Add vLLM-Omni library info to Hugging Face Hub requests (#5381)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>

tests/benchmarks/test_accuracy_bench_utils.py

@@ -1,9 +1,11 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 # ruff: noqa: E402, I001
 import argparse
 import math
 import os
 import sys
-import types
 from pathlib import Path
 
 import pytest
@@ -106,11 +108,9 @@ def test_resolve_seed_tts_root_downloads_only_requested_locale(monkeypatch, tmp_
         captured["allow_patterns"] = allow_patterns
         return str(downloaded_root)
 
-    monkeypatch.setitem(
-        sys.modules,
-        "huggingface_hub",
-        types.SimpleNamespace(snapshot_download=fake_snapshot_download),
-    )
+    from vllm_omni.transformers_utils import repo_utils
+
+    monkeypatch.setattr(repo_utils.hf_api(), "snapshot_download", fake_snapshot_download)
 
     resolved = resolve_seed_tts_root(
         "zhaochenyang20/seed-tts-eval",

tests/dfx/perf/scripts/run_diffusion_benchmark.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 """
 Performance benchmark CI runner for diffusion models.
 
@@ -356,6 +359,8 @@ def _resolve_offline_model(model: str) -> str:
     """
     import huggingface_hub
 
+    from vllm_omni.transformers_utils.repo_utils import hf_api
+
     if not model or os.path.isdir(model):
         return model
 
@@ -380,9 +385,7 @@ def _resolve_offline_model(model: str) -> str:
     if len(parts) >= 3:
         repo_id = "/".join(parts[:2])
         subfolder = "/".join(parts[2:])
-        from huggingface_hub import snapshot_download
-
-        snapshot_root = snapshot_download(
+        snapshot_root = hf_api().snapshot_download(
             repo_id,
             allow_patterns=[f"{subfolder}/**"],
             local_files_only=huggingface_hub.constants.HF_HUB_OFFLINE,
@@ -391,9 +394,7 @@ def _resolve_offline_model(model: str) -> str:
 
     if not huggingface_hub.constants.HF_HUB_OFFLINE:
         return model
-    from huggingface_hub import snapshot_download
-
-    return snapshot_download(model, local_files_only=True)
+    return hf_api().snapshot_download(model, local_files_only=True)
 
 
 class DiffusionServer:

tests/diffusion/model_loader/test_diffusers_loader.py

@@ -13,7 +13,6 @@ from unittest.mock import MagicMock
 import pytest
 import torch
 import torch.nn as nn
-from huggingface_hub import snapshot_download
 from safetensors.torch import save_file
 from vllm.config.load import LoadConfig
 
@@ -32,6 +31,7 @@ from vllm_omni.diffusion.models.helios import HeliosPipeline
 from vllm_omni.diffusion.models.host_weight_contract import FinalLayoutModelContract
 from vllm_omni.diffusion.registry import initialize_model
 from vllm_omni.quantization.component_config import ComponentQuantizationConfig
+from vllm_omni.transformers_utils.repo_utils import hf_api
 
 pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
 
@@ -41,7 +41,7 @@ model_path = "hf-internal-testing/tiny-helios-modular-pipe"
 @pytest.fixture(scope="module")
 def prefetch_helios_model():
     """Downloads the tiny helios model prior to running a test."""
-    snapshot_download(model_path)
+    hf_api().snapshot_download(model_path)
 
 
 @pytest.fixture(scope="function")

tests/diffusion/model_loader/test_hub_prefetch.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
 
 import contextlib
 
@@ -14,10 +14,10 @@ pytestmark = [pytest.mark.core_model, pytest.mark.diffusion, pytest.mark.cpu]
 def test_prefetch_subfolders_propagates_revision(monkeypatch):
     calls = []
 
-    def fake_snapshot_download(**kwargs):
+    def fake_snapshot_download(self, **kwargs):
         calls.append(kwargs)
 
-    monkeypatch.setattr(huggingface_hub, "snapshot_download", fake_snapshot_download)
+    monkeypatch.setattr(huggingface_hub.HfApi, "snapshot_download", fake_snapshot_download)
     monkeypatch.setattr(hub_prefetch, "_repo_prefetch_lock", lambda _model: contextlib.nullcontext())
 
     hub_prefetch.prefetch_subfolders(

tests/diffusion/models/cosmos3/test_cosmos3_sound_tokenizer.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
 
@@ -81,9 +81,8 @@ def test_from_config_loads_local_diffusers_component(tmp_path, monkeypatch: pyte
 
 
 def test_from_config_downloads_component_from_hf_repo(tmp_path, monkeypatch: pytest.MonkeyPatch) -> None:
-    import huggingface_hub
-
     from vllm_omni.diffusion.models.cosmos3 import sound_tokenizer
+    from vllm_omni.transformers_utils import repo_utils
 
     cache_dir = tmp_path / "hf"
     _write_component(cache_dir, checkpoint_name=DIFFUSERS_SOUND_TOKENIZER_CHECKPOINT_NAME)
@@ -95,7 +94,7 @@ def test_from_config_downloads_component_from_hf_repo(tmp_path, monkeypatch: pyt
         calls.append((repo_id, revision, allow_patterns))
         return str(cache_dir)
 
-    monkeypatch.setattr(huggingface_hub, "snapshot_download", fake_snapshot_download)
+    monkeypatch.setattr(repo_utils.hf_api(), "snapshot_download", fake_snapshot_download)
 
     sound_tokenizer.Cosmos3SoundTokenizer.from_config(
         SimpleNamespace(

tests/diffusion/models/ltx2/test_ltx2_pipeline.py

@@ -10,6 +10,7 @@ from dataclasses import replace
 from types import SimpleNamespace
 from typing import Any
 
+import huggingface_hub
 import numpy as np
 import pytest
 import torch
@@ -302,11 +303,11 @@ def test_ltx_artifact_uses_source_revision_and_hub_fallback(
     filename = "ltx-sidecar.safetensors"
     calls = []
 
-    def fake_download(**kwargs):
+    def fake_download(self, **kwargs):
         calls.append(kwargs)
         return "/cache/ltx-sidecar.safetensors"
 
-    monkeypatch.setattr(ltx2_components, "hf_hub_download", fake_download)
+    monkeypatch.setattr(huggingface_hub.HfApi, "hf_hub_download", fake_download)
 
     assert (
         resolve_ltx_artifact(
@@ -331,7 +332,9 @@ def test_ltx_artifact_prefers_model_root(tmp_path, monkeypatch):
     filename = "ltx-sidecar.safetensors"
     expected = tmp_path / filename
     expected.write_bytes(b"sidecar")
-    monkeypatch.setattr(ltx2_components, "hf_hub_download", lambda **_kwargs: pytest.fail("unexpected Hub lookup"))
+    monkeypatch.setattr(
+        huggingface_hub.HfApi, "hf_hub_download", lambda *_args, **_kwargs: pytest.fail("unexpected Hub lookup")
+    )
 
     assert resolve_ltx_artifact(
         str(tmp_path),
@@ -345,11 +348,11 @@ def test_ltx_artifact_prefers_model_root(tmp_path, monkeypatch):
 def test_ltx_artifact_local_model_missing_sidecar_falls_back_to_hub(tmp_path, monkeypatch):
     calls = []
 
-    def fake_download(**kwargs):
+    def fake_download(self, **kwargs):
         calls.append(kwargs)
         return "/cache/ltx-sidecar.safetensors"
 
-    monkeypatch.setattr(ltx2_components, "hf_hub_download", fake_download)
+    monkeypatch.setattr(huggingface_hub.HfApi, "hf_hub_download", fake_download)
 
     assert (
         resolve_ltx_artifact(

tests/diffusion/models/minimax_h3/test_minimax_h3_quantization_quality.py

@@ -42,10 +42,10 @@ _QUALITY_CONFIG = QualityTestConfig(
 
 
 def _resolve_fl2va_model_ref() -> str:
-    from huggingface_hub import snapshot_download
+    from vllm_omni.transformers_utils.repo_utils import hf_api
 
     repo_root = Path(
-        snapshot_download(
+        hf_api().snapshot_download(
             repo_id=_MINIMAX_H3_REPO,
             revision=_MINIMAX_H3_REVISION,
             allow_patterns=["FL2VA/**"],
@@ -253,13 +253,13 @@ 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(*, repo_id, revision, allow_patterns):
+    def fake_snapshot_download(self, *, 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.snapshot_download", fake_snapshot_download)
+    monkeypatch.setattr("huggingface_hub.HfApi.snapshot_download", fake_snapshot_download)
     assert _resolve_fl2va_model_ref() == str(fl2va_root)
 
 

tests/e2e/accuracy/ltx/test_ltx_official_similarity.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
 
 """E2E accuracy guard against a pinned Lightricks LTX pipeline revision.
 
@@ -22,11 +22,11 @@ from typing import Literal
 import numpy as np
 import pytest
 import torch
-from huggingface_hub import hf_hub_download, snapshot_download
 from torchmetrics.image import PeakSignalNoiseRatio, StructuralSimilarityIndexMeasure
 
 from tests.e2e.accuracy.helpers import reset_artifact_dir
 from tests.helpers.mark import hardware_test
+from vllm_omni.transformers_utils.repo_utils import hf_api
 
 OFFICIAL_REPOSITORY = "https://github.com/Lightricks/LTX-2.git"
 OFFICIAL_REVISION = "9377758131b1ffde4b7f766804590a6617bf2ab9"
@@ -326,7 +326,7 @@ def _resolve_model(case: LTXAccuracyCase) -> Path:
     if revision is None and model_id == case.model_id:
         revision = case.model_revision
     return Path(
-        snapshot_download(
+        hf_api().snapshot_download(
             repo_id=model_id,
             revision=revision,
             allow_patterns=[
@@ -359,7 +359,7 @@ def _resolve_gemma_root(case: LTXAccuracyCase, model: Path) -> Path:
         if configured_model and Path(configured_model).is_dir():
             return Path(configured_model)
         return Path(
-            snapshot_download(
+            hf_api().snapshot_download(
                 repo_id=case.gemma_model_id,
                 revision=case.gemma_model_revision,
                 allow_patterns=[
@@ -383,7 +383,7 @@ def _resolve_artifact(artifact: LTXArtifact, model: Path | None = None) -> Path:
         if model_path.is_file():
             return model_path
     return Path(
-        hf_hub_download(
+        hf_api().hf_hub_download(
             repo_id=artifact.repo_id,
             repo_type=None if artifact.repo_type == "model" else artifact.repo_type,
             filename=artifact.filename,

tests/e2e/features/rlhf_test/test_verl_omni_e2e.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
 
 """E2E test that follows the EXACT same flow as
 ``verl-omni/tests/workers/rollout/rollout_vllm/test_vllm_omni_generate.py``
@@ -35,7 +35,6 @@ from uuid import uuid4
 import pytest
 import ray
 import torch
-from huggingface_hub import snapshot_download
 from omegaconf import OmegaConf
 from transformers import AutoTokenizer
 
@@ -46,6 +45,7 @@ from tests.e2e.features.helpers.verl_omni_server import (
     vLLMOmniHttpServerLocal,
 )
 from tests.helpers.mark import hardware_test
+from vllm_omni.transformers_utils.repo_utils import hf_api
 
 MODEL = "tiny-random/Qwen-Image"
 TOKENIZER_MODEL = "Qwen/Qwen2-1.5B-Instruct"
@@ -55,7 +55,7 @@ _MIN_PROMPT_TOKENS = 35
 def _resolve_model_path(repo_id: str) -> str:
     if os.path.isdir(repo_id):
         return repo_id
-    return snapshot_download(repo_id=repo_id)
+    return hf_api().snapshot_download(repo_id=repo_id)
 
 
 @lru_cache(maxsize=1)

tests/e2e/offline_inference/test_cosyvoice3_expansion.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
 """
 Offline E2E smoke test for CosyVoice3 zero-shot reference inference.
 
@@ -20,7 +20,6 @@ from pathlib import Path
 import numpy as np
 import pytest
 import soundfile as sf
-from huggingface_hub import snapshot_download
 from vllm.sampling_params import SamplingParams
 
 from tests.helpers.mark import hardware_test
@@ -29,6 +28,7 @@ from tests.helpers.runtime import OmniRunner
 from tests.helpers.stage_config import get_deploy_config_path
 from vllm_omni.model_executor.models.cosyvoice3.tokenizer import get_qwen_tokenizer
 from vllm_omni.transformers_utils.configs.cosyvoice3 import CosyVoice3Config
+from vllm_omni.transformers_utils.repo_utils import hf_api
 
 MODEL = "FunAudioLLM/Fun-CosyVoice3-0.5B-2512"
 MODEL_DIR_ENV = "VLLM_OMNI_COSYVOICE3_MODEL_DIR"
@@ -58,7 +58,7 @@ def _resolve_model_dir() -> Path:
     override = os.environ.get(MODEL_DIR_ENV)
     if override:
         return Path(override).expanduser().resolve()
-    return Path(snapshot_download(MODEL, allow_patterns=["*"]))
+    return Path(hf_api().snapshot_download(MODEL, allow_patterns=["*"]))
 
 
 def _reference_zero_shot_stage0_sampling(*, text: str) -> SamplingParams:

tests/e2e/offline_inference/test_mammoth_moda2_expansion.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 """
 End-to-end test for MammothModa2 text-to-image generation.
 
@@ -19,12 +22,12 @@ from pathlib import Path
 
 import pytest
 import torch
-from huggingface_hub import snapshot_download
 from vllm.sampling_params import SamplingParams
 
 from tests.helpers.mark import hardware_test
 from tests.helpers.runtime import OmniRunner
 from tests.helpers.stage_config import get_deploy_config_path
+from vllm_omni.transformers_utils.repo_utils import hf_api
 
 # ---------------------------------------------------------------------------
 # Constants
@@ -64,7 +67,7 @@ _PIXEL_SAMPLE_COORDS = [
 # Helpers
 # ---------------------------------------------------------------------------
 def _load_t2i_gen_config(repo_id: str) -> dict:
-    weights_dir = Path(snapshot_download(repo_id))
+    weights_dir = Path(hf_api().snapshot_download(repo_id))
     cfg_path = weights_dir / "t2i_generation_config.json"
     if not cfg_path.exists():
         pytest.skip(f"t2i_generation_config.json not found at {cfg_path}")

tests/e2e/offline_inference/test_moss_tts_realtime.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
 """E2E offline inference tests for MOSS-TTS-Realtime (MossTTSRealtime, 1.7B).
 
 Uses the standard omni_runner + pytestmark pattern (one module-scoped engine
@@ -160,12 +160,13 @@ def _build_request(ref_audio_path: str, text: str) -> dict:
     Frees the codec model before returning to avoid competing with the
     running vllm engine for GPU/CPU memory.
     """
-    from huggingface_hub import snapshot_download
     from transformers import AutoModel, AutoTokenizer
 
+    from vllm_omni.transformers_utils.repo_utils import hf_api
+
     # Step 1: locate the realtime processor module in the snapshot.
     try:
-        snap_dir = Path(snapshot_download(repo_id=MODEL))
+        snap_dir = Path(hf_api().snapshot_download(repo_id=MODEL))
     except Exception as exc:
         msg = f"Cannot locate snapshot for {MODEL}: {exc}"
         if os.environ.get("MOSS_TTS_SKIP_ON_NET_FAIL"):

tests/e2e/online_serving/helpers/minicpmo_4_5_duplex.py

@@ -12,7 +12,6 @@ from pathlib import Path
 from types import SimpleNamespace
 
 import pytest
-from huggingface_hub import snapshot_download
 
 from tests.helpers.runtime import OmniServerParams, get_model_prefix
 from tests.helpers.stage_config import (
@@ -20,6 +19,7 @@ from tests.helpers.stage_config import (
     get_deploy_duplex_max_sessions,
     modify_stage_config,
 )
+from vllm_omni.transformers_utils.repo_utils import hf_api
 
 MODEL = "openbmb/MiniCPM-o-4_5"
 DEPLOY_CONFIG_REL = "minicpmo_4_5.yaml"
@@ -111,7 +111,7 @@ def resolve_ref_audio() -> Path:
     model_prefix = get_model_prefix()
     model_root = Path(model_prefix) / MODEL if model_prefix else Path(MODEL)
     if not model_root.is_dir():
-        model_root = Path(snapshot_download(MODEL, local_files_only=True))
+        model_root = Path(hf_api().snapshot_download(MODEL, local_files_only=True))
     ref_audio = model_root / REF_AUDIO_RELATIVE_PATH
     if not ref_audio.is_file():
         raise FileNotFoundError(f"MiniCPM-o checkpoint ref audio is missing: {ref_audio}")

tests/engine/test_arg_utils.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 """
 Tests for Omni config utils. For stability, these tests should largely be
 invariant to the specific attributes of vLLM config except in cases where we
@@ -63,7 +66,7 @@ def test_full_payload_capability_reaches_omni_model_config(monkeypatch):
     class ConnectorWorker:
         model_runner_cls = ConnectorRunner
 
-    worker_module.ConnectorWorker = ConnectorWorker
+    setattr(worker_module, "ConnectorWorker", ConnectorWorker)
     monkeypatch.setitem(sys.modules, worker_module.__name__, worker_module)
 
     monkeypatch.setattr(OmniEngineArgs, "_ensure_omni_models_registered", lambda _self: None)
@@ -95,7 +98,7 @@ def test_stage_without_connector_configuration_accepts_plain_runner(monkeypatch)
     class WorkerWithoutConnector:
         model_runner_cls = object
 
-    worker_module.WorkerWithoutConnector = WorkerWithoutConnector
+    setattr(worker_module, "WorkerWithoutConnector", WorkerWithoutConnector)
     monkeypatch.setitem(sys.modules, worker_module.__name__, worker_module)
 
     args = OmniEngineArgs(
@@ -113,7 +116,7 @@ def test_full_payload_capability_requires_selected_worker_connector(monkeypatch)
     class WorkerWithoutConnector:
         model_runner_cls = object
 
-    worker_module.WorkerWithoutConnector = WorkerWithoutConnector
+    setattr(worker_module, "WorkerWithoutConnector", WorkerWithoutConnector)
     monkeypatch.setitem(sys.modules, worker_module.__name__, worker_module)
 
     with pytest.raises(ValueError, match="does not provide an Omni connector model runner"):
@@ -130,7 +133,7 @@ def test_full_payload_capability_validates_platform_selected_worker(monkeypatch)
     class WorkerWithoutConnector:
         model_runner_cls = object
 
-    worker_module.WorkerWithoutConnector = WorkerWithoutConnector
+    setattr(worker_module, "WorkerWithoutConnector", WorkerWithoutConnector)
     monkeypatch.setitem(sys.modules, worker_module.__name__, worker_module)
     monkeypatch.setattr(
         current_omni_platform,
@@ -293,7 +296,7 @@ def test_remote_tokenizer_subfolder_download_does_not_report_failure(tmp_path, m
     baseline_config = Mock()
     warning = mocker.patch("vllm_omni.engine.arg_utils.logger.warning")
 
-    monkeypatch.setattr("huggingface_hub.snapshot_download", lambda *args, **kwargs: str(tmp_path))
+    monkeypatch.setattr("huggingface_hub.HfApi.snapshot_download", lambda *args, **kwargs: str(tmp_path))
     monkeypatch.setattr(OmniEngineArgs, "_patch_empty_hf_config", lambda *args, **kwargs: None)
     monkeypatch.setattr(EngineArgs, "create_model_config", lambda _self: baseline_config)
     monkeypatch.setattr(
@@ -339,6 +342,7 @@ def test_patch_missing_local_hf_config(tmp_path):
 def test_non_missing_local_hf_config_error_reaches_parent_loader(tmp_path, monkeypatch, config_entry):
     """Non-missing config errors must reach vLLM's normal loader."""
     config_path = tmp_path / "config.json"
+    loader_error: Exception
     if config_entry == "malformed":
         config_path.write_text("{not valid json", encoding="utf-8")
         loader_error = json.JSONDecodeError("invalid config", "{not valid json", 1)

tests/helpers/__init__.py

@@ -29,12 +29,13 @@ def skip_if_gated_repo_inaccessible(
     and we skip cleanly.
     """
     try:
-        from huggingface_hub import hf_hub_download
         from huggingface_hub.errors import GatedRepoError, RepositoryNotFoundError
+
+        from vllm_omni.transformers_utils.repo_utils import hf_api
     except Exception:
         return
     try:
-        hf_hub_download(repo_id=repo_id, filename=filename, revision=revision)
+        hf_api().hf_hub_download(repo_id=repo_id, filename=filename, revision=revision)
     except GatedRepoError as exc:
         import pytest
 

tests/model_executor/models/minicpmo_4_5/test_code2wav_model_dir.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
 """Unit tests for MiniCPM-o 4.5 code2wav model-dir resolution (#5442).
 
 In hub/CI deployments ``model_config.model`` is a repo id rather than a local
@@ -27,7 +27,7 @@ def _no_hub(monkeypatch):
     def _fail(*args, **kwargs):  # pragma: no cover - must not be reached
         raise AssertionError("snapshot_download must not be called here")
 
-    monkeypatch.setattr(huggingface_hub, "snapshot_download", _fail)
+    monkeypatch.setattr(huggingface_hub.HfApi, "snapshot_download", _fail)
 
 
 def test_local_directory_is_returned_unchanged(tmp_path, monkeypatch):
@@ -38,13 +38,13 @@ def test_local_directory_is_returned_unchanged(tmp_path, monkeypatch):
 def test_repo_id_resolves_via_snapshot_download(tmp_path, monkeypatch):
     calls = {}
 
-    def _fake_snapshot_download(model_ref, revision=None, allow_patterns=None):
+    def _fake_snapshot_download(self, model_ref, revision=None, allow_patterns=None):
         calls["model_ref"] = model_ref
         calls["revision"] = revision
         calls["allow_patterns"] = allow_patterns
         return str(tmp_path / "snapshot")
 
-    monkeypatch.setattr(huggingface_hub, "snapshot_download", _fake_snapshot_download)
+    monkeypatch.setattr(huggingface_hub.HfApi, "snapshot_download", _fake_snapshot_download)
     resolved = _resolve_model_dir("openbmb/MiniCPM-o-4_5", revision="abc123")
     assert resolved == str(tmp_path / "snapshot")
     assert calls["model_ref"] == "openbmb/MiniCPM-o-4_5"
@@ -56,7 +56,7 @@ def test_snapshot_download_failure_propagates(monkeypatch):
     def _raise(*args, **kwargs):
         raise FileNotFoundError("offline and not cached")
 
-    monkeypatch.setattr(huggingface_hub, "snapshot_download", _raise)
+    monkeypatch.setattr(huggingface_hub.HfApi, "snapshot_download", _raise)
     with pytest.raises(FileNotFoundError):
         _resolve_model_dir("openbmb/MiniCPM-o-4_5")
 

tests/model_executor/models/omnivoice/test_nonverbal_tags_tokenize.py

@@ -1,12 +1,15 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 from __future__ import annotations
 
 import pytest
-from huggingface_hub import hf_hub_download
 from tokenizers import Tokenizer as HFTokenizer
 
 from vllm_omni.diffusion.models.omnivoice.pipeline_omnivoice import (
     _tokenize_with_nonverbal_tags,
 )
+from vllm_omni.transformers_utils.repo_utils import hf_api
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
 
@@ -18,7 +21,7 @@ def _ids(tok: HFTokenizer, s: str) -> list[int]:
 class TestNonVerbalTags:
     @classmethod
     def setup_class(cls):
-        tokenizer_path = hf_hub_download(repo_id="k2-fsa/OmniVoice", filename="tokenizer.json")
+        tokenizer_path = hf_api().hf_hub_download(repo_id="k2-fsa/OmniVoice", filename="tokenizer.json")
         cls.tokenizer = HFTokenizer.from_file(tokenizer_path)
 
     def test_plain_text_fallback(self):