(self, **kwargs):
+    def fake_snapshot_download(**kwargs):
         calls.append(kwargs)
 
-    monkeypatch.setattr(huggingface_hub.HfApi, "snapshot_download", fake_snapshot_download)
+    patch_hf_snapshot_download(monkeypatch, fake_snapshot_download, hf_home=tmp_path)
     monkeypatch.setattr(hub_prefetch, "_repo_prefetch_lock", lambda _model: contextlib.nullcontext())
 
     hub_prefetch.prefetch_subfolders(

tests/diffusion/models/hunyuan_image3/test_hunyuan_image3_it2i_multi_image.py

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 """Multi-image input regression for HunyuanImage3 IT2I prompt construction.
 
 The official HunyuanImage-3.0-Instruct supports up to 3 reference images
@@ -41,7 +41,7 @@ pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
 
 
 class FakeTokenizer:
-    """Recording fake tokenizer mirroring the one in test_prompt_utils.
+    """Recording fake tokenizer mirroring the one in test_hunyuan_image3_prompt_utils.
 
     Special token ids: `<|startoftext|>`=1, `<img>`=2, `<think>`=3,
     `<recaption>`=4. encode() returns one id per character starting at

tests/diffusion/models/hunyuan_image3/test_hunyuan_image3_prompt_utils.py

@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
 from __future__ import annotations
 
 import ast

tests/diffusion/models/magi2/test_pipeline_magi2.py

@@ -171,7 +171,7 @@ def test_huggingface_url_resolves_pinned_snapshot(tmp_path, monkeypatch):
         else pytest.fail("unexpected snapshot request")
     )
     monkeypatch.setattr(
-        "vllm.transformers_utils.repo_utils.hf_api",
+        "vllm_omni.transformers_utils.repo_utils.hf_api",
         lambda: api,
     )
     assert _resolve_checkpoint_root(

tests/diffusion/models/minimax_h3/test_minimax_h3_quantization_quality.py

@@ -16,6 +16,7 @@ from tests.diffusion.quantization.test_quantization_quality import (
     _maybe_save_output,
 )
 from tests.helpers.mark import hardware_marks
+from tests.helpers.mock import patch_hf_snapshot_download
 
 _MINIMAX_H3_REPO = "MiniMaxAI/MiniMax-H3"
 _MINIMAX_H3_REVISION = "48d93ede732756e404a3b1b2f3b3a9b5a22f6cfc"
@@ -258,13 +259,13 @@ def test_resolve_fl2va_model_ref(tmp_path, monkeypatch):
     fl2va_root.mkdir()
     (fl2va_root / "model_index.json").write_text("{}", encoding="utf-8")
 
-    def fake_snapshot_download(self, *, repo_id, revision, allow_patterns):
+    def fake_snapshot_download(*, repo_id, revision, allow_patterns):
         assert repo_id == _MINIMAX_H3_REPO
         assert revision == _MINIMAX_H3_REVISION
         assert allow_patterns == ["FL2VA/**"]
         return str(tmp_path)
 
-    monkeypatch.setattr("huggingface_hub.HfApi.snapshot_download", fake_snapshot_download)
+    patch_hf_snapshot_download(monkeypatch, fake_snapshot_download, hf_home=tmp_path)
     assert _resolve_fl2va_model_ref() == str(fl2va_root)