Move GenerationMixin to separate file (#5254)

* separate_generation_code

* isort

* renamed

* rename_files

* move_shapelit

src/transformers/__init__.py

@@ -169,7 +169,8 @@ if is_sklearn_available():
 
 # Modeling
 if is_torch_available():
-    from .modeling_utils import PreTrainedModel, prune_layer, Conv1D, top_k_top_p_filtering, apply_chunking_to_forward
+    from .generation_utils import top_k_top_p_filtering
+    from .modeling_utils import PreTrainedModel, prune_layer, Conv1D, apply_chunking_to_forward
     from .modeling_auto import (
         AutoModel,
         AutoModelForPreTraining,
@@ -406,9 +407,9 @@ if is_torch_available():
 
 # TensorFlow
 if is_tf_available():
+    from .generation_tf_utils import tf_top_k_top_p_filtering
     from .modeling_tf_utils import (
         shape_list,
-        tf_top_k_top_p_filtering,
         TFPreTrainedModel,
         TFSequenceSummary,
         TFSharedEmbeddings,