fix

Author: zhangyuqin1998

paddlenlp/trainer/training_args.py

@@ -938,6 +938,10 @@ def __post_init__(self):
                 world_size % (self.tensor_parallel_degree * self.pipeline_parallel_degree) == 0
             ), f"Total world_size:{world_size} shoule be devided by tensor_parallel_degree: {self.tensor_parallel_degree} and pipeline_parallel_degree: {self.pipeline_parallel_degree}."
 
+            assert not (
+                sep_parallel_degree > 1 and context_parallel_degree > 1
+            ), f"sep parallel and context parallel cannot be used together, sep_parallel_degree:{sep_parallel_degree}, context_parallel_degree:{context_parallel_degree}."
+
             if self.sharding_parallel_degree == -1:
                 if len(self.sharding) > 0:
                     self.sharding_parallel_degree = world_size // (