fix circular dependencies when using Lora (bigscience-workshop#284)

Author: zhangsheng377

megatron/core/tensor_parallel/layers.py

@@ -503,7 +503,7 @@ def __init__(self, input_size, output_size, *,
                  skip_bias_add=False,
                  skip_weight_param_allocation: bool=False,
                  moe=False, enable_expert_tensor_parallelism=False):
-        super(ColumnParallelLinear, self).__init__()
+        torch.nn.Module.__init__(self)
 
         # Keep input parameters
         self.input_size = input_size
@@ -691,7 +691,7 @@ def __init__(self, input_size: int, output_size: int, *,
                  keep_master_weight_for_test: bool = False,
                  skip_bias_add: bool = False,
                  moe=False, enable_expert_tensor_parallelism=False):
-        super(RowParallelLinear, self).__init__()
+        torch.nn.Module.__init__(self)
 
         # Keep input parameters
         self.input_size = input_size