[Attention] Update attention imports (#29540)

Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
This commit is contained in:
Matthew Bonanni
2025-11-27 11:19:09 -05:00
committed by GitHub
parent cd007a53b4
commit fc1d8be3dc
38 changed files with 63 additions and 126 deletions
+1 -5
View File
@@ -8,16 +8,14 @@ from typing import TYPE_CHECKING
import torch
import vllm.envs as envs
from vllm.attention.backends.registry import AttentionBackendEnum
from vllm.logger import init_logger
from vllm.utils.torch_utils import cuda_device_count_stateless
from .interface import DeviceCapability, Platform, PlatformEnum
if TYPE_CHECKING:
from vllm.attention.backends.registry import AttentionBackendEnum
from vllm.config import VllmConfig
else:
AttentionBackendEnum = None
logger = init_logger(__name__)
@@ -196,7 +194,6 @@ class RocmPlatform(Platform):
from importlib.util import find_spec
from vllm._aiter_ops import rocm_aiter_ops
from vllm.attention.backends.registry import AttentionBackendEnum
if rocm_aiter_ops.is_mha_enabled():
# Note: AITER FA is only supported for Qwen-VL models.
@@ -222,7 +219,6 @@ class RocmPlatform(Platform):
attn_type: str | None = None,
) -> str:
from vllm._aiter_ops import rocm_aiter_ops
from vllm.attention.backends.registry import AttentionBackendEnum
if use_sparse:
if kv_cache_dtype.startswith("fp8"):