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 11, 22:30 UTC
[Misc] Fix circular import in vllm.transformers_utils.config (#30179)
Signed-off-by: Ye (Charlotte) Qi <yeq@meta.com>
Ye (Charlotte) Qi committed
Dec 6, 2025 at 01:24 UTC
92c35abb242babbf592390960fb5a4155261e017
1 file changed
+2
-1
vllm/transformers_utils/config.py
+2
-1
@@ -25,7 +25,6 @@ from transformers.models.auto.tokenization_auto import get_tokenizer_config
25
from transformers.utils import CONFIG_NAME as HF_CONFIG_NAME
26
27
from vllm import envs
28
-from vllm.config.utils import getattr_iter
28
from vllm.logger import init_logger
29
from vllm.transformers_utils.utils import parse_safetensors_file_metadata
30
@@ -305,6 +304,8 @@ def set_default_rope_theta(config: PretrainedConfig, default_theta: float) -> No
304
305
def patch_rope_parameters(config: PretrainedConfig) -> None:
306
"""Provide backwards compatibility for RoPE."""
307
+ from vllm.config.utils import getattr_iter
308
+
309
rope_theta_names = ("rope_theta", "rotary_emb_base")
310
rope_theta = getattr_iter(config, rope_theta_names, None)
311
if Version(version("transformers")) < Version("5.0.0.dev0"):