diff --git a/tests/lora/test_utils.py b/tests/lora/test_utils.py
index bec12eeeb48d..4a9ba66069d7 100644
--- a/tests/lora/test_utils.py
+++ b/tests/lora/test_utils.py
@@ -175,7 +175,7 @@ def test_get_adapter_absolute_path_local_existing(mock_abspath, mock_exist):
     assert get_adapter_absolute_path(path) == absolute_path
 
 
-@patch("huggingface_hub.snapshot_download")
+@patch("huggingface_hub.HfApi.snapshot_download")
 @patch("os.path.exists")
 def test_get_adapter_absolute_path_huggingface(mock_exist, mock_snapshot_download):
     # Hugging Face model identifier
@@ -186,7 +186,7 @@ def test_get_adapter_absolute_path_huggingface(mock_exist, mock_snapshot_downloa
     assert get_adapter_absolute_path(path) == absolute_path
 
 
-@patch("huggingface_hub.snapshot_download")
+@patch("huggingface_hub.HfApi.snapshot_download")
 @patch("os.path.exists")
 def test_get_adapter_absolute_path_huggingface_error(
     mock_exist, mock_snapshot_download
diff --git a/vllm/assets/video.py b/vllm/assets/video.py
index 9ec2e4d16770..72cd196c68f4 100644
--- a/vllm/assets/video.py
+++ b/vllm/assets/video.py
@@ -7,10 +7,10 @@
 
 import numpy as np
 import numpy.typing as npt
-from huggingface_hub import hf_hub_download
 from PIL import Image
 
 from vllm.multimodal.media.audio import load_audio_pyav
+from vllm.transformers_utils.repo_utils import hf_api
 
 from .base import get_cache_dir
 
@@ -27,7 +27,7 @@ def download_video_asset(filename: str) -> str:
     video_path = video_directory / filename
     video_path_str = str(video_path)
     if not video_path.exists():
-        video_path_str = hf_hub_download(
+        video_path_str = hf_api().hf_hub_download(
             repo_id="raushan-testing-hf/videos-test",
             filename=filename,
             repo_type="dataset",
diff --git a/vllm/benchmarks/datasets/datasets.py b/vllm/benchmarks/datasets/datasets.py
index 70545a3b4959..af62f049ff3a 100644
--- a/vllm/benchmarks/datasets/datasets.py
+++ b/vllm/benchmarks/datasets/datasets.py
@@ -31,7 +31,6 @@
 
 import numpy as np
 import pybase64 as base64
-from huggingface_hub import snapshot_download
 from PIL import Image
 from typing_extensions import deprecated
 
@@ -46,6 +45,7 @@
 from vllm.multimodal.audio import get_audio_duration
 from vllm.multimodal.image import convert_image_mode
 from vllm.tokenizers import TokenizerLike
+from vllm.transformers_utils.repo_utils import hf_api
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 from vllm.utils.import_utils import PlaceholderModule
 from vllm.utils.mistral import is_mistral_tokenizer
@@ -3334,7 +3334,10 @@ def __init__(self, **kwargs) -> None:
         self._remote_path_root = (
             f"https://huggingface.co/datasets/{self.hf_name}/resolve/main"
         )
-        self._local_path_root = snapshot_download(self.hf_name, repo_type="dataset")
+        self._local_path_root = hf_api().snapshot_download(
+            self.hf_name,
+            repo_type="dataset",
+        )
 
     def sample(
         self,
diff --git a/vllm/lora/utils.py b/vllm/lora/utils.py
index 7b68c1f952aa..d5c9a1a6ff8a 100644
--- a/vllm/lora/utils.py
+++ b/vllm/lora/utils.py
@@ -4,7 +4,6 @@
 import os
 from typing import TYPE_CHECKING
 
-import huggingface_hub
 import regex as re
 from huggingface_hub.utils import HfHubHTTPError, HFValidationError
 from torch import nn
@@ -37,6 +36,7 @@
 from vllm.model_executor.layers.fused_moe import FusedMoE
 from vllm.model_executor.layers.linear import LinearBase
 from vllm.model_executor.utils import get_moe_expert_mapping, get_packed_modules_mapping
+from vllm.transformers_utils.repo_utils import hf_api
 
 if TYPE_CHECKING:
     from vllm.model_executor.layers.logits_processor import LogitsProcessor
@@ -340,7 +340,9 @@ def get_adapter_absolute_path(lora_path: str) -> str:
         error_log = "Error downloading the ModelScope model"
     else:
         # Otherwise, we assume the path is a Hugging Face Hub repo.
-        download_fn = lambda: huggingface_hub.snapshot_download(repo_id=lora_path)
+        download_fn = lambda: hf_api().snapshot_download(
+            repo_id=lora_path,
+        )
         download_exceptions = (HfHubHTTPError, HFValidationError)
         error_log = "Error downloading the HuggingFace model"
 
diff --git a/vllm/model_executor/model_loader/bitsandbytes_loader.py b/vllm/model_executor/model_loader/bitsandbytes_loader.py
index 81526415ff2d..4d9ae267e084 100644
--- a/vllm/model_executor/model_loader/bitsandbytes_loader.py
+++ b/vllm/model_executor/model_loader/bitsandbytes_loader.py
@@ -10,7 +10,6 @@
 
 import numpy as np
 import torch
-from huggingface_hub import HfApi
 from packaging import version
 from torch import nn
 from transformers.utils import SAFE_WEIGHTS_INDEX_NAME
@@ -48,6 +47,7 @@
     set_weight_attrs,
 )
 from vllm.platforms import current_platform
+from vllm.transformers_utils.repo_utils import hf_api
 from vllm.utils.torch_utils import set_default_torch_dtype
 
 logger = init_logger(__name__)
@@ -97,8 +97,7 @@ def _get_weight_files(
                 if weight_files:
                     return model_name_or_path, weight_files, pattern
         else:
-            hf_api = HfApi()
-            repo_files = hf_api.list_repo_files(repo_id=model_name_or_path)
+            repo_files = hf_api().list_repo_files(repo_id=model_name_or_path)
             for pattern in allowed_patterns:
                 matching_files = fnmatch.filter(repo_files, pattern)
                 if matching_files:
diff --git a/vllm/model_executor/model_loader/gguf_loader.py b/vllm/model_executor/model_loader/gguf_loader.py
index 6148caa98746..2db5efd0e5b3 100644
--- a/vllm/model_executor/model_loader/gguf_loader.py
+++ b/vllm/model_executor/model_loader/gguf_loader.py
@@ -8,7 +8,6 @@
 import regex as re
 import torch
 import torch.nn as nn
-from huggingface_hub import hf_hub_download
 from transformers import AutoModelForCausalLM, AutoModelForImageTextToText
 
 from vllm.config import ModelConfig, VllmConfig
@@ -27,6 +26,7 @@
     gguf_quant_weights_iterator_multi,
 )
 from vllm.transformers_utils.gguf_utils import detect_gguf_multimodal
+from vllm.transformers_utils.repo_utils import hf_api
 from vllm.utils.torch_utils import set_default_torch_dtype
 
 if TYPE_CHECKING:
@@ -57,7 +57,7 @@ def _prepare_weights(self, model_config: ModelConfig):
         # repo id/filename.gguf
         if "/" in model_name_or_path and model_name_or_path.endswith(".gguf"):
             repo_id, filename = model_name_or_path.rsplit("/", 1)
-            return hf_hub_download(
+            return hf_api().hf_hub_download(
                 repo_id=repo_id,
                 filename=filename,
                 revision=model_config.revision,
diff --git a/vllm/model_executor/model_loader/tensorizer.py b/vllm/model_executor/model_loader/tensorizer.py
index 37d37d55f543..736b2134604b 100644
--- a/vllm/model_executor/model_loader/tensorizer.py
+++ b/vllm/model_executor/model_loader/tensorizer.py
@@ -16,7 +16,6 @@
 
 import regex as re
 import torch
-from huggingface_hub import snapshot_download
 from torch import nn
 from torch.utils._python_dispatch import TorchDispatchMode
 from transformers import PretrainedConfig
@@ -26,6 +25,7 @@
 from vllm.logger import init_logger
 from vllm.model_executor.layers.vocab_parallel_embedding import VocabParallelEmbedding
 from vllm.platforms import current_platform
+from vllm.transformers_utils.repo_utils import hf_api
 from vllm.utils.argparse_utils import FlexibleArgumentParser
 from vllm.utils.import_utils import PlaceholderModule
 
@@ -629,7 +629,7 @@ def serialize_extra_artifacts(
         )
 
     with tempfile.TemporaryDirectory() as tmpdir:
-        snapshot_download(
+        hf_api().snapshot_download(
             served_model_name,
             local_dir=tmpdir,
             ignore_patterns=[
diff --git a/vllm/model_executor/model_loader/weight_utils.py b/vllm/model_executor/model_loader/weight_utils.py
index 972271e8c308..cbb191ebb62b 100644
--- a/vllm/model_executor/model_loader/weight_utils.py
+++ b/vllm/model_executor/model_loader/weight_utils.py
@@ -23,7 +23,6 @@
 import numpy as np
 import regex as re
 import torch
-from huggingface_hub import HfFileSystem, hf_hub_download, snapshot_download
 from safetensors.torch import load, load_file, safe_open, save_file
 from tqdm.auto import tqdm
 from transformers.utils import SAFE_WEIGHTS_INDEX_NAME
@@ -46,6 +45,7 @@
 )
 from vllm.platforms import current_platform
 from vllm.tracing import instrument
+from vllm.transformers_utils.repo_utils import hf_api, hf_fs
 from vllm.utils.import_utils import PlaceholderModule
 
 try:
@@ -373,7 +373,7 @@ def get_quant_config(
     if not is_local:
         # Download the config files.
         with get_lock(model_config.model, load_config.download_dir):
-            hf_folder = snapshot_download(
+            hf_folder = hf_api().snapshot_download(
                 model_config.model,
                 revision=model_config.revision,
                 allow_patterns="*.json",
@@ -431,7 +431,7 @@ def get_sparse_attention_config(
     if not is_local:
         # Download the config files.
         with get_lock(model_name_or_path, load_config.download_dir):
-            hf_folder = snapshot_download(
+            hf_folder = hf_api().snapshot_download(
                 model_name_or_path,
                 revision=model_config.revision,
                 allow_patterns="*.json",
@@ -534,7 +534,7 @@ def download_weights_from_hf(
         # Attempt to reduce allow_patterns to a single pattern
         # so we only have to call snapshot_download once.
         try:
-            fs = HfFileSystem()
+            fs = hf_fs()
             file_list = fs.ls(
                 os.path.join(model_name_or_path, subfolder or ""),
                 detail=False,
@@ -546,7 +546,7 @@ def download_weights_from_hf(
             # unnecessary files (e.g., from subdirectories like "original/").
             index_file = f"{model_name_or_path}/{SAFE_WEIGHTS_INDEX_NAME}"
             if "*.safetensors" in allow_patterns and index_file in file_list:
-                index_path = hf_hub_download(
+                index_path = hf_api().hf_hub_download(
                     repo_id=model_name_or_path,
                     filename=SAFE_WEIGHTS_INDEX_NAME,
                     cache_dir=cache_dir,
@@ -582,7 +582,7 @@ def download_weights_from_hf(
     with get_lock(model_name_or_path, cache_dir):
         start_time = time.perf_counter()
         for allow_pattern in allow_patterns:
-            hf_folder = snapshot_download(
+            hf_folder = hf_api().snapshot_download(
                 model_name_or_path,
                 allow_patterns=allow_pattern,
                 ignore_patterns=ignore_patterns,
@@ -631,7 +631,7 @@ def download_safetensors_index_file_from_hf(
     with get_lock(model_name_or_path, cache_dir):
         try:
             # Download the safetensors index file.
-            hf_hub_download(
+            hf_api().hf_hub_download(
                 repo_id=model_name_or_path,
                 filename=index_file,
                 cache_dir=cache_dir,
diff --git a/vllm/plugins/lora_resolvers/hf_hub_resolver.py b/vllm/plugins/lora_resolvers/hf_hub_resolver.py
index c152d4c29467..57ccdc2e48d8 100644
--- a/vllm/plugins/lora_resolvers/hf_hub_resolver.py
+++ b/vllm/plugins/lora_resolvers/hf_hub_resolver.py
@@ -3,13 +3,12 @@
 import asyncio
 import os
 
-from huggingface_hub import HfApi, snapshot_download
-
 import vllm.envs as envs
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
 from vllm.lora.resolver import LoRAResolverRegistry
 from vllm.plugins.lora_resolvers.filesystem_resolver import FilesystemResolver
+from vllm.transformers_utils.repo_utils import hf_api
 
 logger = init_logger(__name__)
 
@@ -49,7 +48,7 @@ async def resolve_lora(
             return None
 
         repo_path = await asyncio.to_thread(
-            snapshot_download,
+            hf_api().snapshot_download,
             repo_id=maybe_repo,
             allow_patterns=f"{maybe_subpath}/*" if maybe_subpath != "." else "*",
         )
@@ -110,7 +109,10 @@ async def _get_adapter_dirs(self, repo_name: str) -> set[str]:
         Args:
             repo_name: Name of the HF hub repo to inspect.
         """
-        repo_files = await asyncio.to_thread(HfApi().list_repo_files, repo_id=repo_name)
+        repo_files = await asyncio.to_thread(
+            hf_api().list_repo_files,
+            repo_id=repo_name,
+        )
         adapter_dirs = {
             os.path.dirname(name)
             for name in repo_files
diff --git a/vllm/tokenizers/grok2.py b/vllm/tokenizers/grok2.py
index 61fa1107e2a3..612af5374082 100644
--- a/vllm/tokenizers/grok2.py
+++ b/vllm/tokenizers/grok2.py
@@ -8,7 +8,6 @@
 from pathlib import Path
 from typing import Any, Literal, overload
 
-from huggingface_hub import hf_hub_download
 from huggingface_hub.utils import (
     EntryNotFoundError,
     HfHubHTTPError,
@@ -20,6 +19,7 @@
 
 from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
 from vllm.logger import init_logger
+from vllm.transformers_utils.repo_utils import hf_api
 
 from .protocol import TokenizerLike
 
@@ -70,7 +70,7 @@ def _maybe_load_tokenizer_config(
         return {}
 
     try:
-        config_file = hf_hub_download(
+        config_file = hf_api().hf_hub_download(
             repo_id=repo_id,
             filename="tokenizer_config.json",
             revision=revision,
@@ -208,7 +208,7 @@ def from_pretrained(
             repo_id = None
         else:
             vocab_file = Path(
-                hf_hub_download(
+                hf_api().hf_hub_download(
                     repo_id=str(path_or_repo_id),
                     filename="tokenizer.tok.json",
                     revision=revision,
diff --git a/vllm/tokenizers/kimi_audio.py b/vllm/tokenizers/kimi_audio.py
index d2b0a2a557ef..10f57a7f7ded 100644
--- a/vllm/tokenizers/kimi_audio.py
+++ b/vllm/tokenizers/kimi_audio.py
@@ -10,13 +10,13 @@
 
 import pybase64
 import tiktoken
-from huggingface_hub import hf_hub_download
 from transformers import AddedToken, BatchEncoding
 from transformers.utils import chat_template_utils as hf_chat_utils
 
 from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
 from vllm.logger import init_logger
 from vllm.tokenizers.protocol import TokenizerLike
+from vllm.transformers_utils.repo_utils import hf_api
 
 logger = init_logger(__name__)
 
@@ -78,7 +78,7 @@ def from_pretrained(
 
             # Try to download tiktoken.model or tokenizer.model
             try:
-                vocab_path = hf_hub_download(
+                vocab_path = hf_api().hf_hub_download(
                     repo_id=repo_id,
                     filename="tiktoken.model",
                     revision=revision,
@@ -87,7 +87,7 @@ def from_pretrained(
                 vocab_file = Path(vocab_path)
             except Exception:
                 try:
-                    vocab_path = hf_hub_download(
+                    vocab_path = hf_api().hf_hub_download(
                         repo_id=repo_id,
                         filename="tokenizer.model",
                         revision=revision,
@@ -101,7 +101,7 @@ def from_pretrained(
 
             # Also download tokenizer_config.json if available
             with contextlib.suppress(Exception):
-                hf_hub_download(
+                hf_api().hf_hub_download(
                     repo_id=repo_id,
                     filename="tokenizer_config.json",
                     revision=revision,
diff --git a/vllm/transformers_utils/config.py b/vllm/transformers_utils/config.py
index ffe5941bcc08..f940739a96c3 100644
--- a/vllm/transformers_utils/config.py
+++ b/vllm/transformers_utils/config.py
@@ -12,7 +12,7 @@
 
 import huggingface_hub
 import torch
-from huggingface_hub import constants, get_safetensors_metadata
+from huggingface_hub import constants
 from packaging.version import Version
 from safetensors.torch import _TYPES as _SAFETENSORS_TO_TORCH_DTYPE
 from transformers import GenerationConfig, PretrainedConfig
@@ -43,6 +43,7 @@
 from .repo_utils import (
     file_or_path_exists,
     get_hf_file_to_dict,
+    hf_api,
     list_repo_files,
     try_get_local_file,
     with_retry,
@@ -1143,7 +1144,7 @@ def try_get_safetensors_metadata(
     revision: str | None = None,
 ):
     get_safetensors_metadata_partial = partial(
-        get_safetensors_metadata, model, revision=revision
+        hf_api().get_safetensors_metadata, model, revision=revision
     )
 
     try:
@@ -1237,7 +1238,7 @@ def get_safetensors_params_metadata(
     # local HF cache: weights may already be cached from a prior run, and
     # weight loading itself uses the same cache.
     try:
-        local_dir = huggingface_hub.snapshot_download(
+        local_dir = hf_api().snapshot_download(
             repo_id=model,
             revision=revision,
             allow_patterns=["*.safetensors"],
diff --git a/vllm/transformers_utils/repo_utils.py b/vllm/transformers_utils/repo_utils.py
index 704d505617f2..8385057e9111 100644
--- a/vllm/transformers_utils/repo_utils.py
+++ b/vllm/transformers_utils/repo_utils.py
@@ -12,8 +12,7 @@
 from typing import TypeVar
 
 import huggingface_hub
-from huggingface_hub import hf_hub_download, try_to_load_from_cache
-from huggingface_hub import list_repo_files as hf_list_repo_files
+from huggingface_hub import HfApi, try_to_load_from_cache
 from huggingface_hub.utils import (
     EntryNotFoundError,
     HfHubHTTPError,
@@ -24,9 +23,31 @@
 
 from vllm import envs
 from vllm.logger import init_logger
+from vllm.version import __version__ as VLLM_VERSION
 
 logger = init_logger(__name__)
 
+_hf_api: HfApi | None = None
+
+
+def hf_api() -> HfApi:
+    """Return a shared HfApi instance tagged with vLLM's library info."""
+    global _hf_api
+    if _hf_api is None:
+        _hf_api = HfApi(
+            library_name="vllm",
+            library_version=VLLM_VERSION,
+        )
+    return _hf_api
+
+
+def hf_fs() -> "huggingface_hub.HfFileSystem":
+    """Return a fresh HfFileSystem tagged with vLLM's library info."""
+    return huggingface_hub.HfFileSystem(
+        library_name="vllm",
+        library_version=VLLM_VERSION,
+    )
+
 
 _R = TypeVar("_R")
 
@@ -80,7 +101,7 @@ def lookup_files() -> list[str]:
                     revision=revision,
                     token=os.getenv("MODELSCOPE_API_TOKEN", None),
                 )
-            return hf_list_repo_files(
+            return hf_api().list_repo_files(
                 repo_id, revision=revision, repo_type=repo_type, token=token
             )
         except huggingface_hub.errors.OfflineModeIsEnabled:
@@ -215,9 +236,10 @@ def get_model_path(model: str | Path, revision: str | None = None):
 
         return snapshot_download(model_id=model, **common_kwargs)
 
-    from huggingface_hub import snapshot_download
-
-    return snapshot_download(repo_id=model, **common_kwargs)
+    return hf_api().snapshot_download(
+        repo_id=model,
+        **common_kwargs,
+    )
 
 
 def _try_download_from_hf_hub(
@@ -231,7 +253,13 @@ def _try_download_from_hf_hub(
     if Path(model).is_dir():
         return None
     try:
-        return Path(hf_hub_download(model, file_name, revision=revision))
+        return Path(
+            hf_api().hf_hub_download(
+                model,
+                file_name,
+                revision=revision,
+            )
+        )
     except huggingface_hub.errors.OfflineModeIsEnabled:
         return None
     except (