A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 03:30 UTC
[Misc] add handler HF_TOKEN is emptry string (#20369)
Signed-off-by: rongfu.leng <rongfu.leng@daocloud.io>
rongfu.leng committed
Jul 3, 2025 at 00:14 UTC
139508a418a4185d50eef087d054b8115cb77542
1 file changed
+22
-6
vllm/transformers_utils/config.py
+22
-6
@@ -56,6 +56,22 @@ MISTRAL_CONFIG_NAME = "params.json"
56
57
logger = init_logger(__name__)
58
59
+
60
+def _get_hf_token() -> Optional[str]:
61
+ """
62
+ Get the HuggingFace token from environment variable.
63
+
64
+ Returns None if the token is not set, is an empty string,
65
+ or contains only whitespace.
66
+ This follows the same pattern as huggingface_hub library which
67
+ treats empty string tokens as None to avoid authentication errors.
68
+ """
69
+ token = os.getenv('HF_TOKEN')
70
+ if token and token.strip():
71
+ return token
72
+ return None
73
+
74
+
75
_CONFIG_REGISTRY_OVERRIDE_HF: dict[str, type[PretrainedConfig]] = {
76
"mllama": MllamaConfig
77
}
@@ -195,7 +211,7 @@ def file_or_path_exists(model: Union[str, Path], config_name: str,
211
return file_exists(str(model),
212
config_name,
213
revision=revision,
198
- token=os.getenv('HF_TOKEN', None))
214
+ token=_get_hf_token())
215
216
217
def patch_rope_scaling(config: PretrainedConfig) -> None:
@@ -322,7 +338,7 @@ def get_config(
338
model,
339
revision=revision,
340
code_revision=code_revision,
325
- token=os.getenv('HF_TOKEN', None),
341
+ token=_get_hf_token(),
342
**kwargs,
343
)
344
@@ -334,7 +350,7 @@ def get_config(
350
model,
351
revision=revision,
352
code_revision=code_revision,
337
- token=os.getenv('HF_TOKEN', None),
353
+ token=_get_hf_token(),
354
**kwargs,
355
)
356
else:
@@ -344,7 +360,7 @@ def get_config(
360
trust_remote_code=trust_remote_code,
361
revision=revision,
362
code_revision=code_revision,
347
- token=os.getenv('HF_TOKEN', None),
363
+ token=_get_hf_token(),
364
**kwargs,
365
)
366
except ValueError as e:
@@ -571,7 +587,7 @@ def get_sentence_transformer_tokenizer_config(model: str,
587
# If model is on HuggingfaceHub, get the repo files
588
repo_files = list_repo_files(model,
589
revision=revision,
574
- token=os.getenv('HF_TOKEN', None))
590
+ token=_get_hf_token())
591
except Exception:
592
repo_files = []
593
@@ -862,7 +878,7 @@ def try_get_safetensors_metadata(
878
get_safetensors_metadata,
879
model,
880
revision=revision,
865
- token=os.getenv('HF_TOKEN', None),
881
+ token=_get_hf_token(),
882
)
883
884
try: