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 14, 05:30 UTC
Remove unused imports (#26972)
Signed-off-by: Lukas Geiger <lukas.geiger94@gmail.com>
Lukas Geiger committed
Oct 17, 2025 at 03:51 UTC
4d055ef46521bdd9d9ca79de0bcfc943c4909afa
8 files changed
-13
vllm/model_executor/layers/mamba/linear_attn.py
-4
@@ -10,7 +10,6 @@ if TYPE_CHECKING:
10
from typing import TYPE_CHECKING
11
12
import torch
13
-import torch.distributed
13
import torch.nn.functional as F
14
from einops import rearrange
15
from torch import nn
@@ -41,9 +40,6 @@ from vllm.v1.attention.backends.linear_attn import LinearAttentionMetadata
40
if TYPE_CHECKING:
41
from vllm.attention.backends.abstract import AttentionBackend
42
44
-import torch
45
-import torch.distributed
46
-
43
44
class MiniMaxText01RMSNormTP(CustomOp):
45
name = "MiniMaxText01RMSNormTP"
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py
-1
@@ -848,7 +848,6 @@ class CompressedTensorsW8A8Fp8MoEMethod(CompressedTensorsMoEMethod):
848
# Property to determine if AITER is used
849
if self.rocm_aiter_moe_enabled:
850
from vllm.model_executor.layers.fused_moe.rocm_aiter_fused_moe import ( # noqa E501
851
- rocm_aiter_fused_experts,
851
shuffle_weights,
852
)
853
vllm/model_executor/models/gemma3n_mm.py
-1
@@ -58,7 +58,6 @@ from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsTransc
58
from .utils import (
59
AutoWeightsLoader,
60
WeightsMapper,
61
- flatten_bn,
61
init_vllm_registered_model,
62
maybe_prefix,
63
)
vllm/model_executor/models/minimax_text_01.py
-1
@@ -11,7 +11,6 @@ if TYPE_CHECKING:
11
12
import regex as re
13
import torch
14
-import torch.distributed
14
from torch import nn
15
from transformers import MiniMaxConfig
16
vllm/multimodal/utils.py
-2
@@ -31,13 +31,11 @@ if TYPE_CHECKING:
31
from .inputs import (
32
BatchedTensorInputs,
33
MultiModalKwargsItem,
34
- MultiModalKwargsItems,
34
MultiModalPlaceholderDict,
35
)
36
else:
37
BatchedTensorInputs = Any
38
MultiModalKwargsItem = Any
40
- MultiModalKwargsItems = Any
39
MultiModalPlaceholderDict = Any
40
41
global_thread_pool = ThreadPoolExecutor(
vllm/transformers_utils/tokenizer.py
-2
@@ -21,11 +21,9 @@ from vllm.transformers_utils.utils import check_gguf_file
21
22
if TYPE_CHECKING:
23
from vllm.config import ModelConfig
24
- from vllm.lora.request import LoRARequest
24
from vllm.transformers_utils.tokenizer_base import TokenizerBase
25
else:
26
ModelConfig = Any
28
- LoRARequest = Any
27
TokenizerBase = Any
28
29
logger = init_logger(__name__)
vllm/utils/__init__.py
-1
@@ -900,7 +900,6 @@ def _cuda_device_count_stateless(cuda_visible_devices: str | None = None) -> int
900
# c1cd946818442aca8c7f812b16d187ce1586c3bc/
901
# torch/cuda/__init__.py#L831C1-L831C17
902
import torch.cuda
903
- import torch.version
903
904
from vllm.platforms import current_platform
905
vllm/v1/worker/tpu_worker.py
-1
@@ -7,7 +7,6 @@ from collections.abc import Callable
7
from typing import Any, TypeVar
8
9
import torch
10
-import torch.distributed
10
import torch.nn as nn
11
12
import vllm.envs as envs