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/diffusion/models/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
 """Cosmos3 sound tokenizer integration.
 
 The tokenizer decodes model-generated sound latents for video+audio output. It
@@ -338,9 +338,9 @@ class Cosmos3SoundTokenizer:
 
         model_root = str(model_path) if model_path and os.path.isdir(model_path) else None
         if model_root is None and model_path and not explicit_avae_path:
-            from huggingface_hub import snapshot_download
+            from vllm_omni.transformers_utils.repo_utils import hf_api
 
-            model_root = snapshot_download(
+            model_root = hf_api().snapshot_download(
                 repo_id=str(model_path),
                 revision=getattr(od_config, "revision", None),
                 allow_patterns=[

vllm_omni/diffusion/models/dreamzero/pipeline_dreamzero.py

@@ -21,7 +21,6 @@ from contextlib import contextmanager
 import numpy as np
 import torch
 import torch.nn as nn
-from huggingface_hub import hf_hub_download
 from transformers import AutoTokenizer, UMT5Config, UMT5EncoderModel
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 
@@ -67,6 +66,7 @@ from vllm_omni.experimental.world_models.session_state import (
     resolve_session_state_config,
 )
 from vllm_omni.inputs.data import OmniDiffusionSamplingParams
+from vllm_omni.transformers_utils.repo_utils import hf_api
 
 logger = logging.getLogger(__name__)
 MAX_DREAMZERO_SESSIONS = 64
@@ -581,7 +581,7 @@ class DreamZeroPipeline(nn.Module, CFGParallelMixin):
                 return json.load(f)
 
         try:
-            json_path = hf_hub_download(model_path, relative_path)
+            json_path = hf_api().hf_hub_download(model_path, relative_path)
             with open(json_path) as f:
                 return json.load(f)
         except Exception:

vllm_omni/diffusion/models/helios/pipeline_helios.py

@@ -76,9 +76,9 @@ def load_json_config(model_path: str, subfolder: str, filename: str, local_files
                 return json.load(f)
     else:
         try:
-            from huggingface_hub import hf_hub_download
+            from vllm_omni.transformers_utils.repo_utils import hf_api
 
-            config_path = hf_hub_download(
+            config_path = hf_api().hf_hub_download(
                 repo_id=model_path,
                 filename=f"{subfolder}/{filename}",
             )

vllm_omni/diffusion/models/longcat_video/pipeline_longcat_video_avatar.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
 
@@ -376,9 +376,9 @@ def _ensure_local_dir(model: str | os.PathLike[str], allow_patterns: list[str] |
     model_path = Path(model)
     if model_path.exists():
         return model_path
-    from huggingface_hub import snapshot_download
+    from vllm_omni.transformers_utils.repo_utils import hf_api
 
-    return Path(snapshot_download(str(model), allow_patterns=allow_patterns))
+    return Path(hf_api().snapshot_download(str(model), allow_patterns=allow_patterns))
 
 
 def _ensure_avatar_model_index(model_dir: Path, use_int8: bool) -> None:

vllm_omni/diffusion/models/ltx2/ltx2_components.py

@@ -20,7 +20,6 @@ from diffusers.pipelines.ltx2 import LTX2TextConnectors
 from diffusers.pipelines.ltx2.latent_upsampler import LTX2LatentUpsamplerModel
 from diffusers.pipelines.ltx2.vocoder import LTX2Vocoder
 from diffusers.video_processor import VideoProcessor
-from huggingface_hub import hf_hub_download
 from safetensors import safe_open
 from safetensors.torch import load_file
 from transformers import AutoModelForImageTextToText, AutoTokenizer, Gemma3ForConditionalGeneration
@@ -32,6 +31,7 @@ from vllm_omni.diffusion.distributed.utils import get_local_device
 from vllm_omni.diffusion.model_loader.diffusers_loader import DiffusersPipelineLoader
 from vllm_omni.diffusion.model_loader.hub_prefetch import from_pretrained_with_prefetch, prefetch_subfolders
 from vllm_omni.diffusion.offloader.module_collector import ModuleDiscovery
+from vllm_omni.transformers_utils.repo_utils import hf_api
 
 if TYPE_CHECKING:
     from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
@@ -279,7 +279,7 @@ def resolve_ltx_artifact(
     # independently pinned artifact revision.
     revision = model_revision if model == repo_id else artifact_revision
     try:
-        return hf_hub_download(
+        return hf_api().hf_hub_download(
             repo_id=repo_id,
             filename=filename,
             revision=revision,
@@ -351,7 +351,7 @@ def _load_ltx_metadata_json(model: str, filename: str, revision: str | None = No
             return {}
     else:
         try:
-            path = hf_hub_download(repo_id=model, filename=filename, revision=revision)
+            path = hf_api().hf_hub_download(repo_id=model, filename=filename, revision=revision)
         except Exception:
             return {}
     try:
@@ -529,7 +529,7 @@ def _detect_vocoder_output_sample_rate(model: str, revision: str | None = None)
     vocoder_config_path = os.path.join(model, "vocoder", "config.json")
     if not os.path.exists(vocoder_config_path):
         try:
-            vocoder_config_path = hf_hub_download(model, "vocoder/config.json", revision=revision)
+            vocoder_config_path = hf_api().hf_hub_download(model, "vocoder/config.json", revision=revision)
         except Exception:
             return None
     try:
@@ -839,7 +839,7 @@ def load_transformer_config(
         if not os.path.exists(config_path):
             raise FileNotFoundError(f"LTX transformer config not found: {config_path}")
     else:
-        config_path = hf_hub_download(
+        config_path = hf_api().hf_hub_download(
             repo_id=model_path,
             filename=f"{subfolder}/config.json",
             revision=revision,

vllm_omni/diffusion/models/omnivoice/pipeline_omnivoice.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
 """
 OmniVoice TTS Pipeline for vLLM-Omni diffusion engine.
 
@@ -141,9 +141,9 @@ class OmniVoicePipeline(nn.Module, SupportAudioOutput):
 
         # Resolve model path (HF hub ID → local cache)
         if not os.path.isdir(self.model_path):
-            from huggingface_hub import snapshot_download
+            from vllm_omni.transformers_utils.repo_utils import hf_api
 
-            self.model_path = snapshot_download(self.model_path)
+            self.model_path = hf_api().snapshot_download(self.model_path)
 
         # Load OmniVoice config
         config_path = os.path.join(self.model_path, "config.json")

vllm_omni/diffusion/models/pi0/pipeline_pi0.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
 """π0 (Pi-Zero) VLA pipeline for vllm-omni.
 
 Entry point for ``DiffusionEngine.step() → pipeline.forward(req)``. Mirrors the
@@ -94,9 +94,9 @@ class Pi0Pipeline(nn.Module):
             return None
         if os.path.isdir(model):
             return model
-        from huggingface_hub import snapshot_download
+        from vllm_omni.transformers_utils.repo_utils import hf_api
 
-        return snapshot_download(
+        return hf_api().snapshot_download(
             repo_id=model,
             allow_patterns=["*.json", "*.safetensors", "*.model", "tokenizer*"],
         )

vllm_omni/diffusion/models/sensenova_u1/pipeline_sensenova_u1.py

@@ -27,7 +27,6 @@ import numpy as np
 import torch
 import torch.nn as nn
 import torchvision.transforms as T
-from huggingface_hub import snapshot_download
 from PIL import Image
 from transformers import AutoTokenizer
 from vllm.logger import init_logger
@@ -109,7 +108,9 @@ def _resolve_model_path(model_path: str) -> str:
     """Resolve a HuggingFace model ID or local path to a local directory."""
     if os.path.isdir(model_path):
         return model_path
-    return snapshot_download(model_path)
+    from vllm_omni.transformers_utils.repo_utils import hf_api
+
+    return hf_api().snapshot_download(model_path)
 
 
 def _round_by_factor(number: float, factor: int) -> int:

vllm_omni/diffusion/models/utils.py

@@ -168,8 +168,8 @@ def _load_json(model_path: str, filename: str, local_files_only: bool = True) ->
         with open(path) as f:
             return json.load(f)
     else:
-        from huggingface_hub import hf_hub_download
+        from vllm_omni.transformers_utils.repo_utils import hf_api
 
-        cached = hf_hub_download(repo_id=model_path, filename=filename)
+        cached = hf_api().hf_hub_download(repo_id=model_path, filename=filename)
         with open(cached) as f:
             return json.load(f)

vllm_omni/diffusion/models/wan2_2/pipeline_wan2_2.py

@@ -156,9 +156,9 @@ def load_transformer_config(model_path: str, subfolder: str = "transformer", loc
     else:
         # Try to download config from HF Hub
         try:
-            from huggingface_hub import hf_hub_download
+            from vllm_omni.transformers_utils.repo_utils import hf_api
 
-            config_path = hf_hub_download(
+            config_path = hf_api().hf_hub_download(
                 repo_id=model_path,
                 filename=f"{subfolder}/config.json",
             )
@@ -364,9 +364,9 @@ class Wan22Pipeline(
         else:
             # For remote models, download and read model_index.json
             try:
-                from huggingface_hub import hf_hub_download
+                from vllm_omni.transformers_utils.repo_utils import hf_api
 
-                model_index_path = hf_hub_download(repo_id=model, filename="model_index.json")
+                model_index_path = hf_api().hf_hub_download(repo_id=model, filename="model_index.json")
                 with open(model_index_path) as f:
                     model_index = json.load(f)
                     self.expand_timesteps = model_index.get("expand_timesteps", False)