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>

vllm_omni/benchmarks/data_modules/daily_omni_dataset.py

@@ -170,7 +170,7 @@ def ensure_daily_omni_hub_videos_dir(repo_id: str) -> Path:
         raise ValueError("repo_id is required to download Daily-Omni Videos.tar")
 
     try:
-        from huggingface_hub import hf_hub_download
+        from vllm_omni.transformers_utils.repo_utils import hf_api
     except ImportError as e:
         raise ImportError(
             "Daily-Omni Hub media download requires huggingface_hub. "
@@ -180,7 +180,7 @@ def ensure_daily_omni_hub_videos_dir(repo_id: str) -> Path:
     tar_path: Path | None = None
     for fname in ("Videos.tar", "videos.tar"):
         try:
-            tar_path = Path(hf_hub_download(repo_id=rid, filename=fname, repo_type="dataset"))
+            tar_path = Path(hf_api().hf_hub_download(repo_id=rid, filename=fname, repo_type="dataset"))
             break
         except Exception:
             continue
@@ -237,14 +237,14 @@ def ensure_daily_omni_hub_root(repo_id: str) -> Path:
         raise ValueError("repo_id is required to download Daily-Omni from Hugging Face")
 
     try:
-        from huggingface_hub import snapshot_download
+        from vllm_omni.transformers_utils.repo_utils import hf_api
     except ImportError as e:
         raise ImportError(
             "Install huggingface_hub to load Daily-Omni from the Hub, or pass a local "
             "--daily-omni-qa-json / --daily-omni-video-dir mirror."
         ) from e
 
-    cache = snapshot_download(
+    cache = hf_api().snapshot_download(
         repo_id=rid,
         repo_type="dataset",
         allow_patterns=_DAILY_OMNI_QA_ALLOW_PATTERNS,

vllm_omni/benchmarks/data_modules/seed_tts_dataset.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 """Seed-TTS zero-shot evaluation-style prompts for ``vllm bench serve``.
 
 Loads rows from the `meta.lst` format used in `BytedanceSpeech/seed-tts-eval`_ (or any
@@ -121,7 +124,7 @@ def resolve_seed_tts_root(dataset_path: str | None, *, explicit_root: str | None
 
     repo_id = dataset_path.strip()
     try:
-        from huggingface_hub import snapshot_download
+        from vllm_omni.transformers_utils.repo_utils import hf_api
     except ImportError as e:
         raise ImportError(
             "Install huggingface_hub to download Seed-TTS from the Hub, or clone the dataset "
@@ -130,7 +133,7 @@ def resolve_seed_tts_root(dataset_path: str | None, *, explicit_root: str | None
     # Download only the requested locale subtree instead of the whole dataset
     # repo. This avoids large, flaky nightly downloads when we only need e.g.
     # ``en/meta.lst`` + ``en/prompt-wavs/**``.
-    cache = snapshot_download(
+    cache = hf_api().snapshot_download(
         repo_id=repo_id,
         repo_type="dataset",
         allow_patterns=[f"{locale}/**"],

vllm_omni/benchmarks/data_modules/seed_tts_eval.py

@@ -1,3 +1,6 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+
 """Seed-TTS WER aligned with Bytedance ``seed-tts-eval`` / ``run_wer.py``.
 
 Matches the published protocol (see Hugging Face dataset card and
@@ -329,7 +332,8 @@ def _ensure_utmos_jit_model() -> Any | None:
             return _utmos_jit_model
         try:
             import torch
-            from huggingface_hub import hf_hub_download
+
+            from vllm_omni.transformers_utils.repo_utils import hf_api
 
             repo = os.environ.get("SEED_TTS_UTMOS_HF_REPO", "balacoon/utmos").strip() or "balacoon/utmos"
             fname = os.environ.get("SEED_TTS_UTMOS_JIT_FILE", "utmos.jit").strip() or "utmos.jit"
@@ -338,7 +342,7 @@ def _ensure_utmos_jit_model() -> Any | None:
                 repo,
                 fname,
             )
-            path = hf_hub_download(repo_id=repo, filename=fname, repo_type="model")
+            path = hf_api().hf_hub_download(repo_id=repo, filename=fname, repo_type="model")
 
             # TODO The model weights in UTMOS must be loaded in cuda:0; otherwise, the model execution will fail.
             want = "cuda:0"