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
Tidy `vllm/config/__init__.py` to only add classes and functions (#26405)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Oct 8, 2025 at 15:10 UTC
2f99f2f506f919922ee699350204a176f8f87c13
18 files changed
+46
-77
tests/distributed/test_context_parallel.py
+1
-1
@@ -15,7 +15,7 @@ from typing import Literal, NamedTuple, Optional
15
16
import pytest
17
18
-from vllm.config import RunnerOption
18
+from vllm.config.model import RunnerOption
19
from vllm.logger import init_logger
20
21
from ..models.registry import HF_EXAMPLE_MODELS
tests/distributed/test_expert_parallel.py
+1
-1
@@ -6,7 +6,7 @@ from typing import Literal, NamedTuple, Optional
6
7
import pytest
8
9
-from vllm.config import RunnerOption
9
+from vllm.config.model import RunnerOption
10
from vllm.logger import init_logger
11
12
from ..utils import compare_two_settings, create_new_process_for_each_test
tests/distributed/test_sequence_parallel.py
+1
-1
@@ -15,7 +15,7 @@ from typing import Literal, NamedTuple, Optional
15
16
import pytest
17
18
-from vllm.config import RunnerOption
18
+from vllm.config.model import RunnerOption
19
from vllm.logger import init_logger
20
21
from ..models.registry import HF_EXAMPLE_MODELS
tests/models/multimodal/generation/vlm_utils/core.py
+1
-1
@@ -7,7 +7,7 @@ from typing import Any, Callable, Optional
7
import torch
8
from transformers.models.auto.auto_factory import _BaseAutoModelClass
9
10
-from vllm.config import RunnerOption
10
+from vllm.config.model import RunnerOption
11
from vllm.transformers_utils.tokenizer import AnyTokenizer
12
13
from .....conftest import HfRunner, VllmRunner
tests/models/multimodal/generation/vlm_utils/types.py
+1
-1
@@ -12,7 +12,7 @@ from pytest import MarkDecorator
12
from transformers import AutoModelForCausalLM
13
from transformers.models.auto.auto_factory import _BaseAutoModelClass
14
15
-from vllm.config import RunnerOption
15
+from vllm.config.model import RunnerOption
16
from vllm.logprobs import SampleLogprobs
17
from vllm.transformers_utils.tokenizer import AnyTokenizer
18
tests/models/registry.py
+1
-1
@@ -10,7 +10,7 @@ import torch
10
from packaging.version import Version
11
from transformers import __version__ as TRANSFORMERS_VERSION
12
13
-from vllm.config import ModelDType, TokenizerMode
13
+from vllm.config.model import ModelDType, TokenizerMode
14
15
16
@dataclass(frozen=True)
tests/models/utils.py
+1
-1
@@ -10,7 +10,7 @@ import torch
10
import torch.nn.functional as F
11
from transformers import PretrainedConfig
12
13
-from vllm.config import ModelConfig, ModelDType, RunnerOption
13
+from vllm.config.model import ModelConfig, ModelDType, RunnerOption
14
from vllm.logprobs import Logprob, PromptLogprobs, SampleLogprobs
15
from vllm.multimodal.processing import InputProcessingContext
16
from vllm.transformers_utils.tokenizer import cached_tokenizer_from_config
tests/v1/attention/utils.py
+1
-1
@@ -15,11 +15,11 @@ from vllm.config import (
15
DeviceConfig,
16
LoadConfig,
17
ModelConfig,
18
- ModelDType,
18
ParallelConfig,
19
SchedulerConfig,
20
VllmConfig,
21
)
22
+from vllm.config.model import ModelDType
23
from vllm.platforms import current_platform
24
from vllm.utils import resolve_obj_by_qualname
25
from vllm.v1.attention.backends.utils import CommonAttentionMetadata
tests/v1/entrypoints/llm/test_struct_output_generate.py
+1
-1
@@ -29,7 +29,7 @@ from vllm.sampling_params import (
29
)
30
31
if TYPE_CHECKING:
32
- from vllm.config import TokenizerMode
32
+ from vllm.config.model import TokenizerMode
33
34
NGRAM_SPEC_CONFIG = {
35
"model": "[ngram]",
tests/v1/metrics/test_ray_metrics.py
+1
-1
@@ -4,7 +4,7 @@
4
import pytest
5
import ray
6
7
-from vllm.config import ModelDType
7
+from vllm.config.model import ModelDType
8
from vllm.sampling_params import SamplingParams
9
from vllm.v1.engine.async_llm import AsyncEngineArgs, AsyncLLM
10
from vllm.v1.metrics.ray_wrappers import RayPrometheusMetric, RayPrometheusStatLogger
tests/v1/sample/test_logprobs.py
+1
-1
@@ -16,7 +16,7 @@ from tests.v1.sample.utils import (
16
get_test_batch,
17
)
18
from vllm import SamplingParams
19
-from vllm.config import LogprobsMode
19
+from vllm.config.model import LogprobsMode
20
21
from ...conftest import HfRunner, VllmRunner
22
vllm/config/__init__.py
+8
-39
@@ -1,42 +1,28 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
4
-from vllm.config.cache import (
5
- BlockSize,
6
- CacheConfig,
7
- CacheDType,
8
- MambaDType,
9
- PrefixCachingHashAlgo,
10
-)
4
+from vllm.config.cache import CacheConfig
5
from vllm.config.compilation import (
6
CompilationConfig,
7
CompilationLevel,
8
CUDAGraphMode,
9
PassConfig,
10
)
17
-from vllm.config.device import Device, DeviceConfig
11
+from vllm.config.device import DeviceConfig
12
from vllm.config.kv_events import KVEventsConfig
13
from vllm.config.kv_transfer import KVTransferConfig
14
from vllm.config.load import LoadConfig
15
from vllm.config.lora import LoRAConfig
16
from vllm.config.model import (
23
- ConvertOption,
24
- HfOverrides,
25
- LogprobsMode,
17
ModelConfig,
27
- ModelDType,
28
- ModelImpl,
29
- RunnerOption,
30
- TaskOption,
31
- TokenizerMode,
18
iter_architecture_defaults,
19
try_match_architecture_defaults,
20
)
35
-from vllm.config.multimodal import MMCacheType, MMEncoderTPMode, MultiModalConfig
36
-from vllm.config.observability import DetailedTraceModules, ObservabilityConfig
37
-from vllm.config.parallel import DistributedExecutorBackend, EPLBConfig, ParallelConfig
21
+from vllm.config.multimodal import MultiModalConfig
22
+from vllm.config.observability import ObservabilityConfig
23
+from vllm.config.parallel import EPLBConfig, ParallelConfig
24
from vllm.config.pooler import PoolerConfig
39
-from vllm.config.scheduler import RunnerType, SchedulerConfig, SchedulerPolicy
25
+from vllm.config.scheduler import SchedulerConfig
26
from vllm.config.speculative import SpeculativeConfig
27
from vllm.config.speech_to_text import SpeechToTextConfig
28
from vllm.config.structured_outputs import StructuredOutputsConfig
@@ -56,20 +42,17 @@ from vllm.config.vllm import (
42
set_current_vllm_config,
43
)
44
45
+# __all__ should only contain classes and functions.
46
+# Types and globals should be imported from their respective modules.
47
__all__ = [
48
# From vllm.config.cache
61
- "BlockSize",
49
"CacheConfig",
63
- "CacheDType",
64
- "MambaDType",
65
- "PrefixCachingHashAlgo",
50
# From vllm.config.compilation
51
"CompilationConfig",
52
"CompilationLevel",
53
"CUDAGraphMode",
54
"PassConfig",
55
# From vllm.config.device
72
- "Device",
56
"DeviceConfig",
57
# From vllm.config.kv_events
58
"KVEventsConfig",
@@ -80,34 +63,20 @@ __all__ = [
63
# From vllm.config.lora
64
"LoRAConfig",
65
# From vllm.config.model
83
- "ConvertOption",
84
- "HfOverrides",
85
- "LogprobsMode",
66
"ModelConfig",
87
- "ModelDType",
88
- "ModelImpl",
89
- "RunnerOption",
90
- "TaskOption",
91
- "TokenizerMode",
67
"iter_architecture_defaults",
68
"try_match_architecture_defaults",
69
# From vllm.config.multimodal
95
- "MMCacheType",
96
- "MMEncoderTPMode",
70
"MultiModalConfig",
71
# From vllm.config.observability
99
- "DetailedTraceModules",
72
"ObservabilityConfig",
73
# From vllm.config.parallel
102
- "DistributedExecutorBackend",
74
"EPLBConfig",
75
"ParallelConfig",
76
# From vllm.config.pooler
77
"PoolerConfig",
78
# From vllm.config.scheduler
108
- "RunnerType",
79
"SchedulerConfig",
110
- "SchedulerPolicy",
80
# From vllm.config.speculative
81
"SpeculativeConfig",
82
# From vllm.config.speech_to_text
vllm/engine/arg_utils.py
+16
-18
@@ -31,43 +31,41 @@ from typing_extensions import TypeIs, deprecated
31
32
import vllm.envs as envs
33
from vllm.config import (
34
- BlockSize,
34
CacheConfig,
36
- CacheDType,
35
CompilationConfig,
36
ConfigType,
39
- ConvertOption,
40
- DetailedTraceModules,
41
- Device,
37
DeviceConfig,
43
- DistributedExecutorBackend,
38
EPLBConfig,
45
- HfOverrides,
39
KVEventsConfig,
40
KVTransferConfig,
41
LoadConfig,
49
- LogprobsMode,
42
LoRAConfig,
51
- MambaDType,
52
- MMEncoderTPMode,
43
ModelConfig,
54
- ModelDType,
44
+ MultiModalConfig,
45
ObservabilityConfig,
46
ParallelConfig,
47
PoolerConfig,
58
- PrefixCachingHashAlgo,
59
- RunnerOption,
48
SchedulerConfig,
61
- SchedulerPolicy,
49
SpeculativeConfig,
50
StructuredOutputsConfig,
64
- TaskOption,
65
- TokenizerMode,
51
VllmConfig,
52
get_attr_docs,
53
)
69
-from vllm.config.multimodal import MMCacheType, MultiModalConfig
70
-from vllm.config.parallel import ExpertPlacementStrategy
54
+from vllm.config.cache import BlockSize, CacheDType, MambaDType, PrefixCachingHashAlgo
55
+from vllm.config.device import Device
56
+from vllm.config.model import (
57
+ ConvertOption,
58
+ HfOverrides,
59
+ LogprobsMode,
60
+ ModelDType,
61
+ RunnerOption,
62
+ TaskOption,
63
+ TokenizerMode,
64
+)
65
+from vllm.config.multimodal import MMCacheType, MMEncoderTPMode
66
+from vllm.config.observability import DetailedTraceModules
67
+from vllm.config.parallel import DistributedExecutorBackend, ExpertPlacementStrategy
68
+from vllm.config.scheduler import SchedulerPolicy
69
from vllm.config.utils import get_field
70
from vllm.logger import init_logger
71
from vllm.platforms import CpuArchEnum, current_platform
vllm/entrypoints/llm.py
+5
-5
@@ -19,18 +19,18 @@ from vllm.beam_search import (
19
)
20
from vllm.config import (
21
CompilationConfig,
22
- ModelDType,
22
+ PoolerConfig,
23
StructuredOutputsConfig,
24
- TokenizerMode,
24
is_init_field,
25
)
27
-from vllm.engine.arg_utils import (
26
+from vllm.config.model import (
27
ConvertOption,
29
- EngineArgs,
28
HfOverrides,
31
- PoolerConfig,
29
+ ModelDType,
30
RunnerOption,
31
+ TokenizerMode,
32
)
33
+from vllm.engine.arg_utils import EngineArgs
34
from vllm.entrypoints.chat_utils import (
35
ChatCompletionMessageParam,
36
ChatTemplateContentFormatOption,
vllm/model_executor/layers/mamba/mamba_utils.py
+2
-1
@@ -4,7 +4,8 @@ from typing import Union
4
5
import torch
6
7
-from vllm.config import MambaDType, ModelDType
7
+from vllm.config.cache import MambaDType
8
+from vllm.config.model import ModelDType
9
from vllm.distributed import divide
10
from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, get_kv_cache_torch_dtype
11
vllm/platforms/tpu.py
+2
-1
@@ -15,7 +15,8 @@ from .interface import Platform, PlatformEnum
15
16
if TYPE_CHECKING:
17
from vllm.attention.backends.registry import _Backend
18
- from vllm.config import BlockSize, ModelConfig, VllmConfig
18
+ from vllm.config import ModelConfig, VllmConfig
19
+ from vllm.config.cache import BlockSize
20
from vllm.pooling_params import PoolingParams
21
else:
22
BlockSize = None
vllm/v1/sample/ops/topk_topp_sampler.py
+1
-1
@@ -8,7 +8,7 @@ import torch.nn as nn
8
from packaging import version
9
10
from vllm import envs
11
-from vllm.config import LogprobsMode
11
+from vllm.config.model import LogprobsMode
12
from vllm.logger import init_logger
13
from vllm.platforms import current_platform
14
vllm/v1/sample/sampler.py
+1
-1
@@ -7,7 +7,7 @@ from typing import Optional
7
import torch
8
import torch.nn as nn
9
10
-from vllm.config import LogprobsMode
10
+from vllm.config.model import LogprobsMode
11
from vllm.utils import is_pin_memory_available
12
from vllm.v1.outputs import LogprobsTensors, SamplerOutput
13
from vllm.v1.sample.metadata import SamplingMetadata