[xpu]feat: support multi-lora on xpu (#20616)

Signed-off-by: yan <yan.ma@intel.com>
This commit is contained in:
Yan Ma
2025-07-08 22:07:10 +08:00
committed by GitHub
parent b942c094e3
commit a4c23314c0
5 changed files with 28 additions and 4 deletions
+11
View File
@@ -58,6 +58,10 @@ class XPUPlatform(Platform):
def get_device_name(cls, device_id: int = 0) -> str:
return torch.xpu.get_device_name(device_id)
@classmethod
def get_punica_wrapper(cls) -> str:
return "vllm.lora.punica_wrapper.punica_gpu.PunicaWrapperGPU"
@classmethod
def get_device_total_memory(cls, device_id: int = 0) -> int:
device_props = torch.xpu.get_device_properties(device_id)
@@ -78,6 +82,13 @@ class XPUPlatform(Platform):
if cache_config and cache_config.block_size is None:
cache_config.block_size = 64
# FIXME: Temporarily forcing eager mode
# remove after t.compile support stabilizes.
if (envs.VLLM_USE_V1 and vllm_config.model_config is not None
and not vllm_config.model_config.enforce_eager):
from vllm.config import CompilationLevel
vllm_config.compilation_config.level = CompilationLevel.NO_COMPILATION # noqa: E501
# Instances created using VllmConfig() typically have model_config as
# None by default. The modification involves adding a check to prevent
# potential null exceptions check and update model config.