Replaced torch.load for loading the pretrained vocab of TransformerXL tokenizer to pickle.load (#6935)

* Replaced torch.load for loading the pretrained vocab of TransformerXL to pickle.load

* Replaced torch.save with pickle.dump when saving the vocabulary

* updating transformer-xl

* uploaded on S3 - compatibility

* fix tests

* style

* Address review comments

Co-authored-by: Thomas Wolf <thomwolf@users.noreply.github.com>
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>

src/transformers/file_utils.py

@@ -203,6 +203,19 @@ def is_faiss_available():
     return _faiss_available
 
 
+def torch_only_method(fn):
+    def wrapper(*args, **kwargs):
+        if not _torch_available:
+            raise ImportError(
+                "You need to install pytorch to use this method or class, "
+                "or activate it with environment variables USE_TORCH=1 and USE_TF=0."
+            )
+        else:
+            return fn(*args, **kwargs)
+
+    return wrapper
+
+
 def is_sklearn_available():
     return _has_sklearn