Fixed TextGenerationPipeline on torch + GPU (#5629)

* Pytorch gpu => cpu proper device

* Memoryless XLNet warning + fixed memories during generation

* Revert "Memoryless XLNet warning + fixed memories during generation"

This reverts commit 3d3251ff

* Took the operations on the generated_sequence out of the ensure_device scope

src/transformers/pipelines.py

@@ -689,6 +689,8 @@ class TextGenerationPipeline(Pipeline):
 
             result = []
             for generated_sequence in output_sequences:
+                if self.framework == "pt" and generated_sequence is not None:
+                    generated_sequence = generated_sequence.cpu()
                 generated_sequence = generated_sequence.numpy().tolist()
                 record = {}
                 if return_tensors: