[NFC] polish colossalai/utils/multi_tensor_apply/multi_tensor_apply.py code style (#1559)

colossalai/pipeline/rpc/PipelineBase.py

@@ -778,4 +778,4 @@ class OneFOneBPipelineEngine(PipelineEngineBase):
                  criterion: Callable = None,
                  checkpoint: bool = False) -> None:
         use_1F1B = True
-        super().__init__(module_partitions, stage_num, num_microbatches, device, use_1F1B, chunk, criterion, checkpoint)
+        super().__init__(module_partitions, stage_num, num_microbatches, device, use_1F1B, chunk, criterion, checkpoint)

colossalai/utils/multi_tensor_apply/multi_tensor_apply.py

@@ -26,13 +26,9 @@ class MultiTensorApply(object):
             raise RuntimeError(
                 "Attempted to call MultiTensorApply method, but MultiTensorApply "
                 "is not available, possibly because Apex was installed without "
-                "--cpp_ext --cuda_ext.  Original import error message:",
-                MultiTensorApply.import_err)
+                "--cpp_ext --cuda_ext.  Original import error message:", MultiTensorApply.import_err)
 
     def __call__(self, op, noop_flag_buffer, tensor_lists, *args):
         self.check_avail()
 
-        return op(self.chunk_size,
-                  noop_flag_buffer,
-                  tensor_lists,
-                  *args)
+        return op(self.chunk_size, noop_flag_buffer, tensor_lists, *args)