Benchmarks (#4912)

* finish benchmark

* fix isort

* fix setup cfg

* retab

* fix time measuring of tf graph mode

* fix tf cuda

* clean code

* better error message

examples/longform-qa/eli5_app.py

@@ -1,11 +1,11 @@
-import numpy as np
-import torch
-
 import faiss
 import nlp
+import numpy as np
+import torch
+from elasticsearch import Elasticsearch
+
 import streamlit as st
 import transformers
-from elasticsearch import Elasticsearch
 from eli5_utils import (
     embed_questions_for_retrieval,
     make_qa_s2s_model,