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 12, 12:15 UTC
[Attention] Remove imports from `vllm/attention/__init__.py` (#29342)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Matthew Bonanni committed
Nov 26, 2025 at 12:53 UTC
430dd4d9eb7e342e28012351df06d93892f86741
96 files changed
+120
-121
docs/contributing/model/basic.md
+1
-1
@@ -29,7 +29,7 @@ The initialization code should look like this:
29
```python
30
from torch import nn
31
from vllm.config import VllmConfig
32
- from vllm.attention import Attention
32
+ from vllm.attention.layer import Attention
33
34
class MyAttention(nn.Module):
35
def __init__(self, vllm_config: VllmConfig, prefix: str):
tests/compile/test_fusion_attn.py
+2
-1
@@ -9,8 +9,9 @@ from tests.compile.backend import LazyInitPass, TestBackend
9
from tests.utils import flat_product
10
from tests.v1.attention.utils import BatchSpec, create_common_attn_metadata
11
from vllm._custom_ops import cutlass_scaled_fp4_mm, scaled_fp4_quant
12
-from vllm.attention import Attention, AttentionMetadata
12
+from vllm.attention.backends.abstract import AttentionMetadata
13
from vllm.attention.backends.registry import AttentionBackendEnum
14
+from vllm.attention.layer import Attention
15
from vllm.attention.selector import global_force_attn_backend_context_manager
16
from vllm.compilation.fusion_attn import ATTN_OP, AttnFusionPass
17
from vllm.compilation.fx_utils import find_op_nodes
tests/compile/test_qk_norm_rope_fusion.py
+2
-1
@@ -5,7 +5,8 @@ import pytest
5
import torch
6
7
from tests.compile.backend import TestBackend
8
-from vllm.attention import Attention, AttentionType
8
+from vllm.attention.backends.abstract import AttentionType
9
+from vllm.attention.layer import Attention
10
from vllm.compilation.matcher_utils import FLASHINFER_ROTARY_OP, RMS_OP, ROTARY_OP
11
from vllm.compilation.noop_elimination import NoOpEliminationPass
12
from vllm.compilation.post_cleanup import PostCleanupPass
tests/kernels/utils.py
+1
-1
@@ -14,7 +14,7 @@ import torch
14
from torch._prims_common import TensorLikeType
15
16
from tests.kernels.quant_utils import native_w8a8_block_matmul
17
-from vllm.attention import AttentionType
17
+from vllm.attention.backends.abstract import AttentionType
18
from vllm.model_executor.layers.activation import SiluAndMul
19
from vllm.model_executor.layers.fused_moe.utils import moe_kernel_quantize_input
20
from vllm.utils import (
tests/v1/worker/test_gpu_model_runner.py
+1
-1
@@ -5,8 +5,8 @@ import numpy as np
5
import pytest
6
import torch
7
8
-from vllm.attention import Attention
8
from vllm.attention.backends.abstract import MultipleOf
9
+from vllm.attention.layer import Attention
10
from vllm.config import (
11
CacheConfig,
12
ModelConfig,
tests/v1/worker/test_utils.py
+2
-2
@@ -7,7 +7,7 @@ from vllm.v1.worker.utils import bind_kv_cache
7
8
9
def test_bind_kv_cache():
10
- from vllm.attention import Attention
10
+ from vllm.attention.layer import Attention
11
12
ctx = {
13
"layers.0.self_attn": Attention(32, 128, 0.1),
@@ -35,7 +35,7 @@ def test_bind_kv_cache():
35
36
37
def test_bind_kv_cache_non_attention():
38
- from vllm.attention import Attention
38
+ from vllm.attention.layer import Attention
39
40
# example from Jamba PP=2
41
ctx = {
vllm/attention/__init__.py
-19
@@ -1,19 +0,0 @@
1
-# SPDX-License-Identifier: Apache-2.0
2
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
-
4
-from vllm.attention.backends.abstract import (
5
- AttentionBackend,
6
- AttentionMetadata,
7
- AttentionType,
8
-)
9
-from vllm.attention.layer import Attention
10
-from vllm.attention.selector import get_attn_backend, get_mamba_attn_backend
11
-
12
-__all__ = [
13
- "Attention",
14
- "AttentionBackend",
15
- "AttentionMetadata",
16
- "AttentionType",
17
- "get_attn_backend",
18
- "get_mamba_attn_backend",
19
-]
vllm/attention/backends/abstract.py
+1
-1
@@ -178,7 +178,7 @@ class AttentionBackend(ABC):
178
By default, only supports decoder attention.
179
Backends should override this to support other attention types.
180
"""
181
- from vllm.attention import AttentionType
181
+ from vllm.attention.backends.abstract import AttentionType
182
183
return attn_type == AttentionType.DECODER
184
vllm/attention/layer.py
+5
-2
@@ -10,8 +10,11 @@ import torch.nn as nn
10
import torch.nn.functional as F
11
12
import vllm.envs as envs
13
-from vllm.attention import AttentionType
14
-from vllm.attention.backends.abstract import AttentionBackend, MLAAttentionImpl
13
+from vllm.attention.backends.abstract import (
14
+ AttentionBackend,
15
+ AttentionType,
16
+ MLAAttentionImpl,
17
+)
18
from vllm.attention.backends.registry import AttentionBackendEnum
19
from vllm.attention.selector import get_attn_backend
20
from vllm.attention.utils.kv_sharing_utils import validate_kv_sharing_target
vllm/compilation/fusion_attn.py
+1
-1
@@ -10,7 +10,7 @@ from torch import fx
10
from torch._higher_order_ops.auto_functionalize import auto_functionalized
11
from torch._inductor.pattern_matcher import PatternMatcherPass
12
13
-from vllm.attention import Attention
13
+from vllm.attention.layer import Attention
14
from vllm.config import VllmConfig, get_layers_from_vllm_config
15
from vllm.logger import init_logger
16
from vllm.model_executor.layers.quantization.utils.quant_utils import (
vllm/compilation/qk_norm_rope_fusion.py
+1
-1
@@ -9,7 +9,7 @@ from torch import fx
9
from torch._higher_order_ops.auto_functionalize import auto_functionalized
10
from torch._inductor.pattern_matcher import PatternMatcherPass
11
12
-from vllm.attention import Attention
12
+from vllm.attention.layer import Attention
13
from vllm.config import VllmConfig, get_layers_from_vllm_config
14
from vllm.logger import init_logger
15
from vllm.model_executor.layers.rotary_embedding import RotaryEmbedding
vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
+1
-1
@@ -20,7 +20,7 @@ import torch
20
import zmq
21
22
from vllm import envs
23
-from vllm.attention import AttentionBackend
23
+from vllm.attention.backends.abstract import AttentionBackend
24
from vllm.attention.backends.registry import AttentionBackendEnum
25
from vllm.attention.selector import get_attn_backend
26
from vllm.config import VllmConfig
vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py
+2
-1
@@ -8,7 +8,8 @@ from typing import Any, ClassVar
8
9
import torch
10
11
-from vllm.attention import Attention, AttentionBackend, AttentionMetadata
11
+from vllm.attention.backends.abstract import AttentionBackend, AttentionMetadata
12
+from vllm.attention.layer import Attention
13
from vllm.config import VllmConfig, get_layers_from_vllm_config
14
from vllm.distributed.kv_events import BlockRemoved, BlockStored, KVCacheEvent
15
from vllm.distributed.kv_transfer.kv_connector.v1 import (
vllm/model_executor/layers/mamba/linear_attn.py
+1
-1
@@ -8,7 +8,7 @@ import torch.nn.functional as F
8
from einops import rearrange
9
from torch import nn
10
11
-from vllm.attention import AttentionMetadata
11
+from vllm.attention.backends.abstract import AttentionMetadata
12
from vllm.config import CacheConfig, ModelConfig, get_current_vllm_config
13
from vllm.distributed.communication_op import tensor_model_parallel_all_reduce
14
from vllm.distributed.parallel_state import (
vllm/model_executor/model_loader/utils.py
+1
-2
@@ -11,8 +11,7 @@ import torch
11
from torch import nn
12
from typing_extensions import assert_never
13
14
-from vllm.attention import Attention
15
-from vllm.attention.layer import MLAAttention
14
+from vllm.attention.layer import Attention, MLAAttention
15
from vllm.config import ModelConfig, VllmConfig, set_current_vllm_config
16
from vllm.logger import init_logger
17
from vllm.model_executor.layers.quantization.base_config import (
vllm/model_executor/models/afmoe.py
+2
-1
@@ -9,7 +9,8 @@ from itertools import islice
9
import torch
10
from torch import nn
11
12
-from vllm.attention import Attention, AttentionType
12
+from vllm.attention.backends.abstract import AttentionType
13
+from vllm.attention.layer import Attention
14
from vllm.compilation.decorators import support_torch_compile
15
from vllm.config import CacheConfig, VllmConfig, get_current_vllm_config
16
from vllm.distributed import (
vllm/model_executor/models/apertus.py
+2
-1
@@ -32,7 +32,8 @@ import torch
32
from torch import nn
33
from transformers import ApertusConfig
34
35
-from vllm.attention import Attention, AttentionType
35
+from vllm.attention.backends.abstract import AttentionType
36
+from vllm.attention.layer import Attention
37
from vllm.attention.layers.encoder_only_attention import EncoderOnlyAttention
38
from vllm.compilation.decorators import support_torch_compile
39
from vllm.config import CacheConfig, VllmConfig
vllm/model_executor/models/arctic.py
+1
-1
@@ -8,7 +8,7 @@ from itertools import islice
8
import torch
9
from torch import nn
10
11
-from vllm.attention import Attention
11
+from vllm.attention.layer import Attention
12
from vllm.compilation.decorators import support_torch_compile
13
from vllm.config import CacheConfig, VllmConfig
14
from vllm.distributed import (
vllm/model_executor/models/baichuan.py
+1
-1
@@ -29,7 +29,7 @@ import torch
29
from torch import nn
30
from transformers import PretrainedConfig
31
32
-from vllm.attention import Attention
32
+from vllm.attention.layer import Attention
33
from vllm.compilation.decorators import support_torch_compile
34
from vllm.config import CacheConfig, VllmConfig
35
from vllm.distributed import (
vllm/model_executor/models/bailing_moe.py
+1
-1
@@ -32,7 +32,7 @@ import torch.nn.functional as F
32
from torch import nn
33
from transformers.configuration_utils import PretrainedConfig
34
35
-from vllm.attention import Attention
35
+from vllm.attention.layer import Attention
36
from vllm.compilation.decorators import support_torch_compile
37
from vllm.config import CacheConfig, VllmConfig
38
from vllm.distributed import (
vllm/model_executor/models/bloom.py
+1
-1
@@ -27,7 +27,7 @@ import torch
27
from torch import nn
28
from transformers import BloomConfig
29
30
-from vllm.attention import Attention
30
+from vllm.attention.layer import Attention
31
from vllm.compilation.decorators import support_torch_compile
32
from vllm.config import CacheConfig, VllmConfig
33
from vllm.distributed import (
vllm/model_executor/models/chameleon.py
+1
-1
@@ -16,7 +16,7 @@ from transformers import (
16
ChameleonVQVAEConfig,
17
)
18
19
-from vllm.attention import Attention
19
+from vllm.attention.layer import Attention
20
from vllm.config import CacheConfig, VllmConfig
21
from vllm.config.multimodal import BaseDummyOptions
22
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/chatglm.py
+1
-1
@@ -12,7 +12,7 @@ import torch
12
from torch import nn
13
from torch.nn import LayerNorm
14
15
-from vllm.attention import Attention
15
+from vllm.attention.layer import Attention
16
from vllm.compilation.decorators import support_torch_compile
17
from vllm.config import CacheConfig, VllmConfig
18
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/clip.py
+1
-2
@@ -14,8 +14,7 @@ from transformers import (
14
CLIPVisionConfig,
15
)
16
17
-from vllm.attention import Attention
18
-from vllm.attention.layer import MultiHeadAttention
17
+from vllm.attention.layer import Attention, MultiHeadAttention
18
from vllm.config import VllmConfig
19
from vllm.config.multimodal import BaseDummyOptions
20
from vllm.distributed import divide, get_tensor_model_parallel_world_size
vllm/model_executor/models/commandr.py
+1
-1
@@ -30,7 +30,7 @@ import torch
30
from torch import nn
31
from transformers import Cohere2Config, CohereConfig
32
33
-from vllm.attention import Attention
33
+from vllm.attention.layer import Attention
34
from vllm.compilation.decorators import support_torch_compile
35
from vllm.config import CacheConfig, VllmConfig
36
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/dbrx.py
+1
-1
@@ -8,7 +8,7 @@ import torch
8
import torch.nn as nn
9
from transformers import DbrxConfig
10
11
-from vllm.attention import Attention
11
+from vllm.attention.layer import Attention
12
from vllm.config import CacheConfig, VllmConfig
13
from vllm.distributed import (
14
get_pp_group,
vllm/model_executor/models/deepseek_v2.py
+1
-1
@@ -33,8 +33,8 @@ from torch import nn
33
from transformers import DeepseekV2Config, DeepseekV3Config
34
35
from vllm._aiter_ops import rocm_aiter_ops
36
-from vllm.attention import Attention
36
from vllm.attention.backends.abstract import AttentionBackend
37
+from vllm.attention.layer import Attention
38
from vllm.attention.ops.common import pack_seq_triton, unpack_seq_triton
39
from vllm.compilation.decorators import support_torch_compile
40
from vllm.config import CacheConfig, ParallelConfig, VllmConfig, get_current_vllm_config
vllm/model_executor/models/dots1.py
+1
-1
@@ -32,7 +32,7 @@ import torch
32
from torch import nn
33
from transformers import Dots1Config
34
35
-from vllm.attention import Attention
35
+from vllm.attention.layer import Attention
36
from vllm.compilation.decorators import support_torch_compile
37
from vllm.config import CacheConfig, ModelConfig, VllmConfig
38
from vllm.distributed import (
vllm/model_executor/models/ernie45_moe.py
+1
-1
@@ -32,7 +32,7 @@ import torch
32
from torch import nn
33
from transformers import PretrainedConfig
34
35
-from vllm.attention import Attention
35
+from vllm.attention.layer import Attention
36
from vllm.compilation.decorators import support_torch_compile
37
from vllm.config import CacheConfig, VllmConfig, get_current_vllm_config
38
from vllm.distributed import (
vllm/model_executor/models/ernie45_vl_moe.py
+1
-1
@@ -31,7 +31,7 @@ import torch
31
from torch import nn
32
from transformers import PretrainedConfig
33
34
-from vllm.attention import Attention
34
+from vllm.attention.layer import Attention
35
36
# from vllm.compilation.decorators import support_torch_compile
37
from vllm.config import CacheConfig, VllmConfig
vllm/model_executor/models/exaone.py
+1
-1
@@ -32,7 +32,7 @@ import torch
32
from torch import nn
33
from transformers import PretrainedConfig
34
35
-from vllm.attention import Attention
35
+from vllm.attention.layer import Attention
36
from vllm.compilation.decorators import support_torch_compile
37
from vllm.config import CacheConfig, VllmConfig
38
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/exaone4.py
+1
-1
@@ -28,7 +28,7 @@ import torch
28
from torch import nn
29
from transformers import Exaone4Config
30
31
-from vllm.attention import Attention
31
+from vllm.attention.layer import Attention
32
from vllm.compilation.decorators import support_torch_compile
33
from vllm.config import CacheConfig, VllmConfig
34
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/falcon.py
+1
-1
@@ -30,7 +30,7 @@ from torch import nn
30
from torch.nn import LayerNorm
31
from transformers import FalconConfig as HF_FalconConfig
32
33
-from vllm.attention import Attention
33
+from vllm.attention.layer import Attention
34
from vllm.compilation.decorators import support_torch_compile
35
from vllm.config import CacheConfig, VllmConfig
36
from vllm.distributed import (
vllm/model_executor/models/gemma.py
+1
-1
@@ -26,7 +26,7 @@ import torch
26
from torch import nn
27
from transformers import GemmaConfig
28
29
-from vllm.attention import Attention
29
+from vllm.attention.layer import Attention
30
from vllm.compilation.decorators import support_torch_compile
31
from vllm.config import CacheConfig, VllmConfig
32
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/gemma2.py
+1
-1
@@ -23,7 +23,7 @@ import torch
23
from torch import nn
24
from transformers import Gemma2Config
25
26
-from vllm.attention import Attention
26
+from vllm.attention.layer import Attention
27
from vllm.compilation.decorators import support_torch_compile
28
from vllm.config import CacheConfig, VllmConfig
29
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/gemma3.py
+2
-1
@@ -23,7 +23,8 @@ import torch.nn.functional as F
23
from torch import nn
24
from transformers import Gemma3TextConfig
25
26
-from vllm.attention import Attention, AttentionType
26
+from vllm.attention.backends.abstract import AttentionType
27
+from vllm.attention.layer import Attention
28
from vllm.compilation.decorators import support_torch_compile
29
from vllm.config import CacheConfig, VllmConfig
30
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/gemma3n.py
+1
-1
@@ -21,7 +21,7 @@ import torch
21
from torch import nn
22
from transformers.models.gemma3n.configuration_gemma3n import Gemma3nTextConfig
23
24
-from vllm.attention import Attention
24
+from vllm.attention.layer import Attention
25
from vllm.compilation.decorators import support_torch_compile
26
from vllm.config import CacheConfig, VllmConfig
27
from vllm.distributed import get_tensor_model_parallel_world_size
vllm/model_executor/models/glm4.py
+2
-1
@@ -29,7 +29,8 @@ import torch
29
from torch import nn
30
from transformers import Glm4Config
31
32
-from vllm.attention import Attention, AttentionType
32
+from vllm.attention.backends.abstract import AttentionType
33
+from vllm.attention.layer import Attention
34
from vllm.compilation.decorators import support_torch_compile
35
from vllm.config import CacheConfig, VllmConfig
36
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/glm4_moe.py
+1
-1
@@ -31,7 +31,7 @@ import torch
31
from torch import nn
32
from transformers.models.glm4_moe import Glm4MoeConfig
33
34
-from vllm.attention import Attention
34
+from vllm.attention.layer import Attention
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig, get_current_vllm_config
37
from vllm.distributed import (
vllm/model_executor/models/gpt2.py
+1
-1
@@ -27,7 +27,7 @@ import torch
27
from torch import nn
28
from transformers import GPT2Config
29
30
-from vllm.attention import Attention
30
+from vllm.attention.layer import Attention
31
from vllm.compilation.decorators import support_torch_compile
32
from vllm.config import CacheConfig, VllmConfig
33
from vllm.distributed.parallel_state import (
vllm/model_executor/models/gpt_bigcode.py
+1
-1
@@ -28,7 +28,7 @@ import torch
28
from torch import nn
29
from transformers import GPTBigCodeConfig
30
31
-from vllm.attention import Attention
31
+from vllm.attention.layer import Attention
32
from vllm.compilation.decorators import support_torch_compile
33
from vllm.config import CacheConfig, VllmConfig
34
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/gpt_j.py
+1
-1
@@ -26,7 +26,7 @@ import torch
26
from torch import nn
27
from transformers import GPTJConfig
28
29
-from vllm.attention import Attention
29
+from vllm.attention.layer import Attention
30
from vllm.compilation.decorators import support_torch_compile
31
from vllm.config import CacheConfig, VllmConfig
32
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/gpt_neox.py
+1
-1
@@ -26,7 +26,7 @@ import torch
26
from torch import nn
27
from transformers import GPTNeoXConfig
28
29
-from vllm.attention import Attention
29
+from vllm.attention.layer import Attention
30
from vllm.compilation.decorators import support_torch_compile
31
from vllm.config import CacheConfig, VllmConfig
32
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/gpt_oss.py
+2
-1
@@ -7,7 +7,8 @@ import torch.distributed as dist
7
from torch import nn
8
from transformers import GptOssConfig
9
10
-from vllm.attention import Attention, AttentionType
10
+from vllm.attention.backends.abstract import AttentionType
11
+from vllm.attention.layer import Attention
12
from vllm.compilation.decorators import support_torch_compile
13
from vllm.config import CacheConfig, VllmConfig
14
from vllm.distributed import (
vllm/model_executor/models/granite.py
+1
-1
@@ -31,7 +31,7 @@ import torch
31
from torch import nn
32
from transformers import GraniteConfig
33
34
-from vllm.attention import Attention
34
+from vllm.attention.layer import Attention
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig
37
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/granitemoe.py
+1
-1
@@ -31,7 +31,7 @@ from typing import Any
31
import torch
32
from torch import nn
33
34
-from vllm.attention import Attention
34
+from vllm.attention.layer import Attention
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig
37
from vllm.distributed import (
vllm/model_executor/models/grok1.py
+1
-1
@@ -31,7 +31,7 @@ import torch
31
import torch.nn.functional as F
32
from torch import nn
33
34
-from vllm.attention import Attention
34
+from vllm.attention.layer import Attention
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig
37
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/hunyuan_v1.py
+2
-1
@@ -33,7 +33,8 @@ import torch
33
from torch import nn
34
from transformers import PretrainedConfig
35
36
-from vllm.attention import Attention, AttentionType
36
+from vllm.attention.backends.abstract import AttentionType
37
+from vllm.attention.layer import Attention
38
from vllm.compilation.decorators import support_torch_compile
39
from vllm.config import CacheConfig, VllmConfig, get_current_vllm_config
40
from vllm.distributed import (
vllm/model_executor/models/internlm2.py
+1
-1
@@ -10,7 +10,7 @@ import torch
10
from torch import nn
11
from transformers import PretrainedConfig
12
13
-from vllm.attention import Attention
13
+from vllm.attention.layer import Attention
14
from vllm.compilation.decorators import support_torch_compile
15
from vllm.config import CacheConfig, VllmConfig
16
from vllm.distributed import (
vllm/model_executor/models/jais.py
+1
-1
@@ -28,7 +28,7 @@ from itertools import islice
28
import torch
29
from torch import nn
30
31
-from vllm.attention import Attention
31
+from vllm.attention.layer import Attention
32
from vllm.compilation.decorators import support_torch_compile
33
from vllm.config import CacheConfig, VllmConfig
34
from vllm.distributed import (
vllm/model_executor/models/lfm2.py
+1
-1
@@ -7,7 +7,7 @@ import torch
7
import torch.nn as nn
8
from transformers import Lfm2Config
9
10
-from vllm.attention import Attention
10
+from vllm.attention.layer import Attention
11
from vllm.compilation.decorators import support_torch_compile
12
from vllm.config import CacheConfig, ModelConfig, VllmConfig
13
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/lfm2_moe.py
+1
-1
@@ -6,7 +6,7 @@ from itertools import islice
6
import torch
7
import torch.nn as nn
8
9
-from vllm.attention import Attention
9
+from vllm.attention.layer import Attention
10
from vllm.compilation.decorators import support_torch_compile
11
from vllm.config import CacheConfig, ModelConfig, VllmConfig, get_current_vllm_config
12
from vllm.distributed import (
vllm/model_executor/models/llama.py
+2
-1
@@ -31,7 +31,8 @@ import torch
31
from torch import nn
32
from transformers import LlamaConfig
33
34
-from vllm.attention import Attention, AttentionType
34
+from vllm.attention.backends.abstract import AttentionType
35
+from vllm.attention.layer import Attention
36
from vllm.attention.layers.encoder_only_attention import EncoderOnlyAttention
37
from vllm.compilation.decorators import support_torch_compile
38
from vllm.config import CacheConfig, VllmConfig
vllm/model_executor/models/llama4.py
+1
-1
@@ -24,7 +24,7 @@ import torch
24
from torch import nn
25
from transformers import Llama4TextConfig
26
27
-from vllm.attention import Attention
27
+from vllm.attention.layer import Attention
28
from vllm.attention.layers.chunked_local_attention import ChunkedLocalAttention
29
from vllm.compilation.decorators import support_torch_compile
30
from vllm.config import CacheConfig, VllmConfig
vllm/model_executor/models/minicpm.py
+1
-1
@@ -33,7 +33,7 @@ import torch
33
from torch import nn
34
from transformers import PretrainedConfig
35
36
-from vllm.attention import Attention
36
+from vllm.attention.layer import Attention
37
from vllm.compilation.decorators import support_torch_compile
38
from vllm.config import CacheConfig, VllmConfig
39
from vllm.distributed import (
vllm/model_executor/models/minicpm3.py
+1
-1
@@ -29,7 +29,7 @@ import torch
29
from torch import nn
30
from transformers import PretrainedConfig
31
32
-from vllm.attention import Attention
32
+from vllm.attention.layer import Attention
33
from vllm.config import CacheConfig, VllmConfig
34
from vllm.distributed import get_tensor_model_parallel_world_size
35
from vllm.model_executor.layers.layernorm import RMSNorm
vllm/model_executor/models/minimax_m2.py
+1
-1
@@ -30,7 +30,7 @@ import torch
30
from torch import nn
31
from transformers import PretrainedConfig
32
33
-from vllm.attention import Attention
33
+from vllm.attention.layer import Attention
34
from vllm.compilation.decorators import support_torch_compile
35
from vllm.config import CacheConfig, ModelConfig, VllmConfig
36
from vllm.distributed import (
vllm/model_executor/models/minimax_text_01.py
+2
-1
@@ -14,7 +14,8 @@ import torch
14
from torch import nn
15
from transformers import MiniMaxConfig
16
17
-from vllm.attention import Attention, AttentionMetadata
17
+from vllm.attention.backends.abstract import AttentionMetadata
18
+from vllm.attention.layer import Attention
19
from vllm.compilation.decorators import support_torch_compile
20
from vllm.config import CacheConfig, ModelConfig, VllmConfig
21
from vllm.distributed.parallel_state import (
vllm/model_executor/models/mixtral.py
+1
-1
@@ -32,7 +32,7 @@ import torch
32
from torch import nn
33
from transformers import MixtralConfig
34
35
-from vllm.attention import Attention
35
+from vllm.attention.layer import Attention
36
from vllm.compilation.decorators import support_torch_compile
37
from vllm.config import CacheConfig, VllmConfig, get_current_vllm_config
38
from vllm.distributed import (
vllm/model_executor/models/molmo.py
+1
-2
@@ -17,8 +17,7 @@ from transformers import BatchFeature, PretrainedConfig, ProcessorMixin, TensorT
17
from transformers.image_utils import ImageInput
18
from transformers.tokenization_utils_base import TextInput
19
20
-from vllm.attention import Attention
21
-from vllm.attention.layer import MultiHeadAttention
20
+from vllm.attention.layer import Attention, MultiHeadAttention
21
from vllm.compilation.decorators import support_torch_compile
22
from vllm.config import CacheConfig, VllmConfig
23
from vllm.config.multimodal import BaseDummyOptions
vllm/model_executor/models/mpt.py
+1
-1
@@ -10,7 +10,7 @@ import torch
10
import torch.nn as nn
11
from transformers import MptConfig
12
13
-from vllm.attention import Attention
13
+from vllm.attention.layer import Attention
14
from vllm.compilation.decorators import support_torch_compile
15
from vllm.config import CacheConfig, VllmConfig
16
from vllm.distributed import (
vllm/model_executor/models/nemotron.py
+1
-1
@@ -30,7 +30,7 @@ from itertools import islice
30
import torch
31
from torch import nn
32
33
-from vllm.attention import Attention
33
+from vllm.attention.layer import Attention
34
from vllm.compilation.decorators import support_torch_compile
35
from vllm.config import CacheConfig, VllmConfig
36
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/nemotron_nas.py
+1
-1
@@ -31,7 +31,7 @@ import torch
31
from torch import nn
32
from transformers import LlamaConfig
33
34
-from vllm.attention import AttentionType
34
+from vllm.attention.backends.abstract import AttentionType
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig
37
from vllm.distributed import get_pp_group
vllm/model_executor/models/olmo.py
+1
-1
@@ -31,7 +31,7 @@ import torch
31
from torch import nn
32
from transformers import OlmoConfig
33
34
-from vllm.attention import Attention
34
+from vllm.attention.layer import Attention
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig
37
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/olmo2.py
+1
-1
@@ -32,7 +32,7 @@ import torch
32
from torch import nn
33
from transformers import Olmo2Config
34
35
-from vllm.attention import Attention
35
+from vllm.attention.layer import Attention
36
from vllm.compilation.decorators import support_torch_compile
37
from vllm.config import VllmConfig
38
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/olmoe.py
+1
-1
@@ -21,7 +21,7 @@ from itertools import islice
21
import torch
22
from torch import nn
23
24
-from vllm.attention import Attention
24
+from vllm.attention.layer import Attention
25
from vllm.compilation.decorators import support_torch_compile
26
from vllm.config import VllmConfig
27
from vllm.distributed import (
vllm/model_executor/models/openpangu.py
+2
-1
@@ -29,7 +29,8 @@ import torch
29
from torch import nn
30
from transformers import PretrainedConfig
31
32
-from vllm.attention import Attention, AttentionType
32
+from vllm.attention.backends.abstract import AttentionType
33
+from vllm.attention.layer import Attention
34
from vllm.compilation.decorators import support_torch_compile
35
from vllm.config import CacheConfig, ParallelConfig, VllmConfig
36
from vllm.distributed import (
vllm/model_executor/models/opt.py
+1
-1
@@ -27,7 +27,7 @@ import torch
27
from torch import nn
28
from transformers import OPTConfig
29
30
-from vllm.attention import Attention
30
+from vllm.attention.layer import Attention
31
from vllm.compilation.decorators import support_torch_compile
32
from vllm.config import CacheConfig, VllmConfig
33
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/orion.py
+1
-1
@@ -15,7 +15,7 @@ import torch
15
from torch import nn
16
from transformers import PretrainedConfig
17
18
-from vllm.attention import Attention
18
+from vllm.attention.layer import Attention
19
from vllm.compilation.decorators import support_torch_compile
20
from vllm.config import CacheConfig, VllmConfig
21
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/ouro.py
+2
-1
@@ -33,7 +33,8 @@ import torch
33
from torch import nn
34
from transformers import PretrainedConfig
35
36
-from vllm.attention import Attention, AttentionType
36
+from vllm.attention.backends.abstract import AttentionType
37
+from vllm.attention.layer import Attention
38
from vllm.compilation.decorators import support_torch_compile
39
from vllm.config import CacheConfig, VllmConfig
40
from vllm.distributed import get_tensor_model_parallel_world_size
vllm/model_executor/models/persimmon.py
+1
-1
@@ -30,7 +30,7 @@ import torch
30
from torch import nn
31
from transformers import PersimmonConfig
32
33
-from vllm.attention import Attention
33
+from vllm.attention.layer import Attention
34
from vllm.compilation.decorators import support_torch_compile
35
from vllm.config import CacheConfig, VllmConfig
36
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/phi.py
+1
-1
@@ -45,7 +45,7 @@ import torch
45
from torch import nn
46
from transformers import PhiConfig
47
48
-from vllm.attention import Attention
48
+from vllm.attention.layer import Attention
49
from vllm.compilation.decorators import support_torch_compile
50
from vllm.config import CacheConfig, VllmConfig
51
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/phimoe.py
+1
-1
@@ -31,7 +31,7 @@ import torch
31
from torch import nn
32
from transformers.configuration_utils import PretrainedConfig
33
34
-from vllm.attention import Attention
34
+from vllm.attention.layer import Attention
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig
37
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/qwen.py
+1
-1
@@ -16,7 +16,7 @@ import torch
16
from torch import nn
17
from transformers import PretrainedConfig
18
19
-from vllm.attention import Attention
19
+from vllm.attention.layer import Attention
20
from vllm.compilation.decorators import support_torch_compile
21
from vllm.config import CacheConfig, VllmConfig
22
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/qwen2.py
+2
-1
@@ -33,7 +33,8 @@ import torch
33
from torch import nn
34
from transformers import Qwen2Config
35
36
-from vllm.attention import Attention, AttentionType
36
+from vllm.attention.backends.abstract import AttentionType
37
+from vllm.attention.layer import Attention
38
from vllm.attention.layers.encoder_only_attention import EncoderOnlyAttention
39
from vllm.compilation.decorators import support_torch_compile
40
from vllm.config import CacheConfig, VllmConfig
vllm/model_executor/models/qwen2_moe.py
+1
-1
@@ -34,7 +34,7 @@ import torch.nn.functional as F
34
from torch import nn
35
from transformers import Qwen2MoeConfig
36
37
-from vllm.attention import Attention
37
+from vllm.attention.layer import Attention
38
from vllm.compilation.decorators import support_torch_compile
39
from vllm.config import CacheConfig, VllmConfig
40
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/qwen3.py
+2
-1
@@ -30,7 +30,8 @@ import torch
30
from torch import nn
31
from transformers import Qwen3Config
32
33
-from vllm.attention import Attention, AttentionType
33
+from vllm.attention.backends.abstract import AttentionType
34
+from vllm.attention.layer import Attention
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig
37
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/qwen3_moe.py
+1
-1
@@ -31,7 +31,7 @@ from typing import Any
31
import torch
32
from torch import nn
33
34
-from vllm.attention import Attention
34
+from vllm.attention.layer import Attention
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig, get_current_vllm_config
37
from vllm.distributed import (
vllm/model_executor/models/qwen3_next.py
+2
-1
@@ -10,7 +10,8 @@ from einops import rearrange
10
from torch import nn
11
from transformers.activations import ACT2FN
12
13
-from vllm.attention import Attention, AttentionMetadata
13
+from vllm.attention.backends.abstract import AttentionMetadata
14
+from vllm.attention.layer import Attention
15
from vllm.compilation.decorators import support_torch_compile
16
from vllm.config import (
17
CacheConfig,
vllm/model_executor/models/seed_oss.py
+2
-1
@@ -30,7 +30,8 @@ import torch
30
from torch import nn
31
from transformers import PretrainedConfig as SeedOssConfig
32
33
-from vllm.attention import Attention, AttentionType
33
+from vllm.attention.backends.abstract import AttentionType
34
+from vllm.attention.layer import Attention
35
from vllm.compilation.decorators import support_torch_compile
36
from vllm.config import CacheConfig, VllmConfig
37
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/solar.py
+1
-1
@@ -30,7 +30,7 @@ import torch
30
from torch import nn
31
from transformers import PretrainedConfig
32
33
-from vllm.attention import Attention
33
+from vllm.attention.layer import Attention
34
from vllm.compilation.decorators import support_torch_compile
35
from vllm.config import CacheConfig, VllmConfig
36
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/stablelm.py
+1
-1
@@ -29,7 +29,7 @@ import torch
29
from torch import nn
30
from transformers import StableLmConfig
31
32
-from vllm.attention import Attention
32
+from vllm.attention.layer import Attention
33
from vllm.config import CacheConfig, VllmConfig
34
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
35
from vllm.model_executor.layers.activation import SiluAndMul
vllm/model_executor/models/starcoder2.py
+1
-1
@@ -28,7 +28,7 @@ import torch
28
from torch import nn
29
from transformers import Starcoder2Config
30
31
-from vllm.attention import Attention
31
+from vllm.attention.layer import Attention
32
from vllm.compilation.decorators import support_torch_compile
33
from vllm.config import CacheConfig, VllmConfig
34
from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
vllm/model_executor/models/step3_text.py
+1
-1
@@ -9,7 +9,7 @@ from typing import Any
9
import torch
10
from torch import nn
11
12
-from vllm.attention import Attention
12
+from vllm.attention.layer import Attention
13
from vllm.compilation.decorators import support_torch_compile
14
from vllm.config import CacheConfig, ModelConfig, VllmConfig
15
from vllm.distributed import (
vllm/model_executor/models/transformers/base.py
+2
-1
@@ -27,7 +27,8 @@ from torch import nn
27
from transformers import AutoModel
28
from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS
29
30
-from vllm.attention import Attention, AttentionType
30
+from vllm.attention.backends.abstract import AttentionType
31
+from vllm.attention.layer import Attention
32
from vllm.attention.layers.encoder_only_attention import EncoderOnlyAttention
33
from vllm.config.utils import getattr_iter
34
from vllm.distributed import get_pp_group, get_tp_group
vllm/model_executor/models/whisper.py
+2
-2
@@ -16,8 +16,8 @@ from transformers import (
16
)
17
from transformers.models.whisper.modeling_whisper import sinusoids
18
19
-from vllm.attention import Attention, AttentionType
20
-from vllm.attention.layer import MultiHeadAttention
19
+from vllm.attention.backends.abstract import AttentionType
20
+from vllm.attention.layer import Attention, MultiHeadAttention
21
from vllm.attention.layers.cross_attention import CrossAttention
22
from vllm.config import CacheConfig, ModelConfig, SpeechToTextConfig, VllmConfig
23
from vllm.config.multimodal import BaseDummyOptions
vllm/platforms/cuda.py
+1
-1
@@ -335,7 +335,7 @@ class CudaPlatformBase(Platform):
335
use_sparse: bool,
336
attn_type: str | None = None,
337
) -> str:
338
- from vllm.attention import AttentionType
338
+ from vllm.attention.backends.abstract import AttentionType
339
340
if attn_type is None:
341
attn_type = AttentionType.DECODER
vllm/v1/attention/backends/cpu_attn.py
+1
-1
@@ -51,7 +51,7 @@ class CPUAttentionBackend(AttentionBackend):
51
@classmethod
52
def supports_attn_type(cls, attn_type: str) -> bool:
53
"""CPU attention supports decoder and encoder-only attention."""
54
- from vllm.attention import AttentionType
54
+ from vllm.attention.backends.abstract import AttentionType
55
56
return attn_type in (
57
AttentionType.DECODER,
vllm/v1/attention/backends/flash_attn.py
+1
-1
@@ -84,7 +84,7 @@ class FlashAttentionBackend(AttentionBackend):
84
@classmethod
85
def supports_attn_type(cls, attn_type: str) -> bool:
86
"""FlashAttention supports all attention types."""
87
- from vllm.attention import AttentionType
87
+ from vllm.attention.backends.abstract import AttentionType
88
89
return attn_type in (
90
AttentionType.DECODER,
vllm/v1/attention/backends/flex_attention.py
+1
-1
@@ -87,7 +87,7 @@ class FlexAttentionBackend(AttentionBackend):
87
@classmethod
88
def supports_attn_type(cls, attn_type: str) -> bool:
89
"""FlexAttention supports both decoder and encoder-only attention."""
90
- from vllm.attention import AttentionType
90
+ from vllm.attention.backends.abstract import AttentionType
91
92
return attn_type in (AttentionType.DECODER, AttentionType.ENCODER_ONLY)
93
vllm/v1/kv_offload/cpu.py
+1
-1
@@ -4,7 +4,7 @@ from collections.abc import Iterator
4
5
import torch
6
7
-from vllm.attention import AttentionBackend
7
+from vllm.attention.backends.abstract import AttentionBackend
8
from vllm.config import VllmConfig
9
from vllm.platforms import current_platform
10
from vllm.v1.kv_offload.abstract import LoadStoreSpec, OffloadingManager
vllm/v1/kv_offload/spec.py
+1
-1
@@ -11,7 +11,7 @@ from vllm.v1.kv_offload.abstract import LoadStoreSpec, OffloadingManager
11
from vllm.v1.kv_offload.worker.worker import OffloadingHandler
12
13
if TYPE_CHECKING:
14
- from vllm.attention import AttentionBackend
14
+ from vllm.attention.backends.abstract import AttentionBackend
15
from vllm.config import VllmConfig
16
17
logger = init_logger(__name__)
vllm/v1/kv_offload/worker/cpu_gpu.py
+1
-1
@@ -5,7 +5,7 @@ import numpy as np
5
import torch
6
7
from vllm import _custom_ops as ops
8
-from vllm.attention import AttentionBackend
8
+from vllm.attention.backends.abstract import AttentionBackend
9
from vllm.logger import init_logger
10
from vllm.utils.platform_utils import is_pin_memory_available
11
from vllm.v1.kv_offload.mediums import CPULoadStoreSpec, GPULoadStoreSpec
vllm/v1/worker/gpu_model_runner.py
+2
-1
@@ -19,12 +19,13 @@ import torch.nn as nn
19
from tqdm import tqdm
20
21
import vllm.envs as envs
22
-from vllm.attention import Attention, AttentionType
22
from vllm.attention.backends.abstract import (
23
AttentionBackend,
24
AttentionMetadata,
25
+ AttentionType,
26
MultipleOf,
27
)
28
+from vllm.attention.layer import Attention
29
from vllm.compilation.counter import compilation_counter
30
from vllm.compilation.cuda_graph import CUDAGraphWrapper
31
from vllm.compilation.monitor import set_cudagraph_capturing_enabled
vllm/v1/worker/kv_connector_model_runner_mixin.py
+1
-1
@@ -13,7 +13,7 @@ from typing import (
13
14
import torch
15
16
-from vllm.attention import AttentionBackend
16
+from vllm.attention.backends.abstract import AttentionBackend
17
from vllm.config import VllmConfig
18
from vllm.config.cache import CacheDType
19
from vllm.distributed.kv_transfer import (
vllm/v1/worker/tpu_model_runner.py
+1
-2
@@ -17,9 +17,8 @@ import torch_xla.distributed.spmd as xs
17
import torch_xla.runtime as xr
18
19
import vllm.envs as envs
20
-from vllm.attention import Attention
20
from vllm.attention.backends.abstract import AttentionType
22
-from vllm.attention.layer import MLAAttention
21
+from vllm.attention.layer import Attention, MLAAttention
22
from vllm.attention.layers.chunked_local_attention import ChunkedLocalAttention
23
from vllm.compilation.wrapper import TorchCompileWithNoGuardsWrapper
24
from vllm.config import (