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

[platform] Allow platform specify attention backend (#11609)

Browse Source 
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Signed-off-by: Mengqing Cao <cmq0113@163.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
This commit is contained in:
wangxiyuan
2025-01-09 21:46:50 +08:00
committed by GitHub
parent 65097ca0af
commit 405eb8e396
10 changed files with 164 additions and 175 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/platforms/xpu.py
View File

@@ -21,10 +21,13 @@ class XPUPlatform(Platform):
dispatch_key: str = "XPU"
@classmethod
def get_default_attn_backend(cls, selected_backend: _Backend) -> _Backend:
def get_attn_backend_cls(cls, selected_backend: _Backend, head_size: int,
dtype: torch.dtype, kv_cache_dtype: Optional[str],
block_size: int, use_v1: bool) -> str:
if selected_backend != _Backend.IPEX:
logger.info("Cannot use %s backend on XPU.", selected_backend)
return _Backend.IPEX
logger.info("Using IPEX attention backend.")
return "vllm.attention.backends.ipex_attn.IpexAttnBackend"
@staticmethod
def get_device_capability(device_id: int = 0) -> DeviceCapability: