fix for auto_parallel (#10370)

Author: DrownFish19

paddlenlp/transformers/llama/fusion_ops.py

@@ -136,14 +136,8 @@ def rms_norm_fused(x_in, w, eps, use_fast_ln=False):
         fast_ln = try_import("fast_ln")
         return fast_ln.fast_rms_norm(x_in, w, eps)[0]
     else:
-        try:
-            from paddle.incubate.nn.functional import fused_rms_norm
-
-            return fused_rms_norm(x=x_in, norm_weight=w, norm_bias=None, epsilon=eps, begin_norm_axis=2)[0]
-        except ImportError:
-            fused_ln = try_import("fused_ln")
-
-            return fused_ln.fused_rms_norm(x_in, w, eps)[0]
+        fused_ln = try_import("fused_ln")
+        return fused_ln.fused_rms_norm(x_in, w, eps)[0]
 
 
 def fusion_rms_norm(hidden_states, weight, variance_epsilon, use_fast_ln=False):