[1/N][Attention] Restructure attention: move files (#31916)

Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>

vllm/model_executor/models/plamo2.py

@@ -9,7 +9,6 @@ import torch
 from torch import nn
 from transformers import PretrainedConfig
 
-from vllm.attention.backends.abstract import AttentionMetadata
 from vllm.attention.layer import Attention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import VllmConfig, get_current_vllm_config
@@ -66,6 +65,7 @@ from vllm.model_executor.models.utils import (
 from vllm.model_executor.utils import set_weight_attrs
 from vllm.sequence import IntermediateTensors
 from vllm.utils.torch_utils import direct_register_custom_op
+from vllm.v1.attention.backend import AttentionMetadata
 from vllm.v1.attention.backends.mamba2_attn import Mamba2AttentionMetadata
 
 

vllm/model_executor/models/qwen2.py

@@ -33,13 +33,14 @@ import torch
 from torch import nn
 from transformers import Qwen2Config
 
-from vllm.attention.backends.abstract import AttentionType
 from vllm.attention.layer import Attention
-from vllm.attention.layers.encoder_only_attention import EncoderOnlyAttention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, VllmConfig
 from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import SiluAndMul
+from vllm.model_executor.layers.attention.encoder_only_attention import (
+    EncoderOnlyAttention,
+)
 from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
@@ -59,6 +60,7 @@ from vllm.model_executor.model_loader.weight_utils import (
 )
 from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.config import is_interleaved, set_default_rope_theta
+from vllm.v1.attention.backend import AttentionType
 
 from .interfaces import SupportsEagle3, SupportsLoRA, SupportsPP
 from .utils import (

vllm/model_executor/models/qwen2_5_vl.py

@@ -41,8 +41,6 @@ from transformers.models.qwen2_5_vl.configuration_qwen2_5_vl import (
     Qwen2_5_VLVisionConfig,
 )
 
-from vllm.attention.backends.registry import AttentionBackendEnum
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import MultiModalConfig, VllmConfig
 from vllm.distributed import parallel_state
@@ -50,6 +48,7 @@ from vllm.distributed import utils as dist_utils
 from vllm.forward_context import set_forward_context
 from vllm.logger import init_logger
 from vllm.model_executor.layers.activation import get_act_and_mul_fn
+from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
 from vllm.model_executor.layers.conv import Conv3dLayer
 from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
@@ -83,6 +82,7 @@ from vllm.multimodal.processing import PromptReplacement, PromptUpdate
 from vllm.sequence import IntermediateTensors
 from vllm.utils.platform_utils import is_pin_memory_available
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 from .interfaces import (
     MultiModalEmbeddings,

vllm/model_executor/models/qwen2_vl.py

@@ -43,14 +43,13 @@ from transformers.models.qwen2_vl.configuration_qwen2_vl import (
 from transformers.models.qwen2_vl.image_processing_qwen2_vl import smart_resize
 from transformers.models.qwen2_vl.video_processing_qwen2_vl import Qwen2VLVideoProcessor
 
-from vllm.attention.backends.registry import AttentionBackendEnum
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.config import MultiModalConfig, VllmConfig
 from vllm.config.multimodal import BaseDummyOptions
 from vllm.distributed import parallel_state, tensor_model_parallel_all_gather
 from vllm.distributed import utils as dist_utils
 from vllm.logger import init_logger
 from vllm.model_executor.layers.activation import QuickGELU
+from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
 from vllm.model_executor.layers.conv import Conv3dLayer
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,
@@ -90,6 +89,7 @@ from vllm.multimodal.profiling import BaseDummyInputsBuilder
 from vllm.sequence import IntermediateTensors
 from vllm.tokenizers import TokenizerLike
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 from .interfaces import (
     MultiModalEmbeddings,

vllm/model_executor/models/qwen3.py

@@ -30,7 +30,6 @@ import torch
 from torch import nn
 from transformers import Qwen3Config
 
-from vllm.attention.backends.abstract import AttentionType
 from vllm.attention.layer import Attention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, VllmConfig
@@ -44,6 +43,7 @@ from vllm.model_executor.layers.rotary_embedding import get_rope
 from vllm.model_executor.layers.vocab_parallel_embedding import ParallelLMHead
 from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.config import set_default_rope_theta
+from vllm.v1.attention.backend import AttentionType
 
 from .interfaces import SupportsEagle3, SupportsLoRA, SupportsPP
 from .qwen2 import Qwen2MLP as Qwen3MLP

vllm/model_executor/models/qwen3_next.py

@@ -10,7 +10,6 @@ from einops import rearrange
 from torch import nn
 from transformers.activations import ACT2FN
 
-from vllm.attention.backends.abstract import AttentionMetadata
 from vllm.attention.layer import Attention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import (
@@ -75,6 +74,7 @@ from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.configs import Qwen3NextConfig
 from vllm.triton_utils import tl, triton
 from vllm.utils.torch_utils import direct_register_custom_op
+from vllm.v1.attention.backend import AttentionMetadata
 from vllm.v1.attention.backends.gdn_attn import GDNAttentionMetadata
 
 from .interfaces import (

vllm/model_executor/models/qwen3_omni_moe_thinker.py

@@ -46,7 +46,6 @@ from transformers.models.qwen3_omni_moe.processing_qwen3_omni_moe import (
 )
 from transformers.models.whisper import WhisperFeatureExtractor
 
-from vllm.attention.backends.registry import AttentionBackendEnum
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import MultiModalConfig, VllmConfig
 from vllm.distributed import get_pp_group
@@ -75,6 +74,7 @@ from vllm.multimodal.processing import (
     PromptUpdateDetails,
 )
 from vllm.sequence import IntermediateTensors
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 from .interfaces import (
     MultiModalEmbeddings,

vllm/model_executor/models/qwen3_vl.py

@@ -48,7 +48,6 @@ from transformers.models.qwen3_vl.video_processing_qwen3_vl import (
 )
 from transformers.video_utils import VideoMetadata
 
-from vllm.attention.backends.registry import AttentionBackendEnum
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import MultiModalConfig, VllmConfig
 from vllm.config.multimodal import BaseDummyOptions, VideoDummyOptions
@@ -92,6 +91,7 @@ from vllm.multimodal.processing import (
 from vllm.multimodal.profiling import BaseDummyInputsBuilder
 from vllm.sequence import IntermediateTensors
 from vllm.utils.collection_utils import is_list_of
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 from .interfaces import (
     MultiModalEmbeddings,

vllm/model_executor/models/seed_oss.py

@@ -30,7 +30,6 @@ import torch
 from torch import nn
 from transformers import PretrainedConfig as SeedOssConfig
 
-from vllm.attention.backends.abstract import AttentionType
 from vllm.attention.layer import Attention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, VllmConfig
@@ -56,6 +55,7 @@ from vllm.model_executor.model_loader.weight_utils import (
 )
 from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.config import set_default_rope_theta
+from vllm.v1.attention.backend import AttentionType
 
 from .interfaces import SupportsLoRA, SupportsPP
 from .utils import (

vllm/model_executor/models/siglip.py

@@ -15,12 +15,14 @@ from transformers import (
     SiglipVisionConfig,
 )
 
-from vllm.attention.layers.encoder_only_attention import EncoderOnlyAttention
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.config import VllmConfig
 from vllm.config.multimodal import BaseDummyOptions, MultiModalConfig
 from vllm.distributed import divide, get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import get_act_fn
+from vllm.model_executor.layers.attention.encoder_only_attention import (
+    EncoderOnlyAttention,
+)
+from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
 from vllm.model_executor.layers.conv import Conv2dLayer
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,

vllm/model_executor/models/siglip2.py

@@ -10,11 +10,11 @@ from torch import nn
 from torch.nn import functional as F
 from transformers import Siglip2VisionConfig
 
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import MultiModalConfig
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import get_act_fn
+from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,
     QKVParallelLinear,

vllm/model_executor/models/siglip2navit.py

@@ -11,10 +11,10 @@ from torch.nn import functional as F
 from transformers import Siglip2VisionConfig
 from transformers.configuration_utils import PretrainedConfig
 
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.config import MultiModalConfig
 from vllm.distributed import divide, get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import get_act_fn
+from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
 from vllm.model_executor.layers.conv import Conv2dLayer
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,

vllm/model_executor/models/step3_vl.py

@@ -15,11 +15,11 @@ from torchvision import transforms
 from torchvision.transforms.functional import InterpolationMode
 from transformers import BatchFeature, PretrainedConfig, TensorType
 
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.config import VllmConfig
 from vllm.config.multimodal import BaseDummyOptions
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.model_executor.layers.activation import get_act_fn
+from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
 from vllm.model_executor.layers.conv import Conv2dLayer
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,

vllm/model_executor/models/transformers/base.py

@@ -27,13 +27,14 @@ from torch import nn
 from transformers import AutoModel
 from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS
 
-from vllm.attention.backends.abstract import AttentionType
 from vllm.attention.layer import Attention
-from vllm.attention.layers.encoder_only_attention import EncoderOnlyAttention
 from vllm.config.utils import getattr_iter
 from vllm.distributed import get_pp_group, get_tp_group
 from vllm.distributed.utils import get_pp_indices
 from vllm.logger import init_logger
+from vllm.model_executor.layers.attention.encoder_only_attention import (
+    EncoderOnlyAttention,
+)
 from vllm.model_executor.layers.vocab_parallel_embedding import VocabParallelEmbedding
 from vllm.model_executor.models.interfaces import (
     SupportsEagle,
@@ -59,6 +60,7 @@ from vllm.model_executor.models.utils import (
     maybe_prefix,
 )
 from vllm.sequence import IntermediateTensors
+from vllm.v1.attention.backend import AttentionType
 
 if TYPE_CHECKING:
     from transformers import PreTrainedModel

vllm/model_executor/models/vision.py

@@ -10,7 +10,6 @@ from typing import Final, Generic, Literal, Protocol, TypeAlias, TypeVar
 import torch
 from transformers import PretrainedConfig
 
-from vllm.attention.backends.registry import AttentionBackendEnum
 from vllm.config import VllmConfig
 from vllm.distributed import (
     get_tensor_model_parallel_rank,
@@ -19,6 +18,7 @@ from vllm.distributed import (
 )
 from vllm.logger import init_logger
 from vllm.platforms import current_platform
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 logger = init_logger(__name__)
 

vllm/model_executor/models/whisper.py

@@ -18,18 +18,15 @@ from transformers import (
 )
 from transformers.models.whisper.modeling_whisper import sinusoids
 
-from vllm.attention.backends.abstract import (
-    AttentionType,
-)
 from vllm.attention.layer import Attention
-from vllm.attention.layers.cross_attention import CrossAttention
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.config import CacheConfig, ModelConfig, SpeechToTextConfig, VllmConfig
 from vllm.config.multimodal import BaseDummyOptions
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.inputs.data import PromptType
 from vllm.logger import init_logger
 from vllm.model_executor.layers.activation import get_act_fn
+from vllm.model_executor.layers.attention.cross_attention import CrossAttention
+from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,
     QKVParallelLinear,
@@ -62,6 +59,9 @@ from vllm.transformers_utils.processor import cached_processor_from_config
 from vllm.utils.jsontree import json_map_leaves
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
 from vllm.utils.torch_utils import set_default_torch_dtype
+from vllm.v1.attention.backend import (
+    AttentionType,
+)
 
 from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsTranscription
 from .utils import (

vllm/model_executor/models/whisper_utils.py

@@ -9,20 +9,20 @@ import torch
 import torch.nn.functional as F
 from torch import nn
 
-from vllm.attention.backends.abstract import (
+from vllm.attention.layer import Attention
+from vllm.config import CacheConfig, VllmConfig
+from vllm.model_executor.layers.quantization import QuantizationConfig
+from vllm.v1.attention.backend import (
     AttentionBackend,
     AttentionMetadata,
     AttentionType,
 )
-from vllm.attention.layer import Attention
-from vllm.attention.selector import get_attn_backend
-from vllm.config import CacheConfig, VllmConfig
-from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.v1.attention.backends.flash_attn import FlashAttentionBackend
 from vllm.v1.attention.backends.utils import (
     CommonAttentionMetadata,
     subclass_attention_backend_with_overrides,
 )
+from vllm.v1.attention.selector import get_attn_backend
 from vllm.v1.kv_cache_interface import AttentionSpec
 
 # From https://platform.openai.com/docs/guides/speech-to-text/supported-languages

vllm/platforms/cpu.py

@@ -15,16 +15,16 @@ import regex as re
 import torch
 
 from vllm import envs
-from vllm.attention.backends.registry import AttentionBackendEnum
 from vllm.logger import init_logger
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 from .interface import CpuArchEnum, Platform, PlatformEnum
 
 logger = init_logger(__name__)
 
 if TYPE_CHECKING:
-    from vllm.attention.selector import AttentionSelectorConfig
     from vllm.config import VllmConfig
+    from vllm.v1.attention.selector import AttentionSelectorConfig
 else:
     VllmConfig = None
 

vllm/platforms/cuda.py

@@ -14,17 +14,17 @@ from typing_extensions import ParamSpec
 
 # import custom ops, trigger op registration
 import vllm._C  # noqa
-from vllm.attention.backends.registry import AttentionBackendEnum
 from vllm.logger import init_logger
 from vllm.utils.import_utils import import_pynvml
 from vllm.utils.torch_utils import cuda_device_count_stateless
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 from .interface import DeviceCapability, Platform, PlatformEnum
 
 if TYPE_CHECKING:
-    from vllm.attention.selector import AttentionSelectorConfig
     from vllm.config import VllmConfig
     from vllm.config.cache import CacheDType
+    from vllm.v1.attention.selector import AttentionSelectorConfig
 else:
     VllmConfig = None
     CacheDType = None
@@ -148,7 +148,7 @@ class CudaPlatformBase(Platform):
 
     @classmethod
     def check_and_update_config(cls, vllm_config: "VllmConfig") -> None:
-        from vllm.attention.backends.registry import AttentionBackendEnum
+        from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
         parallel_config = vllm_config.parallel_config
         model_config = vllm_config.model_config
@@ -200,7 +200,7 @@ class CudaPlatformBase(Platform):
                 use_cutlass_mla = backend == AttentionBackendEnum.CUTLASS_MLA
                 use_flashinfer_mla = backend == AttentionBackendEnum.FLASHINFER_MLA
 
-            from vllm.attention.ops.flashmla import is_flashmla_dense_supported
+            from vllm.v1.attention.ops.flashmla import is_flashmla_dense_supported
 
             if (
                 use_flashmla

vllm/platforms/interface.py

@@ -13,18 +13,18 @@ import numpy as np
 import torch
 from typing_extensions import deprecated
 
-from vllm.attention.backends.registry import AttentionBackendEnum
 from vllm.logger import init_logger
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 if TYPE_CHECKING:
     from torch.distributed import PrefixStore, ProcessGroup
 
-    from vllm.attention.selector import AttentionSelectorConfig
     from vllm.config import VllmConfig
     from vllm.inputs import ProcessorInputs, PromptType
     from vllm.pooling_params import PoolingParams
     from vllm.sampling_params import SamplingParams
     from vllm.utils.argparse_utils import FlexibleArgumentParser
+    from vllm.v1.attention.selector import AttentionSelectorConfig
 else:
     FlexibleArgumentParser = object