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 6, 02:00 UTC
[XPU] Support Intel XPU hardware information collection in usage stats (#37964)
Signed-off-by: sihao.li <sihao.li@intel.com>
sihao_li committed
Mar 25, 2026 at 01:29 UTC
a5416bc52e5d516008388c833a8d6270e92a894c
1 file changed
+6
vllm/usage/usage_lib.py
+6
@@ -136,6 +136,7 @@ class UsageMessage:
136
self.total_memory: int | None = None
137
self.architecture: str | None = None
138
self.platform: str | None = None
139
+ self.xpu_runtime: str | None = None
140
self.cuda_runtime: str | None = None
141
self.gpu_count: int | None = None
142
self.gpu_type: str | None = None
@@ -201,6 +202,11 @@ class UsageMessage:
202
)
203
if current_platform.is_cuda():
204
self.cuda_runtime = torch.version.cuda
205
+ if current_platform.is_xpu():
206
+ self.xpu_runtime = torch.version.xpu
207
+ self.gpu_count = torch.xpu.device_count()
208
+ self.gpu_type = torch.xpu.get_device_name(0)
209
+ self.gpu_memory_per_device = torch.xpu.get_device_properties(0).total_memory
210
if current_platform.is_tpu(): # noqa: SIM102
211
if not self._report_tpu_inference_usage():
212
logger.exception("Failed to collect TPU information")