diff --git a/代码/code/infer.py b/代码/code/infer.py
index 691f95d..09e3a44 100644
--- a/代码/code/infer.py
+++ b/代码/code/infer.py
@@ -511,11 +511,6 @@ def load_model(ckpt_path, device='cuda:0'):
 
     model.to(dev)
     model.eval()
-
-    # === torch.compile：算子融合 + 减少 kernel launch 开销 ===
-    model = torch.compile(model, mode="reduce-overhead")
-    print("[INFO] torch.compile applied (mode=reduce-overhead)")
-
     print(f"[INFO] Model ready. Device: {dev}")
 
     return model, dev