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 11, 05:00 UTC
[XPU] allow custom workers (e.g. vllm-omni workers) to be used on XPU (#30935)
Signed-off-by: Fanli Lin <fanli.lin@intel.com>
Fanli Lin committed
Dec 19, 2025 at 02:16 UTC
058926d48c2435496839ec8e7e3ee90683ea7791
1 file changed
+4
-1
vllm/platforms/xpu.py
+4
-1
@@ -159,7 +159,10 @@ class XPUPlatform(Platform):
159
160
# check and update parallel config
161
parallel_config = vllm_config.parallel_config
162
- parallel_config.worker_cls = "vllm.v1.worker.xpu_worker.XPUWorker"
162
+ # Only override worker_cls if it's still the default "auto"
163
+ # This allows custom workers (like vllm-omni workers) to be used on XPU
164
+ if parallel_config.worker_cls == "auto":
165
+ parallel_config.worker_cls = "vllm.v1.worker.xpu_worker.XPUWorker"
166
if vllm_config.kv_transfer_config is not None:
167
vllm_config.kv_transfer_config.enable_permute_local_kv = True
168