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/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}")