biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Attention] Refactor CUDA attention backend selection logic (#24794)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
Matthew Bonanni
2025-11-11 06:40:44 -06:00
committed by GitHub
parent 2e78150d24
commit b30dfa03c5
61 changed files with 1338 additions and 1002 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/config/model.py
View File

@@ -45,7 +45,7 @@ if TYPE_CHECKING:
import vllm.model_executor.layers.quantization as me_quant
import vllm.model_executor.models as me_models
from vllm.attention.backends.registry import _Backend
from vllm.attention.backends.registry import AttentionBackendEnum
from vllm.config.load import LoadConfig
from vllm.config.parallel import ParallelConfig
from vllm.model_executor.layers.quantization import QuantizationMethods
@@ -53,7 +53,7 @@ if TYPE_CHECKING:
else:
PretrainedConfig = Any
_Backend = Any
AttentionBackendEnum = Any
me_quant = LazyLoader(
"model_executor", globals(), "vllm.model_executor.layers.quantization"
)
@@ -302,7 +302,7 @@ class ModelConfig:
mm_processor_cache_type: InitVar[MMCacheType | None] = None
mm_shm_cache_max_object_size_mb: InitVar[int | None] = None
mm_encoder_tp_mode: InitVar[MMEncoderTPMode | None] = None
mm_encoder_attn_backend: InitVar[_Backend | str | None] = None
mm_encoder_attn_backend: InitVar[AttentionBackendEnum | str | None] = None
interleave_mm_strings: InitVar[bool | None] = None
skip_mm_profiling: InitVar[bool | None] = None
video_pruning_rate: InitVar[float | None] = None
@@ -420,7 +420,7 @@ class ModelConfig:
mm_processor_cache_type: MMCacheType | None,
mm_shm_cache_max_object_size_mb: int | None,
mm_encoder_tp_mode: MMEncoderTPMode | None,
mm_encoder_attn_backend: _Backend | str | None,
mm_encoder_attn_backend: AttentionBackendEnum | str | None,
interleave_mm_strings: bool | None,
skip_mm_profiling: bool | None,
video_pruning_rate: float | None,