del wait in sharding for npu (#34637)

python/paddle/distributed/fleet/meta_optimizers/sharding_optimizer.py

@@ -379,7 +379,8 @@ class ShardingOptimizer(MetaOptimizerBase):
                   'w') as f:
             f.writelines(str(main_block.program))
 
-        # GPU and NPU need to wait server ready
+        # GPU need to wait server ready, GPU and NPU is Layered connection
+        if not core.is_compiled_with_npu():
             self._wait()
         return optimize_ops, params_grads