diff --git a/python/sglang/multimodal_gen/runtime/loader/fsdp_load.py b/python/sglang/multimodal_gen/runtime/loader/fsdp_load.py
index ae40df3f15f6..3ca0592be320 100644
--- a/python/sglang/multimodal_gen/runtime/loader/fsdp_load.py
+++ b/python/sglang/multimodal_gen/runtime/loader/fsdp_load.py
@@ -269,7 +269,7 @@ def load_model_from_full_model_state_dict(
                 meta_sharded_param.placements,
             )
             if cpu_offload:
-                sharded_tensor = sharded_tensor.to("cpu", pin_memory=True)
+                sharded_tensor = sharded_tensor.to("cpu")
         sharded_sd[target_param_name] = nn.Parameter(sharded_tensor)
 
     model.reverse_param_names_mapping = reverse_param_names_mapping