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

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

vllm/model_executor/layers/mamba/short_conv.py

@@ -4,7 +4,6 @@
 
 import torch
 
-from vllm.attention.backends.abstract import AttentionMetadata
 from vllm.config import CacheConfig, ModelConfig, get_current_vllm_config
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.forward_context import ForwardContext, get_forward_context
@@ -24,6 +23,7 @@ from vllm.model_executor.layers.mamba.ops.causal_conv1d import (
     causal_conv1d_update,
 )
 from vllm.utils.torch_utils import direct_register_custom_op
+from vllm.v1.attention.backend import AttentionMetadata
 from vllm.v1.attention.backends.short_conv_attn import ShortConvAttentionMetadata
 
 

vllm/model_executor/models/afmoe.py

@@ -9,7 +9,6 @@ from itertools import islice
 import torch
 from torch import nn
 
-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, get_current_vllm_config
@@ -50,6 +49,7 @@ from vllm.model_executor.models.utils import (
     maybe_prefix,
 )
 from vllm.sequence import IntermediateTensors
+from vllm.v1.attention.backend import AttentionType
 
 logger = init_logger(__name__)
 

vllm/model_executor/models/aimv2.py

@@ -8,10 +8,10 @@ from collections.abc import Iterable
 import torch
 import torch.nn as nn
 
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.distributed import get_tensor_model_parallel_world_size
 from vllm.distributed.utils import divide
 from vllm.model_executor.layers.activation import SiluAndMul
+from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
 from vllm.model_executor.layers.conv import Conv2dLayer
 from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (

vllm/model_executor/models/apertus.py

@@ -32,13 +32,14 @@ import torch
 from torch import nn
 from transformers import ApertusConfig
 
-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 XIELU
+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 (
     ColumnParallelLinear,
@@ -57,6 +58,7 @@ from vllm.model_executor.model_loader.weight_utils import (
     maybe_remap_kv_scale_name,
 )
 from vllm.sequence import IntermediateTensors
+from vllm.v1.attention.backend import AttentionType
 
 from .interfaces import SupportsLoRA, SupportsPP
 from .utils import (

vllm/model_executor/models/bert.py

@@ -7,11 +7,13 @@ import torch
 from torch import nn
 from transformers import BertConfig
 
-from vllm.attention.layers.encoder_only_attention import EncoderOnlyAttention
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, PoolerConfig, VllmConfig
 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.encoder_only_attention import (
+    EncoderOnlyAttention,
+)
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,
     QKVParallelLinear,

vllm/model_executor/models/bert_with_rope.py

@@ -6,7 +6,6 @@ import torch
 from torch import nn
 from transformers import PretrainedConfig
 
-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 (
@@ -16,6 +15,9 @@ from vllm.distributed import (
     tensor_model_parallel_all_reduce,
 )
 from vllm.model_executor.layers.activation import get_act_and_mul_fn, get_act_fn
+from vllm.model_executor.layers.attention.encoder_only_attention import (
+    EncoderOnlyAttention,
+)
 from vllm.model_executor.layers.fused_moe import activation_without_mul, fused_topk
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,

vllm/model_executor/models/blip.py

@@ -9,9 +9,9 @@ import torch
 import torch.nn as nn
 from transformers import Blip2VisionConfig, BlipVisionConfig
 
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 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/clip.py

@@ -15,11 +15,11 @@ from transformers import (
 )
 
 from vllm.attention.layer import Attention
-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.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/config.py

@@ -4,12 +4,12 @@ from copy import deepcopy
 from math import lcm
 from typing import TYPE_CHECKING
 
-from vllm.attention.backends.registry import AttentionBackendEnum
 from vllm.logger import init_logger
 from vllm.model_executor.models import ModelRegistry
 from vllm.platforms import current_platform
 from vllm.utils.math_utils import cdiv, round_up
 from vllm.utils.torch_utils import STR_DTYPE_TO_TORCH_DTYPE
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 from vllm.v1.kv_cache_interface import FullAttentionSpec, MambaSpec, MLAAttentionSpec
 
 if TYPE_CHECKING:

vllm/model_executor/models/deepencoder.py

@@ -18,8 +18,8 @@ import torch.nn as nn
 import torch.nn.functional as F
 from transformers import CLIPVisionConfig
 
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.config import MultiModalConfig
+from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
 from vllm.model_executor.layers.conv import Conv2dLayer
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader

vllm/model_executor/models/deepseek_v2.py

@@ -33,9 +33,7 @@ from torch import nn
 from transformers import DeepseekV2Config, DeepseekV3Config
 
 from vllm._aiter_ops import rocm_aiter_ops
-from vllm.attention.backends.abstract import AttentionBackend
 from vllm.attention.layer import Attention
-from vllm.attention.ops.common import pack_seq_triton, unpack_seq_triton
 from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, ParallelConfig, VllmConfig, get_current_vllm_config
 from vllm.distributed import (
@@ -78,10 +76,12 @@ from vllm.platforms import current_platform
 from vllm.sequence import IntermediateTensors
 from vllm.utils.deep_gemm import fp8_mqa_logits, fp8_paged_mqa_logits
 from vllm.utils.torch_utils import direct_register_custom_op
+from vllm.v1.attention.backend import AttentionBackend
 from vllm.v1.attention.backends.mla.indexer import (
     DeepseekV32IndexerBackend,
     DeepseekV32IndexerMetadata,
 )
+from vllm.v1.attention.ops.common import pack_seq_triton, unpack_seq_triton
 from vllm.v1.kv_cache_interface import KVCacheSpec, MLAAttentionSpec
 from vllm.v1.worker.workspace import current_workspace_manager
 
@@ -679,7 +679,9 @@ def sparse_attn_indexer(
             )
             fp8_mqa_logits_func = fp8_mqa_logits
             if current_platform.is_rocm():
-                from vllm.attention.ops.rocm_aiter_mla_sparse import rocm_fp8_mqa_logits
+                from vllm.v1.attention.ops.rocm_aiter_mla_sparse import (
+                    rocm_fp8_mqa_logits,
+                )
 
                 fp8_mqa_logits_func = rocm_fp8_mqa_logits
             logits = fp8_mqa_logits_func(
@@ -729,7 +731,7 @@ def sparse_attn_indexer(
         num_padded_tokens = batch_size * next_n
         fp8_paged_mqa_logits_func = fp8_paged_mqa_logits
         if current_platform.is_rocm():
-            from vllm.attention.ops.rocm_aiter_mla_sparse import (
+            from vllm.v1.attention.ops.rocm_aiter_mla_sparse import (
                 rocm_fp8_paged_mqa_logits,
             )
 

vllm/model_executor/models/dots_ocr.py

@@ -8,10 +8,6 @@ import torch.nn as nn
 from torch.nn import LayerNorm
 from transformers.models.qwen2_vl import Qwen2VLProcessor
 
-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 utils as dist_utils
@@ -20,6 +16,9 @@ from vllm.distributed.parallel_state import (
     get_tensor_model_parallel_world_size,
 )
 from vllm.model_executor.layers.activation import SiluAndMul
+from vllm.model_executor.layers.attention.mm_encoder_attention import (
+    MMEncoderAttention,
+)
 from vllm.model_executor.layers.conv import Conv2dLayer
 from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
@@ -59,6 +58,7 @@ from vllm.multimodal.inputs import MultiModalDataDict
 from vllm.sequence import IntermediateTensors
 from vllm.transformers_utils.configs.dotsocr import DotsOCRConfig, DotsVisionConfig
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 from .vision import run_dp_sharded_mrope_vision_model
 

vllm/model_executor/models/ernie45_vl.py

@@ -36,16 +36,15 @@ import torch.nn.functional as F
 from einops import rearrange
 from transformers import BatchFeature
 
-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, VideoDummyOptions
 from vllm.distributed import parallel_state
 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.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
     ColumnParallelLinear,
@@ -74,6 +73,7 @@ from vllm.multimodal.processing import (
 from vllm.multimodal.profiling import BaseDummyInputsBuilder
 from vllm.sequence import IntermediateTensors
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 from .ernie45_vl_moe import Ernie4_5_VLMoeForCausalLM
 from .interfaces import (

vllm/model_executor/models/gemma3.py

@@ -22,13 +22,15 @@ import torch
 from torch import nn
 from transformers import Gemma3TextConfig
 
-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
 from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
 from vllm.logger import init_logger
 from vllm.model_executor.layers.activation import GeluAndMul
+from vllm.model_executor.layers.attention.encoder_only_attention import (
+    EncoderOnlyAttention,
+)
 from vllm.model_executor.layers.layernorm import GemmaRMSNorm
 from vllm.model_executor.layers.linear import (
     MergedColumnParallelLinear,
@@ -47,8 +49,8 @@ from vllm.model_executor.model_loader.weight_utils import (
     maybe_remap_kv_scale_name,
 )
 from vllm.sequence import IntermediateTensors
+from vllm.v1.attention.backend import AttentionType
 
-from ...attention.layers.encoder_only_attention import EncoderOnlyAttention
 from .interfaces import SupportsLoRA, SupportsPP
 from .utils import (
     AutoWeightsLoader,

vllm/model_executor/models/glm4.py

@@ -29,7 +29,6 @@ import torch
 from torch import nn
 from transformers import Glm4Config
 
-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
@@ -41,6 +40,7 @@ from vllm.model_executor.layers.quantization import QuantizationConfig
 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.v1.attention.backend import AttentionType
 
 from .interfaces import SupportsLoRA, SupportsPP
 from .llama import LlamaMLP as Glm4MLP

vllm/model_executor/models/glm4_1v.py

@@ -46,15 +46,14 @@ from transformers.models.glm4v.image_processing_glm4v import (
 from transformers.models.glm4v.video_processing_glm4v import Glm4vVideoProcessor
 from transformers.video_utils import VideoMetadata
 
-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, VideoDummyOptions
 from vllm.distributed import get_tensor_model_parallel_world_size, parallel_state
 from vllm.distributed import utils as dist_utils
 from vllm.logger import init_logger
+from vllm.model_executor.layers.attention.mm_encoder_attention import (
+    MMEncoderAttention,
+)
 from vllm.model_executor.layers.conv import Conv2dLayer, Conv3dLayer
 from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (
@@ -89,6 +88,7 @@ from vllm.multimodal.processing import (
 from vllm.multimodal.profiling import BaseDummyInputsBuilder
 from vllm.sequence import IntermediateTensors
 from vllm.utils.tensor_schema import TensorSchema, TensorShape
+from vllm.v1.attention.backends.registry import AttentionBackendEnum
 
 from ..layers.activation import SiluAndMul
 from .interfaces import (

vllm/model_executor/models/glm4v.py

@@ -19,11 +19,11 @@ from transformers import BatchFeature, PreTrainedTokenizer, TensorType
 from transformers.image_utils import ImageInput
 from transformers.tokenization_utils_base import TextInput
 
-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 SiluAndMul, 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/glmasr.py

@@ -11,12 +11,12 @@ from transformers import BatchFeature
 from transformers.models.glmasr import GlmAsrConfig, GlmAsrProcessor
 from transformers.models.whisper import WhisperFeatureExtractor
 
-from vllm.attention.layers.mm_encoder_attention import MMEncoderAttention
 from vllm.config import ModelConfig, SpeechToTextConfig, VllmConfig
 from vllm.config.multimodal import BaseDummyOptions
 from vllm.distributed.parallel_state import get_tensor_model_parallel_world_size
 from vllm.inputs.data import PromptType
 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/gpt_oss.py

@@ -7,7 +7,6 @@ import torch.distributed as dist
 from torch import nn
 from transformers import GptOssConfig
 
-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
@@ -37,6 +36,7 @@ from vllm.model_executor.models.utils import sequence_parallel_chunk
 from vllm.platforms import current_platform
 from vllm.sequence import IntermediateTensors
 from vllm.utils.math_utils import cdiv
+from vllm.v1.attention.backend import AttentionType
 
 from .interfaces import SupportsEagle3, SupportsLoRA, SupportsPP
 from .utils import (

vllm/model_executor/models/hunyuan_v1.py

@@ -33,7 +33,6 @@ import torch
 from torch import nn
 from transformers import PretrainedConfig
 
-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, get_current_vllm_config
@@ -65,6 +64,7 @@ from vllm.model_executor.model_loader.weight_utils import (
     maybe_remap_kv_scale_name,
 )
 from vllm.sequence import IntermediateTensors
+from vllm.v1.attention.backend import AttentionType
 
 from .interfaces import MixtureOfExperts, SupportsLoRA, SupportsPP
 from .utils import (