diff --git a/python/sglang/srt/multimodal/processors/base_processor.py b/python/sglang/srt/multimodal/processors/base_processor.py
index 5b6afb15199b..b38231efd9c7 100644
--- a/python/sglang/srt/multimodal/processors/base_processor.py
+++ b/python/sglang/srt/multimodal/processors/base_processor.py
@@ -18,7 +18,16 @@
     MultimodalInputFormat,
 )
 from sglang.srt.server_args import get_global_server_args
-from sglang.srt.utils import envs, is_npu, load_audio, load_image, load_video, logger
+from sglang.srt.utils import (
+    envs,
+    is_cpu,
+    is_npu,
+    is_xpu,
+    load_audio,
+    load_image,
+    load_video,
+    logger,
+)
 from sglang.srt.utils.cuda_ipc_transport_utils import (
     MM_FEATURE_CACHE_SIZE,
     MM_ITEM_MEMORY_POOL_RECYCLE_INTERVAL,
@@ -26,7 +35,9 @@
     MmItemMemoryPool,
 )
 
+_is_cpu = is_cpu()
 _is_npu = is_npu()
+_is_xpu = is_xpu()
 
 SGL_USE_CUDA_IPC = envs.SGLANG_USE_CUDA_IPC_TRANSPORT.get()
 
@@ -317,8 +328,10 @@ def process_mm_data(
             and isinstance(processor.image_processor, BaseImageProcessorFast)
             and not self.server_args.disable_fast_image_processor
         ):
-            if get_global_server_args().rl_on_policy_target is not None:
+            if _is_cpu or get_global_server_args().rl_on_policy_target is not None:
                 kwargs["device"] = "cpu"
+            elif _is_xpu:
+                kwargs["device"] = "xpu"
             elif not _is_npu:
                 kwargs["device"] = "cuda"
             elif processor.__class__.__name__ not in {
@@ -327,6 +340,7 @@ def process_mm_data(
             }:
                 # Note: for qwen-vl, processor has some reshape issue because of dims restriction on Ascend.
                 kwargs["device"] = "npu"
+
         result = processor.__call__(
             text=[input_text],
             padding=True,