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 12, 16:00 UTC
[Bugfix] Use lazy string reference for DeepseekV3Config in config registry (#28958)
Signed-off-by: Luke <yq0536@gmail.com> Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn> Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Luke committed
Nov 23, 2025 at 06:43 UTC
6fb0215eee44cf5e4b28f57e6739ef4a51945127
2 files changed
+8
-2
vllm/transformers_utils/config.py
+2
-2
@@ -27,7 +27,7 @@ from huggingface_hub.utils import (
27
RevisionNotFoundError,
28
)
29
from packaging.version import Version
30
-from transformers import DeepseekV3Config, GenerationConfig, PretrainedConfig
30
+from transformers import GenerationConfig, PretrainedConfig
31
from transformers.configuration_utils import ALLOWED_LAYER_TYPES
32
from transformers.models.auto.image_processing_auto import get_image_processor_config
33
from transformers.models.auto.modeling_auto import (
@@ -84,7 +84,7 @@ _CONFIG_REGISTRY: dict[str, type[PretrainedConfig]] = LazyConfigDict(
84
afmoe="AfmoeConfig",
85
chatglm="ChatGLMConfig",
86
deepseek_vl_v2="DeepseekVLV2Config",
87
- deepseek_v32=DeepseekV3Config,
87
+ deepseek_v32="DeepseekV3Config",
88
flex_olmo="FlexOlmoConfig",
89
kimi_linear="KimiLinearConfig",
90
kimi_vl="KimiVLConfig",
vllm/transformers_utils/configs/__init__.py
+6
@@ -5,8 +5,13 @@ Model configs may be defined in this directory for the following reasons:
5
6
- There is no configuration file defined by HF Hub or Transformers library.
7
- There is a need to override the existing config to support vLLM.
8
+- The HF model_type isn't recognized by the Transformers library but can
9
+ be mapped to an existing Transformers config, such as
10
+ deepseek-ai/DeepSeek-V3.2-Exp.
11
"""
12
13
+from transformers import DeepseekV3Config
14
+
15
from vllm.transformers_utils.configs.afmoe import AfmoeConfig
16
from vllm.transformers_utils.configs.chatglm import ChatGLMConfig
17
from vllm.transformers_utils.configs.deepseek_vl2 import DeepseekVLV2Config
@@ -44,6 +49,7 @@ __all__ = [
49
"AfmoeConfig",
50
"ChatGLMConfig",
51
"DeepseekVLV2Config",
52
+ "DeepseekV3Config",
53
"DotsOCRConfig",
54
"EAGLEConfig",
55
"FlexOlmoConfig",