Tidy `vllm/config/__init__.py` to only add classes and functions (#26405)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>

tests/distributed/test_context_parallel.py

@@ -15,7 +15,7 @@ from typing import Literal, NamedTuple, Optional
 
 import pytest
 
-from vllm.config import RunnerOption
+from vllm.config.model import RunnerOption
 from vllm.logger import init_logger
 
 from ..models.registry import HF_EXAMPLE_MODELS

tests/distributed/test_expert_parallel.py

@@ -6,7 +6,7 @@ from typing import Literal, NamedTuple, Optional
 
 import pytest
 
-from vllm.config import RunnerOption
+from vllm.config.model import RunnerOption
 from vllm.logger import init_logger
 
 from ..utils import compare_two_settings, create_new_process_for_each_test

tests/distributed/test_sequence_parallel.py

@@ -15,7 +15,7 @@ from typing import Literal, NamedTuple, Optional
 
 import pytest
 
-from vllm.config import RunnerOption
+from vllm.config.model import RunnerOption
 from vllm.logger import init_logger
 
 from ..models.registry import HF_EXAMPLE_MODELS

tests/models/multimodal/generation/vlm_utils/core.py

@@ -7,7 +7,7 @@ from typing import Any, Callable, Optional
 import torch
 from transformers.models.auto.auto_factory import _BaseAutoModelClass
 
-from vllm.config import RunnerOption
+from vllm.config.model import RunnerOption
 from vllm.transformers_utils.tokenizer import AnyTokenizer
 
 from .....conftest import HfRunner, VllmRunner

tests/models/multimodal/generation/vlm_utils/types.py

@@ -12,7 +12,7 @@ from pytest import MarkDecorator
 from transformers import AutoModelForCausalLM
 from transformers.models.auto.auto_factory import _BaseAutoModelClass
 
-from vllm.config import RunnerOption
+from vllm.config.model import RunnerOption
 from vllm.logprobs import SampleLogprobs
 from vllm.transformers_utils.tokenizer import AnyTokenizer
 

tests/models/registry.py

@@ -10,7 +10,7 @@ import torch
 from packaging.version import Version
 from transformers import __version__ as TRANSFORMERS_VERSION
 
-from vllm.config import ModelDType, TokenizerMode
+from vllm.config.model import ModelDType, TokenizerMode
 
 
 @dataclass(frozen=True)

tests/models/utils.py

@@ -10,7 +10,7 @@ import torch
 import torch.nn.functional as F
 from transformers import PretrainedConfig
 
-from vllm.config import ModelConfig, ModelDType, RunnerOption
+from vllm.config.model import ModelConfig, ModelDType, RunnerOption
 from vllm.logprobs import Logprob, PromptLogprobs, SampleLogprobs
 from vllm.multimodal.processing import InputProcessingContext
 from vllm.transformers_utils.tokenizer import cached_tokenizer_from_config

tests/v1/attention/utils.py

@@ -15,11 +15,11 @@ from vllm.config import (
     DeviceConfig,
     LoadConfig,
     ModelConfig,
-    ModelDType,
     ParallelConfig,
     SchedulerConfig,
     VllmConfig,
 )
+from vllm.config.model import ModelDType
 from vllm.platforms import current_platform
 from vllm.utils import resolve_obj_by_qualname
 from vllm.v1.attention.backends.utils import CommonAttentionMetadata

tests/v1/entrypoints/llm/test_struct_output_generate.py

@@ -29,7 +29,7 @@ from vllm.sampling_params import (
 )
 
 if TYPE_CHECKING:
-    from vllm.config import TokenizerMode
+    from vllm.config.model import TokenizerMode
 
 NGRAM_SPEC_CONFIG = {
     "model": "[ngram]",

tests/v1/metrics/test_ray_metrics.py

@@ -4,7 +4,7 @@
 import pytest
 import ray
 
-from vllm.config import ModelDType
+from vllm.config.model import ModelDType
 from vllm.sampling_params import SamplingParams
 from vllm.v1.engine.async_llm import AsyncEngineArgs, AsyncLLM
 from vllm.v1.metrics.ray_wrappers import RayPrometheusMetric, RayPrometheusStatLogger

tests/v1/sample/test_logprobs.py

@@ -16,7 +16,7 @@ from tests.v1.sample.utils import (
     get_test_batch,
 )
 from vllm import SamplingParams
-from vllm.config import LogprobsMode
+from vllm.config.model import LogprobsMode
 
 from ...conftest import HfRunner, VllmRunner
 

vllm/config/__init__.py

 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
-from vllm.config.cache import (
-    BlockSize,
-    CacheConfig,
-    CacheDType,
-    MambaDType,
-    PrefixCachingHashAlgo,
-)
+from vllm.config.cache import CacheConfig
 from vllm.config.compilation import (
     CompilationConfig,
     CompilationLevel,
     CUDAGraphMode,
     PassConfig,
 )
-from vllm.config.device import Device, DeviceConfig
+from vllm.config.device import DeviceConfig
 from vllm.config.kv_events import KVEventsConfig
 from vllm.config.kv_transfer import KVTransferConfig
 from vllm.config.load import LoadConfig
 from vllm.config.lora import LoRAConfig
 from vllm.config.model import (
-    ConvertOption,
-    HfOverrides,
-    LogprobsMode,
     ModelConfig,
-    ModelDType,
-    ModelImpl,
-    RunnerOption,
-    TaskOption,
-    TokenizerMode,
     iter_architecture_defaults,
     try_match_architecture_defaults,
 )
-from vllm.config.multimodal import MMCacheType, MMEncoderTPMode, MultiModalConfig
-from vllm.config.observability import DetailedTraceModules, ObservabilityConfig
-from vllm.config.parallel import DistributedExecutorBackend, EPLBConfig, ParallelConfig
+from vllm.config.multimodal import MultiModalConfig
+from vllm.config.observability import ObservabilityConfig
+from vllm.config.parallel import EPLBConfig, ParallelConfig
 from vllm.config.pooler import PoolerConfig
-from vllm.config.scheduler import RunnerType, SchedulerConfig, SchedulerPolicy
+from vllm.config.scheduler import SchedulerConfig
 from vllm.config.speculative import SpeculativeConfig
 from vllm.config.speech_to_text import SpeechToTextConfig
 from vllm.config.structured_outputs import StructuredOutputsConfig
@@ -56,20 +42,17 @@ from vllm.config.vllm import (
     set_current_vllm_config,
 )
 
+# __all__ should only contain classes and functions.
+# Types and globals should be imported from their respective modules.
 __all__ = [
     # From vllm.config.cache
-    "BlockSize",
     "CacheConfig",
-    "CacheDType",
-    "MambaDType",
-    "PrefixCachingHashAlgo",
     # From vllm.config.compilation
     "CompilationConfig",
     "CompilationLevel",
     "CUDAGraphMode",
     "PassConfig",
     # From vllm.config.device
-    "Device",
     "DeviceConfig",
     # From vllm.config.kv_events
     "KVEventsConfig",
@@ -80,34 +63,20 @@ __all__ = [
     # From vllm.config.lora
     "LoRAConfig",
     # From vllm.config.model
-    "ConvertOption",
-    "HfOverrides",
-    "LogprobsMode",
     "ModelConfig",
-    "ModelDType",
-    "ModelImpl",
-    "RunnerOption",
-    "TaskOption",
-    "TokenizerMode",
     "iter_architecture_defaults",
     "try_match_architecture_defaults",
     # From vllm.config.multimodal
-    "MMCacheType",
-    "MMEncoderTPMode",
     "MultiModalConfig",
     # From vllm.config.observability
-    "DetailedTraceModules",
     "ObservabilityConfig",
     # From vllm.config.parallel
-    "DistributedExecutorBackend",
     "EPLBConfig",
     "ParallelConfig",
     # From vllm.config.pooler
     "PoolerConfig",
     # From vllm.config.scheduler
-    "RunnerType",
     "SchedulerConfig",
-    "SchedulerPolicy",
     # From vllm.config.speculative
     "SpeculativeConfig",
     # From vllm.config.speech_to_text

vllm/engine/arg_utils.py

@@ -31,43 +31,41 @@ from typing_extensions import TypeIs, deprecated
 
 import vllm.envs as envs
 from vllm.config import (
-    BlockSize,
     CacheConfig,
-    CacheDType,
     CompilationConfig,
     ConfigType,
-    ConvertOption,
-    DetailedTraceModules,
-    Device,
     DeviceConfig,
-    DistributedExecutorBackend,
     EPLBConfig,
-    HfOverrides,
     KVEventsConfig,
     KVTransferConfig,
     LoadConfig,
-    LogprobsMode,
     LoRAConfig,
-    MambaDType,
-    MMEncoderTPMode,
     ModelConfig,
-    ModelDType,
+    MultiModalConfig,
     ObservabilityConfig,
     ParallelConfig,
     PoolerConfig,
-    PrefixCachingHashAlgo,
-    RunnerOption,
     SchedulerConfig,
-    SchedulerPolicy,
     SpeculativeConfig,
     StructuredOutputsConfig,
-    TaskOption,
-    TokenizerMode,
     VllmConfig,
     get_attr_docs,
 )
-from vllm.config.multimodal import MMCacheType, MultiModalConfig
-from vllm.config.parallel import ExpertPlacementStrategy
+from vllm.config.cache import BlockSize, CacheDType, MambaDType, PrefixCachingHashAlgo
+from vllm.config.device import Device
+from vllm.config.model import (
+    ConvertOption,
+    HfOverrides,
+    LogprobsMode,
+    ModelDType,
+    RunnerOption,
+    TaskOption,
+    TokenizerMode,
+)
+from vllm.config.multimodal import MMCacheType, MMEncoderTPMode
+from vllm.config.observability import DetailedTraceModules
+from vllm.config.parallel import DistributedExecutorBackend, ExpertPlacementStrategy
+from vllm.config.scheduler import SchedulerPolicy
 from vllm.config.utils import get_field
 from vllm.logger import init_logger
 from vllm.platforms import CpuArchEnum, current_platform

vllm/entrypoints/llm.py

@@ -19,18 +19,18 @@ from vllm.beam_search import (
 )
 from vllm.config import (
     CompilationConfig,
-    ModelDType,
+    PoolerConfig,
     StructuredOutputsConfig,
-    TokenizerMode,
     is_init_field,
 )
-from vllm.engine.arg_utils import (
+from vllm.config.model import (
     ConvertOption,
-    EngineArgs,
     HfOverrides,
-    PoolerConfig,
+    ModelDType,
     RunnerOption,
+    TokenizerMode,
 )
+from vllm.engine.arg_utils import EngineArgs
 from vllm.entrypoints.chat_utils import (
     ChatCompletionMessageParam,
     ChatTemplateContentFormatOption,

vllm/model_executor/layers/mamba/mamba_utils.py

@@ -4,7 +4,8 @@ from typing import Union
 
 import torch
 
-from vllm.config import MambaDType, ModelDType
+from vllm.config.cache import MambaDType
+from vllm.config.model import ModelDType
 from vllm.distributed import divide
 from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE, get_kv_cache_torch_dtype
 

vllm/platforms/tpu.py

@@ -15,7 +15,8 @@ from .interface import Platform, PlatformEnum
 
 if TYPE_CHECKING:
     from vllm.attention.backends.registry import _Backend
-    from vllm.config import BlockSize, ModelConfig, VllmConfig
+    from vllm.config import ModelConfig, VllmConfig
+    from vllm.config.cache import BlockSize
     from vllm.pooling_params import PoolingParams
 else:
     BlockSize = None

vllm/v1/sample/ops/topk_topp_sampler.py

@@ -8,7 +8,7 @@ import torch.nn as nn
 from packaging import version
 
 from vllm import envs
-from vllm.config import LogprobsMode
+from vllm.config.model import LogprobsMode
 from vllm.logger import init_logger
 from vllm.platforms import current_platform
 

vllm/v1/sample/sampler.py

@@ -7,7 +7,7 @@ from typing import Optional
 import torch
 import torch.nn as nn
 
-from vllm.config import LogprobsMode
+from vllm.config.model import LogprobsMode
 from vllm.utils import is_pin_memory_available
 from vllm.v1.outputs import LogprobsTensors, SamplerOutput
 from vllm.v1.sample.metadata import SamplingMetadata