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, 01:00 UTC
[Misc] Avoid unnecessary import (#21106)
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
wangxiyuan committed
Jul 17, 2025 at 20:57 UTC
89e3c4e9b46745741ab6b364358449056e4f64dd
2 files changed
+13
-9
vllm/entrypoints/openai/speech_to_text.py
+1
-1
@@ -24,7 +24,6 @@ from vllm.entrypoints.openai.serving_engine import (OpenAIServing,
24
from vllm.entrypoints.openai.serving_models import OpenAIServingModels
25
from vllm.inputs.data import PromptType
26
from vllm.logger import init_logger
27
-from vllm.model_executor.model_loader import get_model_cls
27
from vllm.model_executor.models import SupportsTranscription
28
from vllm.outputs import RequestOutput
29
from vllm.utils import PlaceholderModule
@@ -78,6 +77,7 @@ class OpenAISpeechToText(OpenAIServing):
77
78
@cached_property
79
def model_cls(self) -> type[SupportsTranscription]:
80
+ from vllm.model_executor.model_loader import get_model_cls
81
model_cls = get_model_cls(self.model_config)
82
return cast(type[SupportsTranscription], model_cls)
83
vllm/lora/utils.py
+12
-8
@@ -2,7 +2,7 @@
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
4
import os
5
-from typing import Optional, Union
5
+from typing import TYPE_CHECKING, Optional, Union
6
7
import huggingface_hub
8
import regex as re
@@ -31,10 +31,14 @@ from vllm.lora.layers import (BaseLayerWithLoRA, ColumnParallelLinearWithLoRA,
31
RowParallelLinearWithLoRA,
32
VocabParallelEmbeddingWithLoRA)
33
from vllm.model_executor.layers.linear import LinearBase
34
+
35
# yapf: enable
35
-from vllm.model_executor.layers.logits_processor import LogitsProcessor
36
-from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead
37
-from vllm.model_executor.models.utils import WeightsMapper
36
+
37
+if TYPE_CHECKING:
38
+ from vllm.model_executor.layers.logits_processor import LogitsProcessor
39
+ from vllm.model_executor.layers.vocab_parallel_embedding import (
40
+ ParallelLMHead)
41
+ from vllm.model_executor.models.utils import WeightsMapper
42
43
logger = init_logger(__name__)
44
@@ -75,8 +79,8 @@ def from_layer(layer: nn.Module,
79
80
81
def from_layer_logits_processor(
78
- layer: LogitsProcessor,
79
- lm_head: ParallelLMHead,
82
+ layer: "LogitsProcessor",
83
+ lm_head: "ParallelLMHead",
84
max_loras: int,
85
lora_config: LoRAConfig,
86
model_config: Optional[PretrainedConfig] = None,
@@ -98,8 +102,8 @@ def replace_submodule(model: nn.Module, module_name: str,
102
103
104
def parse_fine_tuned_lora_name(
101
- name: str,
102
- weights_mapper: Optional[WeightsMapper] = None
105
+ name: str,
106
+ weights_mapper: Optional["WeightsMapper"] = None
107
) -> tuple[str, bool, bool]:
108
"""Parse the name of lora weights.
109