From 73f034deca5d58cbe571ef6f08e6d3a1d7781eb0 Mon Sep 17 00:00:00 2001
From: Ryan Sepassi <rsepassi@google.com>
Date: Thu, 8 Feb 2018 15:32:46 -0800
Subject: [PATCH 01/14] Update License year

PiperOrigin-RevId: 185063168
---
 tensor2tensor/__init__.py                     |   2 +-
 tensor2tensor/bin/__init__.py                 |   2 +-
 tensor2tensor/bin/make_tf_configs.py          |   2 +-
 tensor2tensor/bin/t2t_avg_all.py              |   2 +-
 tensor2tensor/bin/t2t_bleu.py                 |   2 +-
 tensor2tensor/bin/t2t_datagen.py              |   2 +-
 tensor2tensor/bin/t2t_decoder.py              |   2 +-
 tensor2tensor/bin/t2t_trainer.py              |   4 +-
 tensor2tensor/bin/t2t_trainer_test.py         |   2 +-
 tensor2tensor/bin/t2t_translate_all.py        |   2 +-
 tensor2tensor/data_generators/__init__.py     |   2 +-
 tensor2tensor/data_generators/algorithmic.py  |   2 +-
 .../data_generators/algorithmic_math.py       |   2 +-
 .../data_generators/algorithmic_math_test.py  |   2 +-
 .../data_generators/algorithmic_test.py       |   2 +-
 tensor2tensor/data_generators/all_problems.py |   2 +-
 .../data_generators/all_problems_test.py      |   2 +-
 tensor2tensor/data_generators/audio.py        |   2 +-
 tensor2tensor/data_generators/audio_test.py   |   2 +-
 tensor2tensor/data_generators/celeba.py       |   2 +-
 tensor2tensor/data_generators/cifar.py        |   2 +-
 tensor2tensor/data_generators/cipher.py       |   2 +-
 .../data_generators/cnn_dailymail.py          |   2 +-
 tensor2tensor/data_generators/desc2code.py    |   2 +-
 .../data_generators/desc2code_test.py         |   2 +-
 tensor2tensor/data_generators/dna_encoder.py  |   2 +-
 .../data_generators/dna_encoder_test.py       |   2 +-
 tensor2tensor/data_generators/fsns.py         |   2 +-
 .../data_generators/gene_expression.py        |   2 +-
 .../data_generators/gene_expression_test.py   |   2 +-
 .../data_generators/generator_utils.py        |   2 +-
 .../data_generators/generator_utils_test.py   |   2 +-
 tensor2tensor/data_generators/gym.py          |   2 +-
 tensor2tensor/data_generators/ice_parsing.py  |   2 +-
 tensor2tensor/data_generators/image_utils.py  |   2 +-
 .../data_generators/image_utils_test.py       |   2 +-
 tensor2tensor/data_generators/imagenet.py     |   2 +-
 tensor2tensor/data_generators/imdb.py         |   2 +-
 tensor2tensor/data_generators/inspect.py      |   2 +-
 tensor2tensor/data_generators/librispeech.py  |   2 +-
 tensor2tensor/data_generators/lm1b.py         |   2 +-
 tensor2tensor/data_generators/mnist.py        |   2 +-
 tensor2tensor/data_generators/mscoco.py       |   2 +-
 tensor2tensor/data_generators/multinli.py     |   2 +-
 tensor2tensor/data_generators/ocr.py          |   2 +-
 tensor2tensor/data_generators/problem.py      |   2 +-
 .../data_generators/problem_hparams.py        |   2 +-
 tensor2tensor/data_generators/ptb.py          |   2 +-
 tensor2tensor/data_generators/snli.py         |   2 +-
 .../data_generators/speech_recognition.py     |   2 +-
 tensor2tensor/data_generators/text_encoder.py |   2 +-
 .../text_encoder_build_subword.py             |   2 +-
 .../data_generators/text_encoder_test.py      |   2 +-
 tensor2tensor/data_generators/tokenizer.py    |   2 +-
 .../data_generators/tokenizer_test.py         |   2 +-
 tensor2tensor/data_generators/translate.py    |   2 +-
 .../data_generators/translate_encs.py         |   2 +-
 .../data_generators/translate_ende.py         |   2 +-
 .../data_generators/translate_enfr.py         |   2 +-
 .../data_generators/translate_enmk.py         |   2 +-
 .../data_generators/translate_enzh.py         |   2 +-
 .../data_generators/translate_test.py         |   2 +-
 tensor2tensor/data_generators/twentybn.py     |   2 +-
 tensor2tensor/data_generators/wiki.py         |   2 +-
 tensor2tensor/data_generators/wsj_parsing.py  |   2 +-
 tensor2tensor/insights/__init__.py            |   2 +-
 tensor2tensor/insights/graph.py               |   2 +-
 .../attention-visualization.html              |   2 +-
 .../attention-visualization.js                |   2 +-
 .../insights/polymer/common-types.js          |   2 +-
 .../polymer/explore_view/explore-view.html    |   2 +-
 .../polymer/explore_view/explore-view.js      |   2 +-
 .../graph-visualization.html                  |   2 +-
 .../graph-visualization.js                    |   2 +-
 tensor2tensor/insights/polymer/index.html     |   2 +-
 .../polymer/insights_app/insights-app.html    |   2 +-
 .../polymer/insights_app/insights-app.js      |   2 +-
 .../language-selector-content.html            |   2 +-
 .../language-selector-content.js              |   2 +-
 .../language_selector/language-selector.html  |   2 +-
 .../language_selector/language-selector.js    |   2 +-
 .../processing-visualization.html             |   2 +-
 .../processing-visualization.js               |   2 +-
 .../polymer/query_card/query-card.html        |   2 +-
 .../insights/polymer/query_card/query-card.js |   2 +-
 .../insights/polymer/tensor2tensor.html       |   2 +-
 .../translation-result.html                   |   2 +-
 .../translation_result/translation-result.js  |   2 +-
 tensor2tensor/insights/query_processor.py     |   2 +-
 tensor2tensor/insights/server.py              |   2 +-
 tensor2tensor/insights/transformer_model.py   |   2 +-
 tensor2tensor/layers/__init__.py              |   2 +-
 tensor2tensor/layers/common_attention.py      |   2 +-
 tensor2tensor/layers/common_attention_test.py |   2 +-
 tensor2tensor/layers/common_hparams.py        |   2 +-
 .../layers/common_image_attention.py          |   2 +-
 tensor2tensor/layers/common_layers.py         |   2 +-
 tensor2tensor/layers/common_layers_test.py    |   2 +-
 tensor2tensor/layers/modalities.py            |   2 +-
 tensor2tensor/layers/modalities_test.py       |   2 +-
 tensor2tensor/layers/rev_block.py             |   2 +-
 tensor2tensor/layers/rev_block_test.py        |   2 +-
 tensor2tensor/models/__init__.py              |   3 +-
 tensor2tensor/models/aligned.py               |   2 +-
 tensor2tensor/models/attention_lm.py          |   2 +-
 tensor2tensor/models/attention_lm_moe.py      |   2 +-
 tensor2tensor/models/basic.py                 |   2 +-
 tensor2tensor/models/bytenet.py               |   2 +-
 tensor2tensor/models/bytenet_test.py          |   2 +-
 tensor2tensor/models/cycle_gan.py             |   2 +-
 tensor2tensor/models/gene_expression.py       |   2 +-
 tensor2tensor/models/gene_expression_test.py  |   2 +-
 tensor2tensor/models/lstm.py                  |   2 +-
 tensor2tensor/models/lstm_test.py             |   2 +-
 tensor2tensor/models/multimodel.py            |   2 +-
 tensor2tensor/models/multimodel_test.py       |   2 +-
 tensor2tensor/models/neural_gpu.py            |   2 +-
 tensor2tensor/models/neural_gpu_test.py       |   2 +-
 tensor2tensor/models/resnet.py                |   2 +-
 tensor2tensor/models/resnet_test.py           |   2 +-
 tensor2tensor/models/revnet.py                |   2 +-
 tensor2tensor/models/revnet_test.py           |   2 +-
 tensor2tensor/models/rl.py                    | 127 ------------------
 tensor2tensor/models/shake_shake.py           |   2 +-
 tensor2tensor/models/slicenet.py              |   2 +-
 tensor2tensor/models/slicenet_test.py         |   2 +-
 tensor2tensor/models/super_lm.py              |   2 +-
 tensor2tensor/models/transformer.py           |   2 +-
 tensor2tensor/models/transformer_moe.py       |   2 +-
 tensor2tensor/models/transformer_revnet.py    |   2 +-
 .../models/transformer_revnet_test.py         |   2 +-
 tensor2tensor/models/transformer_sketch.py    |   2 +-
 tensor2tensor/models/transformer_test.py      |   2 +-
 tensor2tensor/models/transformer_vae.py       |   2 +-
 tensor2tensor/models/vanilla_gan.py           |   2 +-
 tensor2tensor/models/xception.py              |   2 +-
 tensor2tensor/models/xception_test.py         |   2 +-
 tensor2tensor/notebooks/hello_t2t.ipynb       |   2 +-
 tensor2tensor/problems.py                     |   2 +-
 tensor2tensor/problems_test.py                |   2 +-
 tensor2tensor/rl/README.md                    |   2 +-
 tensor2tensor/rl/__init__.py                  |   2 +-
 tensor2tensor/rl/collect.py                   |  25 ++--
 tensor2tensor/rl/envs/__init__.py             |   2 +-
 tensor2tensor/rl/envs/batch_env.py            |   2 +-
 tensor2tensor/rl/envs/in_graph_batch_env.py   |   2 +-
 tensor2tensor/rl/envs/utils.py                |   2 +-
 tensor2tensor/rl/networks.py                  |  69 ++++++++++
 tensor2tensor/rl/ppo.py                       |  45 ++-----
 tensor2tensor/rl/rl_trainer_lib.py            |  65 ++++++---
 tensor2tensor/rl/rl_trainer_lib_test.py       |  17 +--
 tensor2tensor/rl/t2t_rl_trainer.py            |  20 +--
 tensor2tensor/serving/__init__.py             |   2 +-
 tensor2tensor/serving/export.py               |   2 +-
 tensor2tensor/serving/query.py                |   2 +-
 .../test_data/example_usr_dir/__init__.py     |   2 +-
 .../test_data/example_usr_dir/my_submodule.py |   2 +-
 tensor2tensor/utils/__init__.py               |   2 +-
 tensor2tensor/utils/avg_checkpoints.py        |   2 +-
 tensor2tensor/utils/beam_search.py            |   2 +-
 tensor2tensor/utils/beam_search_test.py       |   2 +-
 tensor2tensor/utils/bleu_hook.py              |   2 +-
 tensor2tensor/utils/bleu_hook_test.py         |   2 +-
 .../utils/checkpoint_compatibility_test.py    |   2 +-
 tensor2tensor/utils/cloud_mlengine.py         |   2 +-
 tensor2tensor/utils/cloud_tpu.py              |   2 +-
 tensor2tensor/utils/data_reader.py            |   2 +-
 tensor2tensor/utils/data_reader_test.py       |   2 +-
 tensor2tensor/utils/decoding.py               |   2 +-
 tensor2tensor/utils/devices.py                |   2 +-
 tensor2tensor/utils/diet.py                   |   2 +-
 tensor2tensor/utils/diet_test.py              |   2 +-
 tensor2tensor/utils/expert_utils.py           |   2 +-
 tensor2tensor/utils/expert_utils_test.py      |   2 +-
 tensor2tensor/utils/flags.py                  |   2 +-
 tensor2tensor/utils/get_rouge.py              |   2 +-
 tensor2tensor/utils/metrics.py                |   2 +-
 tensor2tensor/utils/metrics_hook.py           |   2 +-
 tensor2tensor/utils/metrics_hook_test.py      |   2 +-
 tensor2tensor/utils/metrics_test.py           |   2 +-
 tensor2tensor/utils/modality.py               |   2 +-
 tensor2tensor/utils/optimize.py               |   2 +-
 tensor2tensor/utils/registry.py               |   2 +-
 tensor2tensor/utils/registry_test.py          |   2 +-
 tensor2tensor/utils/rouge.py                  |  17 +--
 tensor2tensor/utils/rouge_test.py             |   2 +-
 tensor2tensor/utils/t2t_model.py              |   2 +-
 tensor2tensor/utils/trainer_lib.py            |   2 +-
 tensor2tensor/utils/trainer_lib_test.py       |   2 +-
 tensor2tensor/utils/usr_dir.py                |   2 +-
 tensor2tensor/utils/yellowfin.py              |   2 +-
 tensor2tensor/utils/yellowfin_test.py         |   2 +-
 tensor2tensor/visualization/__init__.py       |   2 +-
 tensor2tensor/visualization/attention.py      |   2 +-
 tensor2tensor/visualization/visualization.py  |   2 +-
 .../visualization/visualization_test.py       |   2 +-
 196 files changed, 332 insertions(+), 432 deletions(-)
 delete mode 100644 tensor2tensor/models/rl.py
 create mode 100644 tensor2tensor/rl/networks.py

diff --git a/tensor2tensor/__init__.py b/tensor2tensor/__init__.py
index 3f714ce1f..4bd418a74 100644
--- a/tensor2tensor/__init__.py
+++ b/tensor2tensor/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/bin/__init__.py b/tensor2tensor/bin/__init__.py
index 3f714ce1f..4bd418a74 100644
--- a/tensor2tensor/bin/__init__.py
+++ b/tensor2tensor/bin/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/bin/make_tf_configs.py b/tensor2tensor/bin/make_tf_configs.py
index ce0d638d6..85ba874f3 100644
--- a/tensor2tensor/bin/make_tf_configs.py
+++ b/tensor2tensor/bin/make_tf_configs.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/bin/t2t_avg_all.py b/tensor2tensor/bin/t2t_avg_all.py
index 6e2a3088e..66bac86fb 100644
--- a/tensor2tensor/bin/t2t_avg_all.py
+++ b/tensor2tensor/bin/t2t_avg_all.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/bin/t2t_bleu.py b/tensor2tensor/bin/t2t_bleu.py
index 5db83965d..40d7ec1cb 100644
--- a/tensor2tensor/bin/t2t_bleu.py
+++ b/tensor2tensor/bin/t2t_bleu.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/bin/t2t_datagen.py b/tensor2tensor/bin/t2t_datagen.py
index 8c91b7ee0..d00c54f59 100644
--- a/tensor2tensor/bin/t2t_datagen.py
+++ b/tensor2tensor/bin/t2t_datagen.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/bin/t2t_decoder.py b/tensor2tensor/bin/t2t_decoder.py
index e7ae6b74a..dd2206752 100644
--- a/tensor2tensor/bin/t2t_decoder.py
+++ b/tensor2tensor/bin/t2t_decoder.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/bin/t2t_trainer.py b/tensor2tensor/bin/t2t_trainer.py
index 018e7c270..82a74d3e5 100644
--- a/tensor2tensor/bin/t2t_trainer.py
+++ b/tensor2tensor/bin/t2t_trainer.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -274,7 +274,7 @@ def save_metadata(hparams):
   # Save hparams as hparams.json
   hparams_fname = os.path.join(output_dir, "hparams.json")
   with tf.gfile.Open(hparams_fname, "w") as f:
-    f.write(hparams.to_json(indent=0, sort_keys=True))
+    f.write(hparams.to_json())
 
 
 def execute_schedule(exp):
diff --git a/tensor2tensor/bin/t2t_trainer_test.py b/tensor2tensor/bin/t2t_trainer_test.py
index b1f38cec5..1f4569e60 100644
--- a/tensor2tensor/bin/t2t_trainer_test.py
+++ b/tensor2tensor/bin/t2t_trainer_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/bin/t2t_translate_all.py b/tensor2tensor/bin/t2t_translate_all.py
index 40046e883..249068dad 100644
--- a/tensor2tensor/bin/t2t_translate_all.py
+++ b/tensor2tensor/bin/t2t_translate_all.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/__init__.py b/tensor2tensor/data_generators/__init__.py
index 3f714ce1f..4bd418a74 100644
--- a/tensor2tensor/data_generators/__init__.py
+++ b/tensor2tensor/data_generators/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/algorithmic.py b/tensor2tensor/data_generators/algorithmic.py
index 3c1d5468f..ddb456cb1 100644
--- a/tensor2tensor/data_generators/algorithmic.py
+++ b/tensor2tensor/data_generators/algorithmic.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/algorithmic_math.py b/tensor2tensor/data_generators/algorithmic_math.py
index 93c8ad117..c3a028b12 100644
--- a/tensor2tensor/data_generators/algorithmic_math.py
+++ b/tensor2tensor/data_generators/algorithmic_math.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/algorithmic_math_test.py b/tensor2tensor/data_generators/algorithmic_math_test.py
index c7fdfa156..ce5310b8e 100644
--- a/tensor2tensor/data_generators/algorithmic_math_test.py
+++ b/tensor2tensor/data_generators/algorithmic_math_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/algorithmic_test.py b/tensor2tensor/data_generators/algorithmic_test.py
index bcd61216e..5d954fd14 100644
--- a/tensor2tensor/data_generators/algorithmic_test.py
+++ b/tensor2tensor/data_generators/algorithmic_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/all_problems.py b/tensor2tensor/data_generators/all_problems.py
index 80734e06d..cf730bc69 100644
--- a/tensor2tensor/data_generators/all_problems.py
+++ b/tensor2tensor/data_generators/all_problems.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/all_problems_test.py b/tensor2tensor/data_generators/all_problems_test.py
index de84a0bf3..2cad7ee44 100644
--- a/tensor2tensor/data_generators/all_problems_test.py
+++ b/tensor2tensor/data_generators/all_problems_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/audio.py b/tensor2tensor/data_generators/audio.py
index d0747a88c..1f92ee939 100644
--- a/tensor2tensor/data_generators/audio.py
+++ b/tensor2tensor/data_generators/audio.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/audio_test.py b/tensor2tensor/data_generators/audio_test.py
index 57e4e1ccc..81a55686d 100644
--- a/tensor2tensor/data_generators/audio_test.py
+++ b/tensor2tensor/data_generators/audio_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/celeba.py b/tensor2tensor/data_generators/celeba.py
index 5b85c2590..cd27e5715 100644
--- a/tensor2tensor/data_generators/celeba.py
+++ b/tensor2tensor/data_generators/celeba.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/cifar.py b/tensor2tensor/data_generators/cifar.py
index 0c03ebc30..881e52a7e 100644
--- a/tensor2tensor/data_generators/cifar.py
+++ b/tensor2tensor/data_generators/cifar.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/cipher.py b/tensor2tensor/data_generators/cipher.py
index 977174880..d7147547b 100644
--- a/tensor2tensor/data_generators/cipher.py
+++ b/tensor2tensor/data_generators/cipher.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/cnn_dailymail.py b/tensor2tensor/data_generators/cnn_dailymail.py
index 636f04a97..2771f4b0c 100644
--- a/tensor2tensor/data_generators/cnn_dailymail.py
+++ b/tensor2tensor/data_generators/cnn_dailymail.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/desc2code.py b/tensor2tensor/data_generators/desc2code.py
index 174bd8107..845cecd70 100644
--- a/tensor2tensor/data_generators/desc2code.py
+++ b/tensor2tensor/data_generators/desc2code.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/desc2code_test.py b/tensor2tensor/data_generators/desc2code_test.py
index 79992296b..61f059789 100644
--- a/tensor2tensor/data_generators/desc2code_test.py
+++ b/tensor2tensor/data_generators/desc2code_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/dna_encoder.py b/tensor2tensor/data_generators/dna_encoder.py
index f084c9cd2..9db20de42 100644
--- a/tensor2tensor/data_generators/dna_encoder.py
+++ b/tensor2tensor/data_generators/dna_encoder.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/dna_encoder_test.py b/tensor2tensor/data_generators/dna_encoder_test.py
index a84f06442..a32e3f2bf 100644
--- a/tensor2tensor/data_generators/dna_encoder_test.py
+++ b/tensor2tensor/data_generators/dna_encoder_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/fsns.py b/tensor2tensor/data_generators/fsns.py
index 938931c4c..8a9cfced4 100644
--- a/tensor2tensor/data_generators/fsns.py
+++ b/tensor2tensor/data_generators/fsns.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/gene_expression.py b/tensor2tensor/data_generators/gene_expression.py
index 477e04017..29d7819f2 100644
--- a/tensor2tensor/data_generators/gene_expression.py
+++ b/tensor2tensor/data_generators/gene_expression.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/gene_expression_test.py b/tensor2tensor/data_generators/gene_expression_test.py
index 797170070..8a7ccd55a 100644
--- a/tensor2tensor/data_generators/gene_expression_test.py
+++ b/tensor2tensor/data_generators/gene_expression_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/generator_utils.py b/tensor2tensor/data_generators/generator_utils.py
index b73b63d74..d25e5c621 100644
--- a/tensor2tensor/data_generators/generator_utils.py
+++ b/tensor2tensor/data_generators/generator_utils.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/generator_utils_test.py b/tensor2tensor/data_generators/generator_utils_test.py
index f058428fb..0460ef2e7 100644
--- a/tensor2tensor/data_generators/generator_utils_test.py
+++ b/tensor2tensor/data_generators/generator_utils_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/gym.py b/tensor2tensor/data_generators/gym.py
index 27be9a71b..d1a1cdc37 100644
--- a/tensor2tensor/data_generators/gym.py
+++ b/tensor2tensor/data_generators/gym.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/ice_parsing.py b/tensor2tensor/data_generators/ice_parsing.py
index fdb53430a..eea2616b5 100644
--- a/tensor2tensor/data_generators/ice_parsing.py
+++ b/tensor2tensor/data_generators/ice_parsing.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/image_utils.py b/tensor2tensor/data_generators/image_utils.py
index 78fdde262..3eb571865 100644
--- a/tensor2tensor/data_generators/image_utils.py
+++ b/tensor2tensor/data_generators/image_utils.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/image_utils_test.py b/tensor2tensor/data_generators/image_utils_test.py
index 5a7afc3b7..875863338 100644
--- a/tensor2tensor/data_generators/image_utils_test.py
+++ b/tensor2tensor/data_generators/image_utils_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/imagenet.py b/tensor2tensor/data_generators/imagenet.py
index 1ca2ec3e2..8fb6679ff 100644
--- a/tensor2tensor/data_generators/imagenet.py
+++ b/tensor2tensor/data_generators/imagenet.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/imdb.py b/tensor2tensor/data_generators/imdb.py
index 22e674098..61587da6e 100644
--- a/tensor2tensor/data_generators/imdb.py
+++ b/tensor2tensor/data_generators/imdb.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/inspect.py b/tensor2tensor/data_generators/inspect.py
index 0293ca9c4..0b113bb98 100644
--- a/tensor2tensor/data_generators/inspect.py
+++ b/tensor2tensor/data_generators/inspect.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/librispeech.py b/tensor2tensor/data_generators/librispeech.py
index 0c59824c1..d14cc8868 100644
--- a/tensor2tensor/data_generators/librispeech.py
+++ b/tensor2tensor/data_generators/librispeech.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/lm1b.py b/tensor2tensor/data_generators/lm1b.py
index cd0eb8e3c..b08ec9e3f 100644
--- a/tensor2tensor/data_generators/lm1b.py
+++ b/tensor2tensor/data_generators/lm1b.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/mnist.py b/tensor2tensor/data_generators/mnist.py
index 9a6792b62..5fefbd476 100644
--- a/tensor2tensor/data_generators/mnist.py
+++ b/tensor2tensor/data_generators/mnist.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/mscoco.py b/tensor2tensor/data_generators/mscoco.py
index 8dc1e1bba..2d58b97b5 100644
--- a/tensor2tensor/data_generators/mscoco.py
+++ b/tensor2tensor/data_generators/mscoco.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/multinli.py b/tensor2tensor/data_generators/multinli.py
index acd3a2c58..e70252005 100644
--- a/tensor2tensor/data_generators/multinli.py
+++ b/tensor2tensor/data_generators/multinli.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/ocr.py b/tensor2tensor/data_generators/ocr.py
index 8e92165bd..924483d5a 100644
--- a/tensor2tensor/data_generators/ocr.py
+++ b/tensor2tensor/data_generators/ocr.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/problem.py b/tensor2tensor/data_generators/problem.py
index a2c330c2d..c5b253c3c 100644
--- a/tensor2tensor/data_generators/problem.py
+++ b/tensor2tensor/data_generators/problem.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/problem_hparams.py b/tensor2tensor/data_generators/problem_hparams.py
index 576a27a79..c69a78eb9 100644
--- a/tensor2tensor/data_generators/problem_hparams.py
+++ b/tensor2tensor/data_generators/problem_hparams.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/ptb.py b/tensor2tensor/data_generators/ptb.py
index 31bc83c0a..19f20c90a 100644
--- a/tensor2tensor/data_generators/ptb.py
+++ b/tensor2tensor/data_generators/ptb.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/snli.py b/tensor2tensor/data_generators/snli.py
index cd4ff723d..8533c72c2 100644
--- a/tensor2tensor/data_generators/snli.py
+++ b/tensor2tensor/data_generators/snli.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/speech_recognition.py b/tensor2tensor/data_generators/speech_recognition.py
index e17e4de85..6d37565b8 100644
--- a/tensor2tensor/data_generators/speech_recognition.py
+++ b/tensor2tensor/data_generators/speech_recognition.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/text_encoder.py b/tensor2tensor/data_generators/text_encoder.py
index d43236945..e974a523d 100644
--- a/tensor2tensor/data_generators/text_encoder.py
+++ b/tensor2tensor/data_generators/text_encoder.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/text_encoder_build_subword.py b/tensor2tensor/data_generators/text_encoder_build_subword.py
index 0c366c896..1bd904f62 100644
--- a/tensor2tensor/data_generators/text_encoder_build_subword.py
+++ b/tensor2tensor/data_generators/text_encoder_build_subword.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/text_encoder_test.py b/tensor2tensor/data_generators/text_encoder_test.py
index 273810684..ff6450dc8 100644
--- a/tensor2tensor/data_generators/text_encoder_test.py
+++ b/tensor2tensor/data_generators/text_encoder_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/tokenizer.py b/tensor2tensor/data_generators/tokenizer.py
index 5cfd7c42e..c023627c2 100644
--- a/tensor2tensor/data_generators/tokenizer.py
+++ b/tensor2tensor/data_generators/tokenizer.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/tokenizer_test.py b/tensor2tensor/data_generators/tokenizer_test.py
index 065a32e91..ac4cd0cca 100644
--- a/tensor2tensor/data_generators/tokenizer_test.py
+++ b/tensor2tensor/data_generators/tokenizer_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/translate.py b/tensor2tensor/data_generators/translate.py
index 95f5844c1..20c5081e7 100644
--- a/tensor2tensor/data_generators/translate.py
+++ b/tensor2tensor/data_generators/translate.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/translate_encs.py b/tensor2tensor/data_generators/translate_encs.py
index ad0fe828d..b5f72fc9a 100644
--- a/tensor2tensor/data_generators/translate_encs.py
+++ b/tensor2tensor/data_generators/translate_encs.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/translate_ende.py b/tensor2tensor/data_generators/translate_ende.py
index 2124be32a..f9be14681 100644
--- a/tensor2tensor/data_generators/translate_ende.py
+++ b/tensor2tensor/data_generators/translate_ende.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/translate_enfr.py b/tensor2tensor/data_generators/translate_enfr.py
index 921834000..18e99d069 100644
--- a/tensor2tensor/data_generators/translate_enfr.py
+++ b/tensor2tensor/data_generators/translate_enfr.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/translate_enmk.py b/tensor2tensor/data_generators/translate_enmk.py
index aa1bac8b1..70630c444 100644
--- a/tensor2tensor/data_generators/translate_enmk.py
+++ b/tensor2tensor/data_generators/translate_enmk.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/translate_enzh.py b/tensor2tensor/data_generators/translate_enzh.py
index 0a645b3bb..27f621c85 100644
--- a/tensor2tensor/data_generators/translate_enzh.py
+++ b/tensor2tensor/data_generators/translate_enzh.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/translate_test.py b/tensor2tensor/data_generators/translate_test.py
index e357e11fc..f28b47818 100644
--- a/tensor2tensor/data_generators/translate_test.py
+++ b/tensor2tensor/data_generators/translate_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/twentybn.py b/tensor2tensor/data_generators/twentybn.py
index c70313767..7d83ce55e 100644
--- a/tensor2tensor/data_generators/twentybn.py
+++ b/tensor2tensor/data_generators/twentybn.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/wiki.py b/tensor2tensor/data_generators/wiki.py
index 828ef2246..4f6dd57bb 100644
--- a/tensor2tensor/data_generators/wiki.py
+++ b/tensor2tensor/data_generators/wiki.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/data_generators/wsj_parsing.py b/tensor2tensor/data_generators/wsj_parsing.py
index 4b1dbdd80..bef82eb1b 100644
--- a/tensor2tensor/data_generators/wsj_parsing.py
+++ b/tensor2tensor/data_generators/wsj_parsing.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/__init__.py b/tensor2tensor/insights/__init__.py
index 3f714ce1f..4bd418a74 100644
--- a/tensor2tensor/insights/__init__.py
+++ b/tensor2tensor/insights/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/graph.py b/tensor2tensor/insights/graph.py
index a733998b8..17e18ea3c 100644
--- a/tensor2tensor/insights/graph.py
+++ b/tensor2tensor/insights/graph.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/attention_visualization/attention-visualization.html b/tensor2tensor/insights/polymer/attention_visualization/attention-visualization.html
index 4ec11ace8..02db7fe09 100644
--- a/tensor2tensor/insights/polymer/attention_visualization/attention-visualization.html
+++ b/tensor2tensor/insights/polymer/attention_visualization/attention-visualization.html
@@ -1,6 +1,6 @@
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/attention_visualization/attention-visualization.js b/tensor2tensor/insights/polymer/attention_visualization/attention-visualization.js
index e738c2629..4e9c6e862 100644
--- a/tensor2tensor/insights/polymer/attention_visualization/attention-visualization.js
+++ b/tensor2tensor/insights/polymer/attention_visualization/attention-visualization.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/common-types.js b/tensor2tensor/insights/polymer/common-types.js
index 13ecf2428..a206c3b9b 100644
--- a/tensor2tensor/insights/polymer/common-types.js
+++ b/tensor2tensor/insights/polymer/common-types.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/explore_view/explore-view.html b/tensor2tensor/insights/polymer/explore_view/explore-view.html
index 97fce423c..f74715016 100644
--- a/tensor2tensor/insights/polymer/explore_view/explore-view.html
+++ b/tensor2tensor/insights/polymer/explore_view/explore-view.html
@@ -1,6 +1,6 @@
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/explore_view/explore-view.js b/tensor2tensor/insights/polymer/explore_view/explore-view.js
index edd50aa06..ee78d47ed 100644
--- a/tensor2tensor/insights/polymer/explore_view/explore-view.js
+++ b/tensor2tensor/insights/polymer/explore_view/explore-view.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/graph_visualization/graph-visualization.html b/tensor2tensor/insights/polymer/graph_visualization/graph-visualization.html
index 930536632..152e7612a 100644
--- a/tensor2tensor/insights/polymer/graph_visualization/graph-visualization.html
+++ b/tensor2tensor/insights/polymer/graph_visualization/graph-visualization.html
@@ -1,6 +1,6 @@
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/graph_visualization/graph-visualization.js b/tensor2tensor/insights/polymer/graph_visualization/graph-visualization.js
index 0945c08a0..c273c348f 100644
--- a/tensor2tensor/insights/polymer/graph_visualization/graph-visualization.js
+++ b/tensor2tensor/insights/polymer/graph_visualization/graph-visualization.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/index.html b/tensor2tensor/insights/polymer/index.html
index e05956599..315dbad24 100644
--- a/tensor2tensor/insights/polymer/index.html
+++ b/tensor2tensor/insights/polymer/index.html
@@ -1,7 +1,7 @@
 <!doctype html>
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/insights_app/insights-app.html b/tensor2tensor/insights/polymer/insights_app/insights-app.html
index 8138177e0..622337c8f 100644
--- a/tensor2tensor/insights/polymer/insights_app/insights-app.html
+++ b/tensor2tensor/insights/polymer/insights_app/insights-app.html
@@ -1,6 +1,6 @@
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/insights_app/insights-app.js b/tensor2tensor/insights/polymer/insights_app/insights-app.js
index b652c2187..65e1ca252 100644
--- a/tensor2tensor/insights/polymer/insights_app/insights-app.js
+++ b/tensor2tensor/insights/polymer/insights_app/insights-app.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/language_selector/language-selector-content.html b/tensor2tensor/insights/polymer/language_selector/language-selector-content.html
index 3abaf7fa4..23bb4f370 100644
--- a/tensor2tensor/insights/polymer/language_selector/language-selector-content.html
+++ b/tensor2tensor/insights/polymer/language_selector/language-selector-content.html
@@ -1,6 +1,6 @@
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/language_selector/language-selector-content.js b/tensor2tensor/insights/polymer/language_selector/language-selector-content.js
index dc7e1de20..c5d783fdc 100644
--- a/tensor2tensor/insights/polymer/language_selector/language-selector-content.js
+++ b/tensor2tensor/insights/polymer/language_selector/language-selector-content.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/language_selector/language-selector.html b/tensor2tensor/insights/polymer/language_selector/language-selector.html
index 963484de9..9e6dbf1be 100644
--- a/tensor2tensor/insights/polymer/language_selector/language-selector.html
+++ b/tensor2tensor/insights/polymer/language_selector/language-selector.html
@@ -1,6 +1,6 @@
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/language_selector/language-selector.js b/tensor2tensor/insights/polymer/language_selector/language-selector.js
index 8441aca9d..919049df0 100644
--- a/tensor2tensor/insights/polymer/language_selector/language-selector.js
+++ b/tensor2tensor/insights/polymer/language_selector/language-selector.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/processing_visualization/processing-visualization.html b/tensor2tensor/insights/polymer/processing_visualization/processing-visualization.html
index 56c75b581..82e2b21e4 100644
--- a/tensor2tensor/insights/polymer/processing_visualization/processing-visualization.html
+++ b/tensor2tensor/insights/polymer/processing_visualization/processing-visualization.html
@@ -1,6 +1,6 @@
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/processing_visualization/processing-visualization.js b/tensor2tensor/insights/polymer/processing_visualization/processing-visualization.js
index 9cc9360ad..69379d88b 100644
--- a/tensor2tensor/insights/polymer/processing_visualization/processing-visualization.js
+++ b/tensor2tensor/insights/polymer/processing_visualization/processing-visualization.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/query_card/query-card.html b/tensor2tensor/insights/polymer/query_card/query-card.html
index d966d1c4f..f670a4cf7 100644
--- a/tensor2tensor/insights/polymer/query_card/query-card.html
+++ b/tensor2tensor/insights/polymer/query_card/query-card.html
@@ -1,6 +1,6 @@
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/query_card/query-card.js b/tensor2tensor/insights/polymer/query_card/query-card.js
index b08a82ae5..c0d25e750 100644
--- a/tensor2tensor/insights/polymer/query_card/query-card.js
+++ b/tensor2tensor/insights/polymer/query_card/query-card.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/tensor2tensor.html b/tensor2tensor/insights/polymer/tensor2tensor.html
index 10cbf3af1..23003a35f 100644
--- a/tensor2tensor/insights/polymer/tensor2tensor.html
+++ b/tensor2tensor/insights/polymer/tensor2tensor.html
@@ -1,7 +1,7 @@
 <!doctype html>
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/translation_result/translation-result.html b/tensor2tensor/insights/polymer/translation_result/translation-result.html
index fdcddcca3..bf152f81d 100644
--- a/tensor2tensor/insights/polymer/translation_result/translation-result.html
+++ b/tensor2tensor/insights/polymer/translation_result/translation-result.html
@@ -1,6 +1,6 @@
 <!--
 @license
-Copyright 2017 The Tensor2Tensor Authors.
+Copyright 2018 The Tensor2Tensor Authors.
 
 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/polymer/translation_result/translation-result.js b/tensor2tensor/insights/polymer/translation_result/translation-result.js
index 98329e3d1..995a5a822 100644
--- a/tensor2tensor/insights/polymer/translation_result/translation-result.js
+++ b/tensor2tensor/insights/polymer/translation_result/translation-result.js
@@ -1,6 +1,6 @@
 /**
  * @license
- * Copyright 2017 The Tensor2Tensor Authors.
+ * Copyright 2018 The Tensor2Tensor Authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/query_processor.py b/tensor2tensor/insights/query_processor.py
index 0aed3a313..905a20c95 100644
--- a/tensor2tensor/insights/query_processor.py
+++ b/tensor2tensor/insights/query_processor.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/server.py b/tensor2tensor/insights/server.py
index ca4dd9544..e61c465ff 100644
--- a/tensor2tensor/insights/server.py
+++ b/tensor2tensor/insights/server.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/insights/transformer_model.py b/tensor2tensor/insights/transformer_model.py
index ca36c5941..d7ac83a0a 100644
--- a/tensor2tensor/insights/transformer_model.py
+++ b/tensor2tensor/insights/transformer_model.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/__init__.py b/tensor2tensor/layers/__init__.py
index 3f714ce1f..4bd418a74 100644
--- a/tensor2tensor/layers/__init__.py
+++ b/tensor2tensor/layers/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/common_attention.py b/tensor2tensor/layers/common_attention.py
index 63bf8d6cd..5badb031d 100644
--- a/tensor2tensor/layers/common_attention.py
+++ b/tensor2tensor/layers/common_attention.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/common_attention_test.py b/tensor2tensor/layers/common_attention_test.py
index d67ef6704..bb84a515d 100644
--- a/tensor2tensor/layers/common_attention_test.py
+++ b/tensor2tensor/layers/common_attention_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/common_hparams.py b/tensor2tensor/layers/common_hparams.py
index 8fbd88bd2..159006b9b 100644
--- a/tensor2tensor/layers/common_hparams.py
+++ b/tensor2tensor/layers/common_hparams.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/common_image_attention.py b/tensor2tensor/layers/common_image_attention.py
index 88fc8ed93..9536125fa 100644
--- a/tensor2tensor/layers/common_image_attention.py
+++ b/tensor2tensor/layers/common_image_attention.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/common_layers.py b/tensor2tensor/layers/common_layers.py
index 7b22dc44b..1bfc97248 100644
--- a/tensor2tensor/layers/common_layers.py
+++ b/tensor2tensor/layers/common_layers.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/common_layers_test.py b/tensor2tensor/layers/common_layers_test.py
index 38c009926..2bf6b4cee 100644
--- a/tensor2tensor/layers/common_layers_test.py
+++ b/tensor2tensor/layers/common_layers_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/modalities.py b/tensor2tensor/layers/modalities.py
index f5788701c..3c54fa339 100644
--- a/tensor2tensor/layers/modalities.py
+++ b/tensor2tensor/layers/modalities.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/modalities_test.py b/tensor2tensor/layers/modalities_test.py
index 87fd2b7e5..1c305ed17 100644
--- a/tensor2tensor/layers/modalities_test.py
+++ b/tensor2tensor/layers/modalities_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/rev_block.py b/tensor2tensor/layers/rev_block.py
index 1684ebf2c..aaacf0c5d 100644
--- a/tensor2tensor/layers/rev_block.py
+++ b/tensor2tensor/layers/rev_block.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/layers/rev_block_test.py b/tensor2tensor/layers/rev_block_test.py
index de7a9bca2..62b167e72 100644
--- a/tensor2tensor/layers/rev_block_test.py
+++ b/tensor2tensor/layers/rev_block_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/__init__.py b/tensor2tensor/models/__init__.py
index 077f90847..2ed6ca883 100644
--- a/tensor2tensor/models/__init__.py
+++ b/tensor2tensor/models/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -35,7 +35,6 @@
 from tensor2tensor.models import neural_gpu
 from tensor2tensor.models import resnet
 from tensor2tensor.models import revnet
-from tensor2tensor.models import rl
 from tensor2tensor.models import shake_shake
 from tensor2tensor.models import slicenet
 from tensor2tensor.models import super_lm
diff --git a/tensor2tensor/models/aligned.py b/tensor2tensor/models/aligned.py
index 0cfa4be86..5f19d4db1 100644
--- a/tensor2tensor/models/aligned.py
+++ b/tensor2tensor/models/aligned.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/attention_lm.py b/tensor2tensor/models/attention_lm.py
index d92400fc8..bb847c1ec 100644
--- a/tensor2tensor/models/attention_lm.py
+++ b/tensor2tensor/models/attention_lm.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/attention_lm_moe.py b/tensor2tensor/models/attention_lm_moe.py
index fcf04f981..c931a4dbe 100644
--- a/tensor2tensor/models/attention_lm_moe.py
+++ b/tensor2tensor/models/attention_lm_moe.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/basic.py b/tensor2tensor/models/basic.py
index 1354259b5..35ae204b5 100644
--- a/tensor2tensor/models/basic.py
+++ b/tensor2tensor/models/basic.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/bytenet.py b/tensor2tensor/models/bytenet.py
index d65580f1f..cf576a0b3 100644
--- a/tensor2tensor/models/bytenet.py
+++ b/tensor2tensor/models/bytenet.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/bytenet_test.py b/tensor2tensor/models/bytenet_test.py
index 8a19ae905..d077a5b69 100644
--- a/tensor2tensor/models/bytenet_test.py
+++ b/tensor2tensor/models/bytenet_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/cycle_gan.py b/tensor2tensor/models/cycle_gan.py
index 0efeaeae5..5f964ef6a 100644
--- a/tensor2tensor/models/cycle_gan.py
+++ b/tensor2tensor/models/cycle_gan.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/gene_expression.py b/tensor2tensor/models/gene_expression.py
index 5e5eb35c5..0235774ab 100644
--- a/tensor2tensor/models/gene_expression.py
+++ b/tensor2tensor/models/gene_expression.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/gene_expression_test.py b/tensor2tensor/models/gene_expression_test.py
index 94cf20ff3..ae16b3383 100644
--- a/tensor2tensor/models/gene_expression_test.py
+++ b/tensor2tensor/models/gene_expression_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/lstm.py b/tensor2tensor/models/lstm.py
index 9348cdc2c..dd6f144ba 100644
--- a/tensor2tensor/models/lstm.py
+++ b/tensor2tensor/models/lstm.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/lstm_test.py b/tensor2tensor/models/lstm_test.py
index f9e472af9..4e35d34cb 100644
--- a/tensor2tensor/models/lstm_test.py
+++ b/tensor2tensor/models/lstm_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/multimodel.py b/tensor2tensor/models/multimodel.py
index 5253e2c98..88ab3950b 100644
--- a/tensor2tensor/models/multimodel.py
+++ b/tensor2tensor/models/multimodel.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/multimodel_test.py b/tensor2tensor/models/multimodel_test.py
index 6f490b280..ff233775a 100644
--- a/tensor2tensor/models/multimodel_test.py
+++ b/tensor2tensor/models/multimodel_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/neural_gpu.py b/tensor2tensor/models/neural_gpu.py
index fe1519344..1d1875743 100644
--- a/tensor2tensor/models/neural_gpu.py
+++ b/tensor2tensor/models/neural_gpu.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/neural_gpu_test.py b/tensor2tensor/models/neural_gpu_test.py
index 99b7f1062..bbf9f1d4d 100644
--- a/tensor2tensor/models/neural_gpu_test.py
+++ b/tensor2tensor/models/neural_gpu_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/resnet.py b/tensor2tensor/models/resnet.py
index 44c88dc8b..30275e7ca 100644
--- a/tensor2tensor/models/resnet.py
+++ b/tensor2tensor/models/resnet.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/resnet_test.py b/tensor2tensor/models/resnet_test.py
index ddcb1627b..12aca39bb 100644
--- a/tensor2tensor/models/resnet_test.py
+++ b/tensor2tensor/models/resnet_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/revnet.py b/tensor2tensor/models/revnet.py
index 3a6c7b32b..66e6a2abf 100644
--- a/tensor2tensor/models/revnet.py
+++ b/tensor2tensor/models/revnet.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/revnet_test.py b/tensor2tensor/models/revnet_test.py
index 4f8cf71b0..68fec94a2 100644
--- a/tensor2tensor/models/revnet_test.py
+++ b/tensor2tensor/models/revnet_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/rl.py b/tensor2tensor/models/rl.py
deleted file mode 100644
index 70622dd67..000000000
--- a/tensor2tensor/models/rl.py
+++ /dev/null
@@ -1,127 +0,0 @@
-# coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-"""Reinforcement learning models and parameters."""
-
-# Dependency imports
-
-import collections
-import functools
-import gym
-import operator
-import tensorflow as tf
-
-from tensor2tensor.layers import common_hparams
-from tensor2tensor.utils import registry
-
-
-@registry.register_hparams
-def ppo_base_v1():
-  """Set of hyperparameters."""
-  hparams = common_hparams.basic_params1()
-  hparams.learning_rate = 1e-4
-  hparams.add_hparam("init_mean_factor", 0.1)
-  hparams.add_hparam("init_logstd", 0.1)
-  hparams.add_hparam("policy_layers", (100, 100))
-  hparams.add_hparam("value_layers", (100, 100))
-  hparams.add_hparam("num_agents", 30)
-  hparams.add_hparam("clipping_coef", 0.2)
-  hparams.add_hparam("gae_gamma", 0.99)
-  hparams.add_hparam("gae_lambda", 0.95)
-  hparams.add_hparam("entropy_loss_coef", 0.01)
-  hparams.add_hparam("value_loss_coef", 1)
-  hparams.add_hparam("optimization_epochs", 15)
-  hparams.add_hparam("epoch_length", 200)
-  hparams.add_hparam("epochs_num", 2000)
-  return hparams
-
-@registry.register_hparams
-def pendulum():
-  hparams = ppo_base_v1()
-  hparams.add_hparam("environment", "Pendulum-v0")
-  hparams.add_hparam("network", feed_forward_gaussian_fun)
-  return hparams
-
-@registry.register_hparams
-def cartpole():
-  hparams = ppo_base_v1()
-  hparams.add_hparam("environment", "CartPole-v0")
-  hparams.add_hparam("network", feed_forward_categorical_fun)
-  return hparams
-
-
-# Neural networks for actor-critic algorithms
-
-NetworkOutput = collections.namedtuple(
-    'NetworkOutput', 'policy, value, action_postprocessing')
-
-
-def feed_forward_gaussian_fun(action_space, config, observations):
-  assert isinstance(action_space, gym.spaces.box.Box), \
-      'Expecting continuous action space.'
-  mean_weights_initializer = tf.contrib.layers.variance_scaling_initializer(
-      factor=config.init_mean_factor)
-  logstd_initializer = tf.random_normal_initializer(config.init_logstd, 1e-10)
-
-  flat_observations = tf.reshape(observations, [
-      tf.shape(observations)[0], tf.shape(observations)[1],
-      functools.reduce(operator.mul, observations.shape.as_list()[2:], 1)])
-
-  with tf.variable_scope('policy'):
-    x = flat_observations
-    for size in config.policy_layers:
-      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
-    mean = tf.contrib.layers.fully_connected(
-        x, action_space.shape[0], tf.tanh,
-        weights_initializer=mean_weights_initializer)
-    logstd = tf.get_variable(
-        'logstd', mean.shape[2:], tf.float32, logstd_initializer)
-    logstd = tf.tile(
-        logstd[None, None],
-        [tf.shape(mean)[0], tf.shape(mean)[1]] + [1] * (mean.shape.ndims - 2))
-  with tf.variable_scope('value'):
-    x = flat_observations
-    for size in config.value_layers:
-      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
-    value = tf.contrib.layers.fully_connected(x, 1, None)[..., 0]
-  mean = tf.check_numerics(mean, 'mean')
-  logstd = tf.check_numerics(logstd, 'logstd')
-  value = tf.check_numerics(value, 'value')
-
-  policy = tf.contrib.distributions.MultivariateNormalDiag(mean,
-                                                           tf.exp(logstd))
-
-  return NetworkOutput(policy, value, lambda a: tf.clip_by_value(a, -2., 2))
-
-
-def feed_forward_categorical_fun(action_space, config, observations):
-  assert isinstance(action_space, gym.spaces.Discrete), \
-      'Expecting discrete action space.'
-  flat_observations = tf.reshape(observations, [
-      tf.shape(observations)[0], tf.shape(observations)[1],
-      functools.reduce(operator.mul, observations.shape.as_list()[2:], 1)])
-  with tf.variable_scope('policy'):
-    x = flat_observations
-    for size in config.policy_layers:
-      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
-    logits = tf.contrib.layers.fully_connected(x, action_space.n,
-                                               activation_fn=None)
-  with tf.variable_scope('value'):
-    x = flat_observations
-    for size in config.value_layers:
-      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
-    value = tf.contrib.layers.fully_connected(x, 1, None)[..., 0]
-  policy = tf.contrib.distributions.Categorical(logits=logits)
-  return NetworkOutput(policy, value, lambda a: a)
diff --git a/tensor2tensor/models/shake_shake.py b/tensor2tensor/models/shake_shake.py
index 31a576338..6adbe5b70 100644
--- a/tensor2tensor/models/shake_shake.py
+++ b/tensor2tensor/models/shake_shake.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/slicenet.py b/tensor2tensor/models/slicenet.py
index 43cfa571e..fe5dab52d 100644
--- a/tensor2tensor/models/slicenet.py
+++ b/tensor2tensor/models/slicenet.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/slicenet_test.py b/tensor2tensor/models/slicenet_test.py
index 55bebf910..2aceb4d5e 100644
--- a/tensor2tensor/models/slicenet_test.py
+++ b/tensor2tensor/models/slicenet_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/super_lm.py b/tensor2tensor/models/super_lm.py
index 079507d2a..caaef23e6 100644
--- a/tensor2tensor/models/super_lm.py
+++ b/tensor2tensor/models/super_lm.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/transformer.py b/tensor2tensor/models/transformer.py
index 061b68ab7..e63a6d4f7 100644
--- a/tensor2tensor/models/transformer.py
+++ b/tensor2tensor/models/transformer.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/transformer_moe.py b/tensor2tensor/models/transformer_moe.py
index 2bf807b19..57d82edf9 100644
--- a/tensor2tensor/models/transformer_moe.py
+++ b/tensor2tensor/models/transformer_moe.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/transformer_revnet.py b/tensor2tensor/models/transformer_revnet.py
index bd4492151..c15ba314f 100644
--- a/tensor2tensor/models/transformer_revnet.py
+++ b/tensor2tensor/models/transformer_revnet.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/transformer_revnet_test.py b/tensor2tensor/models/transformer_revnet_test.py
index 79f8eb1e0..4ea19a0c9 100644
--- a/tensor2tensor/models/transformer_revnet_test.py
+++ b/tensor2tensor/models/transformer_revnet_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/transformer_sketch.py b/tensor2tensor/models/transformer_sketch.py
index e8fe796a8..0a1442d10 100644
--- a/tensor2tensor/models/transformer_sketch.py
+++ b/tensor2tensor/models/transformer_sketch.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/transformer_test.py b/tensor2tensor/models/transformer_test.py
index 1a6134b51..1e903a455 100644
--- a/tensor2tensor/models/transformer_test.py
+++ b/tensor2tensor/models/transformer_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/transformer_vae.py b/tensor2tensor/models/transformer_vae.py
index 59f2d08ba..d40b52c0e 100644
--- a/tensor2tensor/models/transformer_vae.py
+++ b/tensor2tensor/models/transformer_vae.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/vanilla_gan.py b/tensor2tensor/models/vanilla_gan.py
index 459753b95..242ca4d10 100644
--- a/tensor2tensor/models/vanilla_gan.py
+++ b/tensor2tensor/models/vanilla_gan.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/xception.py b/tensor2tensor/models/xception.py
index 401694580..d56321049 100644
--- a/tensor2tensor/models/xception.py
+++ b/tensor2tensor/models/xception.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/models/xception_test.py b/tensor2tensor/models/xception_test.py
index cb4e3544e..8e29b3bd0 100644
--- a/tensor2tensor/models/xception_test.py
+++ b/tensor2tensor/models/xception_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/notebooks/hello_t2t.ipynb b/tensor2tensor/notebooks/hello_t2t.ipynb
index 3b3573098..7d56dbede 100644
--- a/tensor2tensor/notebooks/hello_t2t.ipynb
+++ b/tensor2tensor/notebooks/hello_t2t.ipynb
@@ -29,7 +29,7 @@
         }
       },
       "source": [
-        "# Copyright 2017 Google LLC.\n",
+        "# Copyright 2018 Google LLC.\n",
         "\n",
         "# Licensed under the Apache License, Version 2.0 (the \"License\");\n",
         "# you may not use this file except in compliance with the License.\n",
diff --git a/tensor2tensor/problems.py b/tensor2tensor/problems.py
index 1e94c7bad..3c25abb40 100644
--- a/tensor2tensor/problems.py
+++ b/tensor2tensor/problems.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/problems_test.py b/tensor2tensor/problems_test.py
index de101e6e7..76742aafa 100644
--- a/tensor2tensor/problems_test.py
+++ b/tensor2tensor/problems_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/rl/README.md b/tensor2tensor/rl/README.md
index ff4fca9b9..053119f05 100644
--- a/tensor2tensor/rl/README.md
+++ b/tensor2tensor/rl/README.md
@@ -8,4 +8,4 @@ Currently the only supported algorithm is Proximy Policy Optimization - PPO.
 
 ## Sample usage - training in Pendulum-v0 environment.
 
-```python rl/t2t_rl_trainer.py --hparams_set pendulum [--output_dir dir_location]```
+```python rl/t2t_rl_trainer.py```
diff --git a/tensor2tensor/rl/__init__.py b/tensor2tensor/rl/__init__.py
index 3f714ce1f..4bd418a74 100644
--- a/tensor2tensor/rl/__init__.py
+++ b/tensor2tensor/rl/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/rl/collect.py b/tensor2tensor/rl/collect.py
index 728466ef5..afeefb72f 100644
--- a/tensor2tensor/rl/collect.py
+++ b/tensor2tensor/rl/collect.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -18,27 +18,26 @@
 import tensorflow as tf
 
 
-def define_collect(policy_factory, batch_env, hparams):
+def define_collect(policy_factory, batch_env, config):
   """Collect trajectories."""
-  memory_shape = [hparams.epoch_length] + [batch_env.observ.shape.as_list()[0]]
+  memory_shape = [config.epoch_length] + [batch_env.observ.shape.as_list()[0]]
   memories_shapes_and_types = [
       # observation
       (memory_shape + [batch_env.observ.shape.as_list()[1]], tf.float32),
       (memory_shape, tf.float32),      # reward
       (memory_shape, tf.bool),         # done
-      # action
-      (memory_shape + batch_env.action_shape, batch_env.action_dtype),
+      (memory_shape + batch_env.action_shape, tf.float32),  # action
       (memory_shape, tf.float32),      # pdf
       (memory_shape, tf.float32),      # value function
   ]
   memory = [tf.Variable(tf.zeros(shape, dtype), trainable=False)
             for (shape, dtype) in memories_shapes_and_types]
   cumulative_rewards = tf.Variable(
-      tf.zeros(hparams.num_agents, tf.float32), trainable=False)
+      tf.zeros(config.num_agents, tf.float32), trainable=False)
 
   should_reset_var = tf.Variable(True, trainable=False)
   reset_op = tf.cond(should_reset_var,
-                     lambda: batch_env.reset(tf.range(hparams.num_agents)),
+                     lambda: batch_env.reset(tf.range(config.num_agents)),
                      lambda: 0.0)
   with tf.control_dependencies([reset_op]):
     reset_once_op = tf.assign(should_reset_var, False)
@@ -59,7 +58,7 @@ def step(index, scores_sum, scores_num):
       pdf = policy.prob(action)[0]
       with tf.control_dependencies(simulate_output):
         reward, done = simulate_output
-        done = tf.reshape(done, (hparams.num_agents,))
+        done = tf.reshape(done, (config.num_agents,))
         to_save = [obs_copy, reward, done, action[0, ...], pdf,
                    actor_critic.value[0]]
         save_ops = [tf.scatter_update(memory_slot, index, value)
@@ -83,7 +82,7 @@ def step(index, scores_sum, scores_num):
 
     init = [tf.constant(0), tf.constant(0.0), tf.constant(0)]
     index, scores_sum, scores_num = tf.while_loop(
-        lambda c, _1, _2: c < hparams.epoch_length,
+        lambda c, _1, _2: c < config.epoch_length,
         step,
         init,
         parallel_iterations=1,
@@ -92,9 +91,5 @@ def step(index, scores_sum, scores_num):
                        lambda: scores_sum / tf.cast(scores_num, tf.float32),
                        lambda: 0.)
   printing = tf.Print(0, [mean_score, scores_sum, scores_num], "mean_score: ")
-  with tf.control_dependencies([index, printing]):
-    memory = [tf.identity(mem) for mem in memory]
-    summaries = tf.summary.merge(
-        [tf.summary.scalar("mean_score_this_iter", mean_score),
-         tf.summary.scalar("episodes_finished_this_iter", scores_num)])
-    return memory, summaries
+  with tf.control_dependencies([printing]):
+    return tf.identity(index), memory
diff --git a/tensor2tensor/rl/envs/__init__.py b/tensor2tensor/rl/envs/__init__.py
index 3f714ce1f..4bd418a74 100644
--- a/tensor2tensor/rl/envs/__init__.py
+++ b/tensor2tensor/rl/envs/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/rl/envs/batch_env.py b/tensor2tensor/rl/envs/batch_env.py
index 453348976..14f36a121 100644
--- a/tensor2tensor/rl/envs/batch_env.py
+++ b/tensor2tensor/rl/envs/batch_env.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/rl/envs/in_graph_batch_env.py b/tensor2tensor/rl/envs/in_graph_batch_env.py
index eae0826a3..44e30ddd9 100644
--- a/tensor2tensor/rl/envs/in_graph_batch_env.py
+++ b/tensor2tensor/rl/envs/in_graph_batch_env.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/rl/envs/utils.py b/tensor2tensor/rl/envs/utils.py
index 8171fbe17..4735f72ae 100644
--- a/tensor2tensor/rl/envs/utils.py
+++ b/tensor2tensor/rl/envs/utils.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/rl/networks.py b/tensor2tensor/rl/networks.py
new file mode 100644
index 000000000..e094abe66
--- /dev/null
+++ b/tensor2tensor/rl/networks.py
@@ -0,0 +1,69 @@
+# coding=utf-8
+# Copyright 2018 The Tensor2Tensor Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Neural networks for actor-critic algorithms."""
+
+import collections
+import functools
+import operator
+
+# Dependency imports
+
+import gym
+import tensorflow as tf
+
+
+NetworkOutput = collections.namedtuple(
+    'NetworkOutput', 'policy, value, action_postprocessing')
+
+
+def feed_forward_gaussian_fun(observation_space, action_space, config,
+                              observations):
+  """Feed-forward gaussian."""
+  assert isinstance(observation_space, gym.spaces.box.Box)
+
+  mean_weights_initializer = tf.contrib.layers.variance_scaling_initializer(
+      factor=config.init_mean_factor)
+  logstd_initializer = tf.random_normal_initializer(config.init_logstd, 1e-10)
+
+  flat_observations = tf.reshape(observations, [
+      tf.shape(observations)[0], tf.shape(observations)[1],
+      functools.reduce(operator.mul, observations.shape.as_list()[2:], 1)])
+
+  with tf.variable_scope('policy'):
+    x = flat_observations
+    for size in config.policy_layers:
+      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
+    mean = tf.contrib.layers.fully_connected(
+        x, action_space.shape[0], tf.tanh,
+        weights_initializer=mean_weights_initializer)
+    logstd = tf.get_variable(
+        'logstd', mean.shape[2:], tf.float32, logstd_initializer)
+    logstd = tf.tile(
+        logstd[None, None],
+        [tf.shape(mean)[0], tf.shape(mean)[1]] + [1] * (mean.shape.ndims - 2))
+  with tf.variable_scope('value'):
+    x = flat_observations
+    for size in config.value_layers:
+      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
+    value = tf.contrib.layers.fully_connected(x, 1, None)[..., 0]
+  mean = tf.check_numerics(mean, 'mean')
+  logstd = tf.check_numerics(logstd, 'logstd')
+  value = tf.check_numerics(value, 'value')
+
+  policy = tf.contrib.distributions.MultivariateNormalDiag(mean,
+                                                           tf.exp(logstd))
+
+  return NetworkOutput(policy, value, lambda a: tf.clip_by_value(a, -2., 2))
diff --git a/tensor2tensor/rl/ppo.py b/tensor2tensor/rl/ppo.py
index 9cf83657e..7c26cd074 100644
--- a/tensor2tensor/rl/ppo.py
+++ b/tensor2tensor/rl/ppo.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -32,8 +32,8 @@ def define_ppo_step(observation, action, reward, done, value, old_pdf,
   clipped_ratio = tf.clip_by_value(ratio, 1 - config.clipping_coef,
                                    1 + config.clipping_coef)
 
-  advantage = calculate_generalized_advantage_estimator(
-      reward, value, done, config.gae_gamma, config.gae_lambda)
+  advantage = calculate_discounted_return(
+      reward, value, done, config.gae_gamma, config.gae_lambda) - value
 
   advantage_mean, advantage_variance = tf.nn.moments(advantage, axes=[0, 1],
                                                      keep_dims=True)
@@ -53,11 +53,7 @@ def define_ppo_step(observation, action, reward, done, value, old_pdf,
 
   total_loss = policy_loss + value_loss + entropy_loss
 
-  optimization_op = tf.contrib.layers.optimize_loss(
-      loss=total_loss,
-      global_step=tf.train.get_or_create_global_step(),
-      optimizer=config.optimizer,
-      learning_rate=config.learning_rate)
+  optimization_op = config.optimizer(config.learning_rate).minimize(total_loss)
 
   with tf.control_dependencies([optimization_op]):
     return [tf.identity(x) for x in (policy_loss, value_loss, entropy_loss)]
@@ -83,17 +79,12 @@ def define_ppo_epoch(memory, policy_factory, config):
       [0., 0., 0.],
       parallel_iterations=1)
 
-  summaries = [tf.summary.scalar("policy loss", tf.reduce_mean(policy_loss)),
-               tf.summary.scalar("value loss", tf.reduce_mean(value_loss)),
-               tf.summary.scalar("entropy loss", tf.reduce_mean(entropy_loss))]
+  print_losses = tf.group(
+      tf.Print(0, [tf.reduce_mean(policy_loss)], 'policy loss: '),
+      tf.Print(0, [tf.reduce_mean(value_loss)], 'value loss: '),
+      tf.Print(0, [tf.reduce_mean(entropy_loss)], 'entropy loss: '))
 
-  losses_summary = tf.summary.merge(summaries)
-
-  losses_summary = tf.Print(losses_summary, [tf.reduce_mean(policy_loss)], 'policy loss: ')
-  losses_summary = tf.Print(losses_summary, [tf.reduce_mean(value_loss)], 'value loss: ')
-  losses_summary = tf.Print(losses_summary, [tf.reduce_mean(entropy_loss)], 'entropy loss: ')
-
-  return losses_summary
+  return print_losses
 
 
 def calculate_discounted_return(reward, value, done, discount, unused_lambda):
@@ -109,21 +100,3 @@ def calculate_discounted_return(reward, value, done, discount, unused_lambda):
       1,
       False), [0])
   return tf.check_numerics(return_, 'return')
-
-
-def calculate_generalized_advantage_estimator(reward, value, done, gae_gamma, gae_lambda):
-  """Generalized advantage estimator"""
-
-  #Below is slight wierdness, we set the last reward to 0.
-  # This makes the adventantage to be 0 in the last timestep
-  reward = tf.concat([reward[:-1,:], value[-1:,:]], axis=0)
-  next_value = tf.concat([value[1:,:], tf.zeros_like(value[-1:, :])], axis=0)
-  next_not_done = 1 - tf.cast(tf.concat([done[1:, :], tf.zeros_like(done[-1:, :])], axis=0), tf.float32)
-  delta = reward + gae_gamma * next_value * next_not_done - value
-
-  return_ = tf.reverse(tf.scan(
-      lambda agg, cur: cur[0] + cur[1] * gae_gamma * gae_lambda * agg,
-      [tf.reverse(delta, [0]), tf.reverse(next_not_done, [0])],
-      tf.zeros_like(delta[0, :]),
-      1,  False),  [0])
-  return tf.check_numerics(tf.stop_gradient(return_), 'return')
diff --git a/tensor2tensor/rl/rl_trainer_lib.py b/tensor2tensor/rl/rl_trainer_lib.py
index 0aebf5b73..ee7b2cc83 100644
--- a/tensor2tensor/rl/rl_trainer_lib.py
+++ b/tensor2tensor/rl/rl_trainer_lib.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""Library for training of RL agent with PPO algorithm."""
+"""Training of RL agent with PPO algorithm."""
 
 from __future__ import absolute_import
 
@@ -23,44 +23,65 @@
 
 import gym
 from tensor2tensor.rl import collect
+from tensor2tensor.rl import networks
 from tensor2tensor.rl import ppo
 from tensor2tensor.rl.envs import utils
 
 import tensorflow as tf
 
 
-def define_train(hparams):
+def define_train(policy_lambda, env_lambda, config):
   """Define the training setup."""
-  env_lambda = lambda: gym.make(hparams.environment)
-  policy_lambda = hparams.network
   env = env_lambda()
   action_space = env.action_space
+  observation_space = env.observation_space
 
-  batch_env = utils.define_batch_env(env_lambda, hparams.num_agents)
+  batch_env = utils.define_batch_env(env_lambda, config.num_agents)
 
   policy_factory = tf.make_template(
       "network",
-      functools.partial(policy_lambda, action_space, hparams))
+      functools.partial(policy_lambda, observation_space,
+                        action_space, config))
 
-  memory, collect_summary = collect.define_collect(policy_factory,
-                                                   batch_env, hparams)
-  ppo_summary = ppo.define_ppo_epoch(memory, policy_factory, hparams)
-  summary = tf.summary.merge([collect_summary, ppo_summary])
+  (collect_op, memory) = collect.define_collect(
+      policy_factory, batch_env, config)
 
-  return summary
+  with tf.control_dependencies([collect_op]):
+    ppo_op = ppo.define_ppo_epoch(memory, policy_factory, config)
 
+  return ppo_op
 
-def train(hparams, event_dir=None):
-  summary_op = define_train(hparams)
 
-  if event_dir:
-      summary_writer = tf.summary.FileWriter(event_dir, graph=tf.get_default_graph(), flush_secs=60)
-  else:
-      summary_writer = None
+def train(params):
+  policy_lambda, env_lambda, config = params
+  ppo_op = define_train(policy_lambda, env_lambda, config)
 
   with tf.Session() as sess:
     sess.run(tf.global_variables_initializer())
-    for epoch_index in range(hparams.epochs_num):
-      summary = sess.run(summary_op)
-      if summary_writer:
-        summary_writer.add_summary(summary, epoch_index)
+    for _ in range(config.epochs_num):
+      sess.run(ppo_op)
+
+
+def example_params():
+  """Example hyperparameters."""
+  config = tf.contrib.training.HParams(
+      init_mean_factor=0.1,
+      init_logstd=0.1,
+      policy_layers=(100, 100),
+      value_layers=(100, 100),
+      num_agents=30,
+      clipping_coef=0.2,
+      gae_gamma=0.99,
+      gae_lambda=0.95,
+      entropy_loss_coef=0.01,
+      value_loss_coef=1,
+      optimizer=tf.train.AdamOptimizer,
+      learning_rate=1e-4,
+      optimization_epochs=15,
+      epoch_length=200,
+      epochs_num=2000)
+  return networks.feed_forward_gaussian_fun, pendulum_lambda, config
+
+
+def pendulum_lambda():
+  return gym.make("Pendulum-v0")
diff --git a/tensor2tensor/rl/rl_trainer_lib_test.py b/tensor2tensor/rl/rl_trainer_lib_test.py
index 93481bcb9..0b753a580 100644
--- a/tensor2tensor/rl/rl_trainer_lib_test.py
+++ b/tensor2tensor/rl/rl_trainer_lib_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -17,22 +17,17 @@
 
 # Dependency imports
 
-import tensorflow as tf
-
-from tensor2tensor import models  # pylint: disable=unused-import
 from tensor2tensor.rl import rl_trainer_lib
-from tensor2tensor.utils import trainer_lib
+
+import tensorflow as tf
 
 
 class TrainTest(tf.test.TestCase):
 
   def test_no_crash_pendulum(self):
-    hparams = trainer_lib.create_hparams("pendulum", "epochs_num=10")
-    rl_trainer_lib.train(hparams)
-
-  def test_no_crash_cartpole(self):
-    hparams = trainer_lib.create_hparams("cartpole", "epochs_num=10")
-    rl_trainer_lib.train(hparams)
+    params = rl_trainer_lib.example_params()
+    params[2].epochs_num = 10
+    rl_trainer_lib.train(params)
 
 
 if __name__ == '__main__':
diff --git a/tensor2tensor/rl/t2t_rl_trainer.py b/tensor2tensor/rl/t2t_rl_trainer.py
index 798ca3a8e..6bdbcab06 100644
--- a/tensor2tensor/rl/t2t_rl_trainer.py
+++ b/tensor2tensor/rl/t2t_rl_trainer.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -15,27 +15,15 @@
 
 """Training of RL agent with PPO algorithm."""
 
-import tensorflow as tf
+# Dependency imports
 
-from tensor2tensor import models  # pylint: disable=unused-import
-from tensor2tensor.utils import flags as t2t_flags  # pylint: disable=unused-import
-from tensor2tensor.utils import trainer_lib
 from tensor2tensor.rl import rl_trainer_lib
 
-flags = tf.flags
-FLAGS = flags.FLAGS
-
-# To maintain compatibility with some internal libs, we guard against these flag
-# definitions possibly erroring. Apologies for the ugliness.
-try:
-  flags.DEFINE_string("output_dir", "", "Base output directory for run.")
-except:  # pylint: disable=bare-except
-  pass
+import tensorflow as tf
 
 
 def main(_):
-  hparams = trainer_lib.create_hparams(FLAGS.hparams_set, FLAGS.hparams)
-  rl_trainer_lib.train(hparams, FLAGS.output_dir)
+  rl_trainer_lib.train(rl_trainer_lib.example_params())
 
 
 if __name__ == "__main__":
diff --git a/tensor2tensor/serving/__init__.py b/tensor2tensor/serving/__init__.py
index 3f714ce1f..4bd418a74 100644
--- a/tensor2tensor/serving/__init__.py
+++ b/tensor2tensor/serving/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/serving/export.py b/tensor2tensor/serving/export.py
index 143668437..5fffdbff6 100644
--- a/tensor2tensor/serving/export.py
+++ b/tensor2tensor/serving/export.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/serving/query.py b/tensor2tensor/serving/query.py
index 4140b569e..9c3665fcb 100644
--- a/tensor2tensor/serving/query.py
+++ b/tensor2tensor/serving/query.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/test_data/example_usr_dir/__init__.py b/tensor2tensor/test_data/example_usr_dir/__init__.py
index 9bab20593..61a511e17 100644
--- a/tensor2tensor/test_data/example_usr_dir/__init__.py
+++ b/tensor2tensor/test_data/example_usr_dir/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/test_data/example_usr_dir/my_submodule.py b/tensor2tensor/test_data/example_usr_dir/my_submodule.py
index b6c3579ac..0872ce5ed 100644
--- a/tensor2tensor/test_data/example_usr_dir/my_submodule.py
+++ b/tensor2tensor/test_data/example_usr_dir/my_submodule.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/__init__.py b/tensor2tensor/utils/__init__.py
index 3f714ce1f..4bd418a74 100644
--- a/tensor2tensor/utils/__init__.py
+++ b/tensor2tensor/utils/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/avg_checkpoints.py b/tensor2tensor/utils/avg_checkpoints.py
index 4d1c56eda..6d73de2ac 100644
--- a/tensor2tensor/utils/avg_checkpoints.py
+++ b/tensor2tensor/utils/avg_checkpoints.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/beam_search.py b/tensor2tensor/utils/beam_search.py
index 3c7b8c203..10ee6c1f7 100644
--- a/tensor2tensor/utils/beam_search.py
+++ b/tensor2tensor/utils/beam_search.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/beam_search_test.py b/tensor2tensor/utils/beam_search_test.py
index 13a4e64c5..0d4f35a0a 100644
--- a/tensor2tensor/utils/beam_search_test.py
+++ b/tensor2tensor/utils/beam_search_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/bleu_hook.py b/tensor2tensor/utils/bleu_hook.py
index 3974244ec..fa200a436 100644
--- a/tensor2tensor/utils/bleu_hook.py
+++ b/tensor2tensor/utils/bleu_hook.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/bleu_hook_test.py b/tensor2tensor/utils/bleu_hook_test.py
index bb2677b71..01de21d26 100644
--- a/tensor2tensor/utils/bleu_hook_test.py
+++ b/tensor2tensor/utils/bleu_hook_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/checkpoint_compatibility_test.py b/tensor2tensor/utils/checkpoint_compatibility_test.py
index 3adc64b3e..c95fc8886 100644
--- a/tensor2tensor/utils/checkpoint_compatibility_test.py
+++ b/tensor2tensor/utils/checkpoint_compatibility_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/cloud_mlengine.py b/tensor2tensor/utils/cloud_mlengine.py
index 82ac23a39..7ffb36c55 100644
--- a/tensor2tensor/utils/cloud_mlengine.py
+++ b/tensor2tensor/utils/cloud_mlengine.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/cloud_tpu.py b/tensor2tensor/utils/cloud_tpu.py
index 53dd36bd0..96d011568 100644
--- a/tensor2tensor/utils/cloud_tpu.py
+++ b/tensor2tensor/utils/cloud_tpu.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/data_reader.py b/tensor2tensor/utils/data_reader.py
index 14de558b0..8cece2625 100644
--- a/tensor2tensor/utils/data_reader.py
+++ b/tensor2tensor/utils/data_reader.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/data_reader_test.py b/tensor2tensor/utils/data_reader_test.py
index 85cf58930..457ece27a 100644
--- a/tensor2tensor/utils/data_reader_test.py
+++ b/tensor2tensor/utils/data_reader_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/decoding.py b/tensor2tensor/utils/decoding.py
index 8133eeca4..4e95c7e5a 100644
--- a/tensor2tensor/utils/decoding.py
+++ b/tensor2tensor/utils/decoding.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/devices.py b/tensor2tensor/utils/devices.py
index 4211bca15..a405b498e 100644
--- a/tensor2tensor/utils/devices.py
+++ b/tensor2tensor/utils/devices.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/diet.py b/tensor2tensor/utils/diet.py
index 7ecfba693..7f1915e4d 100644
--- a/tensor2tensor/utils/diet.py
+++ b/tensor2tensor/utils/diet.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/diet_test.py b/tensor2tensor/utils/diet_test.py
index d5815db2e..4e9202440 100644
--- a/tensor2tensor/utils/diet_test.py
+++ b/tensor2tensor/utils/diet_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/expert_utils.py b/tensor2tensor/utils/expert_utils.py
index 9691b5de5..1d465b8e7 100644
--- a/tensor2tensor/utils/expert_utils.py
+++ b/tensor2tensor/utils/expert_utils.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/expert_utils_test.py b/tensor2tensor/utils/expert_utils_test.py
index f9abc72c1..eccad731b 100644
--- a/tensor2tensor/utils/expert_utils_test.py
+++ b/tensor2tensor/utils/expert_utils_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/flags.py b/tensor2tensor/utils/flags.py
index f5d28e9f3..2017cf019 100644
--- a/tensor2tensor/utils/flags.py
+++ b/tensor2tensor/utils/flags.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/get_rouge.py b/tensor2tensor/utils/get_rouge.py
index dc9355b0d..50f0994d9 100644
--- a/tensor2tensor/utils/get_rouge.py
+++ b/tensor2tensor/utils/get_rouge.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/metrics.py b/tensor2tensor/utils/metrics.py
index 724dbb740..1f9e2ed00 100644
--- a/tensor2tensor/utils/metrics.py
+++ b/tensor2tensor/utils/metrics.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/metrics_hook.py b/tensor2tensor/utils/metrics_hook.py
index 964139a42..78681df50 100644
--- a/tensor2tensor/utils/metrics_hook.py
+++ b/tensor2tensor/utils/metrics_hook.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/metrics_hook_test.py b/tensor2tensor/utils/metrics_hook_test.py
index 67c78eb2d..1350679a7 100644
--- a/tensor2tensor/utils/metrics_hook_test.py
+++ b/tensor2tensor/utils/metrics_hook_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/metrics_test.py b/tensor2tensor/utils/metrics_test.py
index 528fd4755..2aee025c1 100644
--- a/tensor2tensor/utils/metrics_test.py
+++ b/tensor2tensor/utils/metrics_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/modality.py b/tensor2tensor/utils/modality.py
index f2525e313..611d4ea04 100644
--- a/tensor2tensor/utils/modality.py
+++ b/tensor2tensor/utils/modality.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/optimize.py b/tensor2tensor/utils/optimize.py
index bc09c009d..6b303d7d0 100644
--- a/tensor2tensor/utils/optimize.py
+++ b/tensor2tensor/utils/optimize.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/registry.py b/tensor2tensor/utils/registry.py
index aaa2be9d2..65caa5181 100644
--- a/tensor2tensor/utils/registry.py
+++ b/tensor2tensor/utils/registry.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/registry_test.py b/tensor2tensor/utils/registry_test.py
index d97dc6bdc..9c7d9d8f2 100644
--- a/tensor2tensor/utils/registry_test.py
+++ b/tensor2tensor/utils/registry_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/rouge.py b/tensor2tensor/utils/rouge.py
index 29c84729f..d8e4349c8 100644
--- a/tensor2tensor/utils/rouge.py
+++ b/tensor2tensor/utils/rouge.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,20 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-# -*- coding: utf-8 -*-
-# Copyright 2017 Google Inc.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#      http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
+# coding=utf-8
 """ROUGe metric implementation.
 
 This is a modified and slightly extended verison of
diff --git a/tensor2tensor/utils/rouge_test.py b/tensor2tensor/utils/rouge_test.py
index 2a8c260e2..a73fd309b 100644
--- a/tensor2tensor/utils/rouge_test.py
+++ b/tensor2tensor/utils/rouge_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/t2t_model.py b/tensor2tensor/utils/t2t_model.py
index aadf5e358..f21562eeb 100644
--- a/tensor2tensor/utils/t2t_model.py
+++ b/tensor2tensor/utils/t2t_model.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/trainer_lib.py b/tensor2tensor/utils/trainer_lib.py
index 0e64f2475..2b0543f37 100644
--- a/tensor2tensor/utils/trainer_lib.py
+++ b/tensor2tensor/utils/trainer_lib.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/trainer_lib_test.py b/tensor2tensor/utils/trainer_lib_test.py
index 5df62d2cb..ef117274e 100644
--- a/tensor2tensor/utils/trainer_lib_test.py
+++ b/tensor2tensor/utils/trainer_lib_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/usr_dir.py b/tensor2tensor/utils/usr_dir.py
index 5edd6f1a2..4aab19ac9 100644
--- a/tensor2tensor/utils/usr_dir.py
+++ b/tensor2tensor/utils/usr_dir.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/yellowfin.py b/tensor2tensor/utils/yellowfin.py
index a3f6a18a1..87b3ec9dd 100644
--- a/tensor2tensor/utils/yellowfin.py
+++ b/tensor2tensor/utils/yellowfin.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/utils/yellowfin_test.py b/tensor2tensor/utils/yellowfin_test.py
index 2130be2b3..4d3a2584a 100644
--- a/tensor2tensor/utils/yellowfin_test.py
+++ b/tensor2tensor/utils/yellowfin_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/visualization/__init__.py b/tensor2tensor/visualization/__init__.py
index b62605264..c2a9550b0 100644
--- a/tensor2tensor/visualization/__init__.py
+++ b/tensor2tensor/visualization/__init__.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/visualization/attention.py b/tensor2tensor/visualization/attention.py
index 11ed75b8d..e2a0a0551 100644
--- a/tensor2tensor/visualization/attention.py
+++ b/tensor2tensor/visualization/attention.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/visualization/visualization.py b/tensor2tensor/visualization/visualization.py
index e92ca518c..f11074a96 100644
--- a/tensor2tensor/visualization/visualization.py
+++ b/tensor2tensor/visualization/visualization.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
diff --git a/tensor2tensor/visualization/visualization_test.py b/tensor2tensor/visualization/visualization_test.py
index b4d6b96f6..c430c815e 100644
--- a/tensor2tensor/visualization/visualization_test.py
+++ b/tensor2tensor/visualization/visualization_test.py
@@ -1,5 +1,5 @@
 # coding=utf-8
-# Copyright 2017 The Tensor2Tensor Authors.
+# Copyright 2018 The Tensor2Tensor Authors.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

From 52509fb243992725ecc3843e51b00e48fdea553d Mon Sep 17 00:00:00 2001
From: Ryan Sepassi <rsepassi@google.com>
Date: Thu, 8 Feb 2018 16:22:12 -0800
Subject: [PATCH 02/14] Fix call order for rouge (issue #541)

PiperOrigin-RevId: 185070260
---
 tensor2tensor/utils/rouge.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tensor2tensor/utils/rouge.py b/tensor2tensor/utils/rouge.py
index d8e4349c8..aea3a5623 100644
--- a/tensor2tensor/utils/rouge.py
+++ b/tensor2tensor/utils/rouge.py
@@ -148,7 +148,7 @@ def rouge_l_fscore(predictions, labels, **unused_kwargs):
   # Convert the outputs and labels to a [batch_size, input_length] tensor.
   outputs = tf.squeeze(outputs, axis=[-1, -2])
   labels = tf.squeeze(labels, axis=[-1, -2])
-  rouge_l_f_score = tf.py_func(rouge_l_sentence_level, (labels, outputs),
+  rouge_l_f_score = tf.py_func(rouge_l_sentence_level, (outputs, labels),
                                tf.float32)
   return rouge_l_f_score, tf.constant(1.0)
 

From a7fbd2eaca1a6c2b87e20776668430af82a83649 Mon Sep 17 00:00:00 2001
From: Ryan Sepassi <rsepassi@google.com>
Date: Thu, 8 Feb 2018 17:33:43 -0800
Subject: [PATCH 03/14] Fix demo VanillaGan model

PiperOrigin-RevId: 185078802
---
 tensor2tensor/models/vanilla_gan.py | 98 +++++++++++++----------------
 1 file changed, 44 insertions(+), 54 deletions(-)

diff --git a/tensor2tensor/models/vanilla_gan.py b/tensor2tensor/models/vanilla_gan.py
index 242ca4d10..1d66bbba2 100644
--- a/tensor2tensor/models/vanilla_gan.py
+++ b/tensor2tensor/models/vanilla_gan.py
@@ -35,10 +35,10 @@
 def generator(z, hparams, reuse=False):
   """Initalizes generator layers."""
 
-  g_h1 = tf.layers.dense(z, hparams.weight_size, activation=tf.nn.relu,
-                         name="l1", reuse=reuse)
-  g_log_prob = tf.layers.dense(g_h1, hparams.input_size, name="logp",
-                               reuse=reuse)
+  g_h1 = tf.layers.dense(
+      z, hparams.hidden_dim, activation=tf.nn.relu, name="l1", reuse=reuse)
+  g_log_prob = tf.layers.dense(
+      g_h1, hparams.height * hparams.width, name="logp", reuse=reuse)
   g_prob = tf.nn.sigmoid(g_log_prob)
 
   return g_prob
@@ -46,9 +46,8 @@ def generator(z, hparams, reuse=False):
 
 def discriminator(x, hparams, reuse=False):
   """Initalizes discriminator layers."""
-
-  d_h1 = tf.layers.dense(x, hparams.weight_size, activation=tf.nn.relu,
-                         name="d_h1", reuse=reuse)
+  d_h1 = tf.layers.dense(
+      x, hparams.hidden_dim, activation=tf.nn.relu, name="d_h1", reuse=reuse)
   d_logit = tf.layers.dense(d_h1, 1, name="d_logit", reuse=reuse)
   d_prob = tf.nn.sigmoid(d_logit)
 
@@ -56,35 +55,38 @@ def discriminator(x, hparams, reuse=False):
 
 
 def reverse_grad(x):
-  return tf.stop_gradient(2*x) - x
+  return tf.stop_gradient(2 * x) - x
 
 
 def vanilla_gan_internal(inputs, hparams, train):
   with tf.variable_scope("vanilla_gan", reuse=tf.AUTO_REUSE):
-    x = common_layers.flatten4d3d(inputs)
-
-    batch_size = tf.shape(inputs)[0]
-    # Currently uses one of three color layers.
-    x = x[:, :, 0]
-    x.set_shape([None, hparams.input_size])
-
-    if train:
-      z = tf.random_uniform(shape=[batch_size,
-                                   hparams.random_sample_size],
-                            minval=-1, maxval=1, name="z")
-    else:
-      z = tf.random_uniform(shape=[1, hparams.random_sample_size],
-                            minval=-1, maxval=1, name="z")
-
+    batch_size, height, width, _ = common_layers.shape_list(inputs)
+    assert height == hparams.height
+    assert width == hparams.width
+
+    # Currently uses only one of RGB
+    x = inputs
+    x = x[:, :, :, 0]
+    x = tf.reshape(x, [batch_size, height * width])
+
+    # Generate a fake image
+    z = tf.random_uniform(
+        shape=[batch_size, hparams.random_sample_size],
+        minval=-1,
+        maxval=1,
+        name="z")
     g_sample = generator(z, hparams)
 
+    # Discriminate on the real image
     d_real, _ = discriminator(x, hparams)
 
-    d_fake, _ = discriminator(reverse_grad(g_sample), hparams,
-                              reuse=True)
-    d_loss = -tf.reduce_mean(tf.log(d_real+hparams.epsilon)
-                             + tf.log(1. - d_fake))
-    g_loss = -tf.reduce_mean(tf.log(d_fake+hparams.epsilon))
+    # Discriminate on the fake image
+    d_fake, _ = discriminator(reverse_grad(g_sample), hparams, reuse=True)
+
+    # GAN losses
+    d_loss = -tf.reduce_mean(
+        tf.log(d_real + hparams.epsilon) + tf.log(1. - d_fake))
+    g_loss = -tf.reduce_mean(tf.log(d_fake + hparams.epsilon))
 
     losses = {}
     losses["discriminator"] = d_loss
@@ -92,31 +94,25 @@ def vanilla_gan_internal(inputs, hparams, train):
     # Include a dummy training loss to skip self.top and self.loss
     losses["training"] = tf.constant(0., dtype=tf.float32)
 
-    z_sampled = tf.random_uniform(shape=[1, hparams.random_sample_size],
-                                  minval=-1, maxval=1, name="z")
-    g_sample = generator(z_sampled, hparams, reuse=True)
-    g_reshaped_sample = tf.reshape(g_sample,
-                                   [1, hparams.height, hparams.width, 1])
-    tf.summary.image("generated", g_reshaped_sample, max_outputs=1)
+    summary_g_image = tf.reshape(g_sample[0, :], [1, height, width, 1])
+    tf.summary.image("generated", summary_g_image, max_outputs=1)
 
     if train:
-      # Returns an empty output, and loss dictionary.
-      return tf.zeros(shape=[1, 1]), losses
+      # Returns an dummy output and the losses dictionary.
+      return tf.zeros([batch_size, 1]), losses
     else:
       return g_sample, losses
 
 
 @registry.register_model
 class VanillaGan(t2t_model.T2TModel):
-  """Simple GAN.
-  """
+  """Simple GAN for demonstration."""
 
   def body(self, features):
     """Computes the generator and discriminator loss.
 
     Args:
-      features: A dictionary of key to Tensor. Each Tensor has shape
-         [batch_size, ?, ?, hidden_size].
+      features: A dictionary of key to Tensor. "inputs" should be an image.
 
     Returns:
       output: Tensor containing one zero. GANs do not make use of the modality
@@ -124,8 +120,8 @@ def body(self, features):
       losses: a dictionary of losses containing the generator and discriminator
         losses.
     """
-    train = self._hparams.mode == tf.estimator.ModeKeys.TRAIN
-    return vanilla_gan_internal(features["targets"], self._hparams, train)
+    train = self.hparams.mode == tf.estimator.ModeKeys.TRAIN
+    return vanilla_gan_internal(features["inputs"], self.hparams, train)
 
   def infer(self,
             features=None,
@@ -135,8 +131,11 @@ def infer(self,
             last_position_only=False,
             alpha=0.0):
     with tf.variable_scope("body/vanilla_gan", reuse=tf.AUTO_REUSE):
-      z = tf.random_uniform(shape=[1, self._hparams.random_sample_size],
-                            minval=-1, maxval=1, name="z")
+      z = tf.random_uniform(
+          shape=[1, self._hparams.random_sample_size],
+          minval=-1,
+          maxval=1,
+          name="z")
 
       g_sample = generator(z, self._hparams)
       return g_sample
@@ -145,22 +144,13 @@ def infer(self,
 @registry.register_hparams
 def vanilla_gan():
   """Basic parameters for a vanilla_gan."""
-
   hparams = common_hparams.basic_params1()
 
   hparams.batch_size = 32
   hparams.label_smoothing = 0.0
-  hparams.add_hparam("startup_steps", 10000)
-
-  hparams.train_steps = 100
-  hparams.add_hparam("weight_size", 128)
+  hparams.add_hparam("hidden_dim", 128)
   hparams.add_hparam("random_sample_size", 100)
   hparams.add_hparam("height", 28)
   hparams.add_hparam("width", 28)
-  hparams.add_hparam("colors", 1)
-  hparams.add_hparam("input_size", 784)
   hparams.add_hparam("epsilon", 1e-4)
-  hparams.learning_rate_warmup_steps = 0
-  hparams.learning_rate = 0.2
-  hparams.learning_rate_decay_scheme = "none"
   return hparams

From 85c60db37f0f7c482feb46a5c8977090ce5a843e Mon Sep 17 00:00:00 2001
From: Lukasz Kaiser <lukaszkaiser@google.com>
Date: Thu, 8 Feb 2018 17:38:29 -0800
Subject: [PATCH 04/14] Correct latent predicton input in transformer ae.

PiperOrigin-RevId: 185079318
---
 tensor2tensor/models/transformer_vae.py | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/tensor2tensor/models/transformer_vae.py b/tensor2tensor/models/transformer_vae.py
index d40b52c0e..ba66fc797 100644
--- a/tensor2tensor/models/transformer_vae.py
+++ b/tensor2tensor/models/transformer_vae.py
@@ -600,8 +600,9 @@ def symbols_to_logits_fn(ids):
 
 def ae_latent_sample(latents_dense, inputs, ed, embed, iters, hparams):
   """Sample from the latent space in the autoencoder."""
-  if hparams.num_decode_blocks < 2:
+  if hparams.num_decode_blocks < 2 and hparams.sampling_temp == 0.0:
     # TODO(lukaszkaiser): beam-search only works in non-blocked mode for now.
+    tf.logging.info("Running beam-search for latents with beam size 1.")
     return ae_latent_sample_beam(latents_dense, inputs, ed, embed, hparams)
   latents_pred = decode_transformer(inputs, ed, latents_dense, hparams, "extra")
   latents_discrete, _ = ae_latent_softmax(latents_pred, None, hparams)
@@ -681,9 +682,9 @@ def ae_transformer_internal(inputs,
       # Extra loss predicting latent code from input. Discrete only.
       if hparams.bottleneck_kind not in ["dense", "vae"]:
         latents_pred = decode_transformer(
-            tf.stop_gradient(inputs) if inputs is not None else None,
-            tf.stop_gradient(ed) if inputs is not None else None,
-            tf.stop_gradient(latents_dense), hparams, "extra",
+            inputs if inputs is not None else None,
+            ed if inputs is not None else None,
+            embed(latents_discrete), hparams, "extra",
             task="translate")
         _, latent_pred_loss = ae_latent_softmax(
             latents_pred, latents_discrete, hparams)
@@ -916,6 +917,7 @@ def transformer_ae_small():
   """Set of hyperparameters."""
   hparams = transformer.transformer_small()
   hparams.batch_size = 2048
+  hparams.learning_rate = 0.2
   hparams.learning_rate_warmup_steps = 4000
   hparams.num_hidden_layers = 3
   hparams.hidden_size = 384
@@ -946,7 +948,6 @@ def transformer_ae_small():
   hparams.add_hparam("v_size", 1024*64)
   hparams.add_hparam("max_context_length", 64)
   hparams.add_hparam("num_compress_steps", 3)
-  hparams.add_hparam("kl_steps", 35000)
   hparams.add_hparam("startup_steps", 10000)
   hparams.add_hparam("mask_startup_steps", 50000)
   hparams.add_hparam("kmeans_lr_factor", 0.002)

From 944cfbed9c4cfbff640dd82ff4384c0d5b01e696 Mon Sep 17 00:00:00 2001
From: Ryan Sepassi <rsepassi@google.com>
Date: Fri, 9 Feb 2018 10:47:51 -0800
Subject: [PATCH 05/14] Update jobId for Cloud ML Engine

PiperOrigin-RevId: 185162214
---
 tensor2tensor/utils/cloud_mlengine.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/tensor2tensor/utils/cloud_mlengine.py b/tensor2tensor/utils/cloud_mlengine.py
index 7ffb36c55..dda6d9654 100644
--- a/tensor2tensor/utils/cloud_mlengine.py
+++ b/tensor2tensor/utils/cloud_mlengine.py
@@ -15,6 +15,7 @@
 
 """Launch on GCP's ML Engine."""
 
+import datetime
 import os
 import shutil
 import sys
@@ -91,7 +92,6 @@ def configure_job():
   """Construct jobSpec for ML Engine job."""
   train_dir = FLAGS.output_dir
   assert train_dir.startswith('gs://')
-  job_name = os.path.basename(train_dir)
 
   # See documentation:
   # https://cloud.google.com/ml-engine/reference/rest/v1/projects.jobs#traininginput
@@ -122,6 +122,8 @@ def configure_job():
   if training_input['scaleTier'] == 'CUSTOM':
     assert 'masterType' in training_input
 
+  timestamp = datetime.datetime.now().strftime('%Y%m%d_%H%M%S')
+  job_name = '%s_%s_t2t_%s' % (FLAGS.model, FLAGS.problems, timestamp)
   job_spec = {'jobId': job_name, 'trainingInput': training_input}
   return job_spec
 

From 4cdbc0980e023c8222cb3a5bde5ccf15520677d2 Mon Sep 17 00:00:00 2001
From: Ryan Sepassi <rsepassi@google.com>
Date: Fri, 9 Feb 2018 11:56:43 -0800
Subject: [PATCH 06/14] Small var name changes

PiperOrigin-RevId: 185172558
---
 tensor2tensor/data_generators/generator_utils.py | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/tensor2tensor/data_generators/generator_utils.py b/tensor2tensor/data_generators/generator_utils.py
index d25e5c621..1074fd327 100644
--- a/tensor2tensor/data_generators/generator_utils.py
+++ b/tensor2tensor/data_generators/generator_utils.py
@@ -83,8 +83,8 @@ def generate_files_distributed(generator,
     counter += 1
     if max_cases and counter > max_cases:
       break
-    sequence_example = to_example(case)
-    writer.write(sequence_example.SerializeToString())
+    example = to_example(case)
+    writer.write(example.SerializeToString())
 
   writer.close()
   return output_file
@@ -151,13 +151,15 @@ def generate_files(generator, output_filenames, max_cases=None):
   writers = [tf.python_io.TFRecordWriter(fname) for fname in output_filenames]
   counter, shard = 0, 0
   for case in generator:
+    if case is None:
+      continue
     if counter > 0 and counter % 100000 == 0:
       tf.logging.info("Generating case %d." % counter)
     counter += 1
     if max_cases and counter > max_cases:
       break
-    sequence_example = to_example(case)
-    writers[shard].write(sequence_example.SerializeToString())
+    example = to_example(case)
+    writers[shard].write(example.SerializeToString())
     shard = (shard + 1) % num_shards
 
   for writer in writers:

From 61906cc72387b2e33dd63590f131ebbe4effc70a Mon Sep 17 00:00:00 2001
From: Ryan Sepassi <rsepassi@google.com>
Date: Fri, 9 Feb 2018 14:09:50 -0800
Subject: [PATCH 07/14] Fix hyperparameter prefixing for ML Engine autotuning

PiperOrigin-RevId: 185190665
---
 tensor2tensor/bin/t2t_trainer.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/tensor2tensor/bin/t2t_trainer.py b/tensor2tensor/bin/t2t_trainer.py
index 82a74d3e5..980376b57 100644
--- a/tensor2tensor/bin/t2t_trainer.py
+++ b/tensor2tensor/bin/t2t_trainer.py
@@ -132,7 +132,7 @@ def set_hparams_from_args(args):
   while i < len(args):
     arg = args[i]
     if arg.startswith(hp_prefix):
-      pairs.append((arg.lstrip(hp_prefix), args[i+1]))
+      pairs.append((arg[len(hp_prefix):], args[i+1]))
       i += 2
     else:
       tf.logging.warn("Found unknown flag: %s", arg)

From 9d4c00f3a74daf2f23c9f66fd9b40117030c2695 Mon Sep 17 00:00:00 2001
From: T2T Team <no-reply@google.com>
Date: Fri, 9 Feb 2018 16:15:46 -0800
Subject: [PATCH 08/14] Shorten the feature name for expected_attentions, too
 verbose.

PiperOrigin-RevId: 185208468
---
 tensor2tensor/models/transformer.py      | 6 +++---
 tensor2tensor/models/transformer_test.py | 2 +-
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/tensor2tensor/models/transformer.py b/tensor2tensor/models/transformer.py
index e63a6d4f7..c67a5fb4d 100644
--- a/tensor2tensor/models/transformer.py
+++ b/tensor2tensor/models/transformer.py
@@ -168,10 +168,10 @@ def body(self, features):
         hparams,
         nonpadding=features_to_nonpadding(features, "targets"))
 
-    expected_attention_weights = features.get("expected_attention_weights")
-    if expected_attention_weights is not None:
+    expected_attentions = features.get("expected_attentions")
+    if expected_attentions is not None:
       attention_loss = common_attention.encoder_decoder_attention_loss(
-          expected_attention_weights, self.attention_weights)
+          expected_attentions, self.attention_weights)
       return decoder_output, {"attention_loss": attention_loss}
 
     return decoder_output
diff --git a/tensor2tensor/models/transformer_test.py b/tensor2tensor/models/transformer_test.py
index 1e903a455..88581ac87 100644
--- a/tensor2tensor/models/transformer_test.py
+++ b/tensor2tensor/models/transformer_test.py
@@ -211,7 +211,7 @@ def testTransformerWithEncoderDecoderAttentionLoss(self):
     model, features = self.getModel(transformer.transformer_small())
     expected_attention_weights = np.random.random_sample(
         size=(BATCH_SIZE, TARGET_LENGTH, INPUT_LENGTH))
-    features["expected_attention_weights"] = tf.constant(
+    features["expected_attentions"] = tf.constant(
         expected_attention_weights, dtype=tf.float32)
     _, extra_loss = model(features)
     with self.test_session() as session:

From d23d241cdfe748b3ff4772403a090997bf8e86c8 Mon Sep 17 00:00:00 2001
From: Lukasz Kaiser <lukaszkaiser@google.com>
Date: Sat, 10 Feb 2018 14:34:39 -0800
Subject: [PATCH 09/14] internal merge

PiperOrigin-RevId: 185267339
---
 tensor2tensor/bin/t2t_trainer.py        |   2 +-
 tensor2tensor/models/rl.py              | 131 ++++++++++++++++++++++++
 tensor2tensor/rl/README.md              |   7 +-
 tensor2tensor/rl/collect.py             |  23 +++--
 tensor2tensor/rl/networks.py            |  69 -------------
 tensor2tensor/rl/ppo.py                 |  50 +++++++--
 tensor2tensor/rl/rl_trainer_lib.py      |  68 +++++-------
 tensor2tensor/rl/rl_trainer_lib_test.py |  12 ++-
 tensor2tensor/rl/t2t_rl_trainer.py      |  15 ++-
 9 files changed, 238 insertions(+), 139 deletions(-)
 create mode 100644 tensor2tensor/models/rl.py
 delete mode 100644 tensor2tensor/rl/networks.py

diff --git a/tensor2tensor/bin/t2t_trainer.py b/tensor2tensor/bin/t2t_trainer.py
index 980376b57..b2545a159 100644
--- a/tensor2tensor/bin/t2t_trainer.py
+++ b/tensor2tensor/bin/t2t_trainer.py
@@ -274,7 +274,7 @@ def save_metadata(hparams):
   # Save hparams as hparams.json
   hparams_fname = os.path.join(output_dir, "hparams.json")
   with tf.gfile.Open(hparams_fname, "w") as f:
-    f.write(hparams.to_json())
+    f.write(hparams.to_json(indent=0, sort_keys=True))
 
 
 def execute_schedule(exp):
diff --git a/tensor2tensor/models/rl.py b/tensor2tensor/models/rl.py
new file mode 100644
index 000000000..4c63b77d0
--- /dev/null
+++ b/tensor2tensor/models/rl.py
@@ -0,0 +1,131 @@
+# coding=utf-8
+# Copyright 2018 The Tensor2Tensor Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Reinforcement learning models and parameters."""
+
+import collections
+import functools
+import operator
+
+# Dependency imports
+
+import gym
+from tensor2tensor.layers import common_hparams
+from tensor2tensor.utils import registry
+
+import tensorflow as tf
+
+
+@registry.register_hparams
+def ppo_base_v1():
+  """Set of hyperparameters."""
+  hparams = common_hparams.basic_params1()
+  hparams.learning_rate = 1e-4
+  hparams.add_hparam("init_mean_factor", 0.1)
+  hparams.add_hparam("init_logstd", 0.1)
+  hparams.add_hparam("policy_layers", (100, 100))
+  hparams.add_hparam("value_layers", (100, 100))
+  hparams.add_hparam("num_agents", 30)
+  hparams.add_hparam("clipping_coef", 0.2)
+  hparams.add_hparam("gae_gamma", 0.99)
+  hparams.add_hparam("gae_lambda", 0.95)
+  hparams.add_hparam("entropy_loss_coef", 0.01)
+  hparams.add_hparam("value_loss_coef", 1)
+  hparams.add_hparam("optimization_epochs", 15)
+  hparams.add_hparam("epoch_length", 200)
+  hparams.add_hparam("epochs_num", 2000)
+  return hparams
+
+
+@registry.register_hparams
+def pendulum_base():
+  hparams = ppo_base_v1()
+  hparams.add_hparam("network", feed_forward_gaussian_fun)
+  return hparams
+
+
+@registry.register_hparams
+def cartpole_base():
+  hparams = ppo_base_v1()
+  hparams.add_hparam("network", feed_forward_categorical_fun)
+  return hparams
+
+
+# Neural networks for actor-critic algorithms
+
+NetworkOutput = collections.namedtuple(
+    "NetworkOutput", "policy, value, action_postprocessing")
+
+
+def feed_forward_gaussian_fun(action_space, config, observations):
+  """Feed-forward Gaussian."""
+  if not isinstance(action_space, gym.spaces.box.Box):
+    raise ValueError("Expecting continuous action space.")
+
+  mean_weights_initializer = tf.contrib.layers.variance_scaling_initializer(
+      factor=config.init_mean_factor)
+  logstd_initializer = tf.random_normal_initializer(config.init_logstd, 1e-10)
+
+  flat_observations = tf.reshape(observations, [
+      tf.shape(observations)[0], tf.shape(observations)[1],
+      functools.reduce(operator.mul, observations.shape.as_list()[2:], 1)])
+
+  with tf.variable_scope("policy"):
+    x = flat_observations
+    for size in config.policy_layers:
+      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
+    mean = tf.contrib.layers.fully_connected(
+        x, action_space.shape[0], tf.tanh,
+        weights_initializer=mean_weights_initializer)
+    logstd = tf.get_variable(
+        "logstd", mean.shape[2:], tf.float32, logstd_initializer)
+    logstd = tf.tile(
+        logstd[None, None],
+        [tf.shape(mean)[0], tf.shape(mean)[1]] + [1] * (mean.shape.ndims - 2))
+  with tf.variable_scope("value"):
+    x = flat_observations
+    for size in config.value_layers:
+      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
+    value = tf.contrib.layers.fully_connected(x, 1, None)[..., 0]
+  mean = tf.check_numerics(mean, "mean")
+  logstd = tf.check_numerics(logstd, "logstd")
+  value = tf.check_numerics(value, "value")
+
+  policy = tf.contrib.distributions.MultivariateNormalDiag(mean,
+                                                           tf.exp(logstd))
+
+  return NetworkOutput(policy, value, lambda a: tf.clip_by_value(a, -2., 2))
+
+
+def feed_forward_categorical_fun(action_space, config, observations):
+  """Feed-forward categorical."""
+  if not isinstance(action_space, gym.spaces.Discrete):
+    raise ValueError("Expecting discrete action space.")
+  flat_observations = tf.reshape(observations, [
+      tf.shape(observations)[0], tf.shape(observations)[1],
+      functools.reduce(operator.mul, observations.shape.as_list()[2:], 1)])
+  with tf.variable_scope("policy"):
+    x = flat_observations
+    for size in config.policy_layers:
+      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
+    logits = tf.contrib.layers.fully_connected(x, action_space.n,
+                                               activation_fn=None)
+  with tf.variable_scope("value"):
+    x = flat_observations
+    for size in config.value_layers:
+      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
+    value = tf.contrib.layers.fully_connected(x, 1, None)[..., 0]
+  policy = tf.contrib.distributions.Categorical(logits=logits)
+  return NetworkOutput(policy, value, lambda a: a)
diff --git a/tensor2tensor/rl/README.md b/tensor2tensor/rl/README.md
index 053119f05..192b5d420 100644
--- a/tensor2tensor/rl/README.md
+++ b/tensor2tensor/rl/README.md
@@ -1,11 +1,12 @@
-# Tensor2Tensor experimental Reinforcement Learning.
+# Tensor2Tensor experimental Model-Based Reinforcement Learning.
 
 The rl package intention is to provide possiblity to run reinforcement
-algorithms within Tensorflow's computation graph. It's very experimental
+algorithms within Tensorflow's computation graph, in order to do model-based
+RL using envoronment models from Tensor2Tensor. It's very experimental
 for now and under heavy development.
 
 Currently the only supported algorithm is Proximy Policy Optimization - PPO.
 
 ## Sample usage - training in Pendulum-v0 environment.
 
-```python rl/t2t_rl_trainer.py```
+```python rl/t2t_rl_trainer.py --problems=Pendulum-v0 --hparams_set pendulum_base [--output_dir dir_location]```
diff --git a/tensor2tensor/rl/collect.py b/tensor2tensor/rl/collect.py
index afeefb72f..8df5138b7 100644
--- a/tensor2tensor/rl/collect.py
+++ b/tensor2tensor/rl/collect.py
@@ -18,26 +18,27 @@
 import tensorflow as tf
 
 
-def define_collect(policy_factory, batch_env, config):
+def define_collect(policy_factory, batch_env, hparams):
   """Collect trajectories."""
-  memory_shape = [config.epoch_length] + [batch_env.observ.shape.as_list()[0]]
+  memory_shape = [hparams.epoch_length] + [batch_env.observ.shape.as_list()[0]]
   memories_shapes_and_types = [
       # observation
       (memory_shape + [batch_env.observ.shape.as_list()[1]], tf.float32),
       (memory_shape, tf.float32),      # reward
       (memory_shape, tf.bool),         # done
-      (memory_shape + batch_env.action_shape, tf.float32),  # action
+      # action
+      (memory_shape + batch_env.action_shape, batch_env.action_dtype),
       (memory_shape, tf.float32),      # pdf
       (memory_shape, tf.float32),      # value function
   ]
   memory = [tf.Variable(tf.zeros(shape, dtype), trainable=False)
             for (shape, dtype) in memories_shapes_and_types]
   cumulative_rewards = tf.Variable(
-      tf.zeros(config.num_agents, tf.float32), trainable=False)
+      tf.zeros(hparams.num_agents, tf.float32), trainable=False)
 
   should_reset_var = tf.Variable(True, trainable=False)
   reset_op = tf.cond(should_reset_var,
-                     lambda: batch_env.reset(tf.range(config.num_agents)),
+                     lambda: batch_env.reset(tf.range(hparams.num_agents)),
                      lambda: 0.0)
   with tf.control_dependencies([reset_op]):
     reset_once_op = tf.assign(should_reset_var, False)
@@ -58,7 +59,7 @@ def step(index, scores_sum, scores_num):
       pdf = policy.prob(action)[0]
       with tf.control_dependencies(simulate_output):
         reward, done = simulate_output
-        done = tf.reshape(done, (config.num_agents,))
+        done = tf.reshape(done, (hparams.num_agents,))
         to_save = [obs_copy, reward, done, action[0, ...], pdf,
                    actor_critic.value[0]]
         save_ops = [tf.scatter_update(memory_slot, index, value)
@@ -82,7 +83,7 @@ def step(index, scores_sum, scores_num):
 
     init = [tf.constant(0), tf.constant(0.0), tf.constant(0)]
     index, scores_sum, scores_num = tf.while_loop(
-        lambda c, _1, _2: c < config.epoch_length,
+        lambda c, _1, _2: c < hparams.epoch_length,
         step,
         init,
         parallel_iterations=1,
@@ -91,5 +92,9 @@ def step(index, scores_sum, scores_num):
                        lambda: scores_sum / tf.cast(scores_num, tf.float32),
                        lambda: 0.)
   printing = tf.Print(0, [mean_score, scores_sum, scores_num], "mean_score: ")
-  with tf.control_dependencies([printing]):
-    return tf.identity(index), memory
+  with tf.control_dependencies([index, printing]):
+    memory = [tf.identity(mem) for mem in memory]
+    summaries = tf.summary.merge(
+        [tf.summary.scalar("mean_score_this_iter", mean_score),
+         tf.summary.scalar("episodes_finished_this_iter", scores_num)])
+    return memory, summaries
diff --git a/tensor2tensor/rl/networks.py b/tensor2tensor/rl/networks.py
deleted file mode 100644
index e094abe66..000000000
--- a/tensor2tensor/rl/networks.py
+++ /dev/null
@@ -1,69 +0,0 @@
-# coding=utf-8
-# Copyright 2018 The Tensor2Tensor Authors.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-"""Neural networks for actor-critic algorithms."""
-
-import collections
-import functools
-import operator
-
-# Dependency imports
-
-import gym
-import tensorflow as tf
-
-
-NetworkOutput = collections.namedtuple(
-    'NetworkOutput', 'policy, value, action_postprocessing')
-
-
-def feed_forward_gaussian_fun(observation_space, action_space, config,
-                              observations):
-  """Feed-forward gaussian."""
-  assert isinstance(observation_space, gym.spaces.box.Box)
-
-  mean_weights_initializer = tf.contrib.layers.variance_scaling_initializer(
-      factor=config.init_mean_factor)
-  logstd_initializer = tf.random_normal_initializer(config.init_logstd, 1e-10)
-
-  flat_observations = tf.reshape(observations, [
-      tf.shape(observations)[0], tf.shape(observations)[1],
-      functools.reduce(operator.mul, observations.shape.as_list()[2:], 1)])
-
-  with tf.variable_scope('policy'):
-    x = flat_observations
-    for size in config.policy_layers:
-      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
-    mean = tf.contrib.layers.fully_connected(
-        x, action_space.shape[0], tf.tanh,
-        weights_initializer=mean_weights_initializer)
-    logstd = tf.get_variable(
-        'logstd', mean.shape[2:], tf.float32, logstd_initializer)
-    logstd = tf.tile(
-        logstd[None, None],
-        [tf.shape(mean)[0], tf.shape(mean)[1]] + [1] * (mean.shape.ndims - 2))
-  with tf.variable_scope('value'):
-    x = flat_observations
-    for size in config.value_layers:
-      x = tf.contrib.layers.fully_connected(x, size, tf.nn.relu)
-    value = tf.contrib.layers.fully_connected(x, 1, None)[..., 0]
-  mean = tf.check_numerics(mean, 'mean')
-  logstd = tf.check_numerics(logstd, 'logstd')
-  value = tf.check_numerics(value, 'value')
-
-  policy = tf.contrib.distributions.MultivariateNormalDiag(mean,
-                                                           tf.exp(logstd))
-
-  return NetworkOutput(policy, value, lambda a: tf.clip_by_value(a, -2., 2))
diff --git a/tensor2tensor/rl/ppo.py b/tensor2tensor/rl/ppo.py
index 7c26cd074..31cea0486 100644
--- a/tensor2tensor/rl/ppo.py
+++ b/tensor2tensor/rl/ppo.py
@@ -18,7 +18,6 @@
 Based on: https://arxiv.org/abs/1707.06347
 """
 
-
 import tensorflow as tf
 
 
@@ -32,8 +31,8 @@ def define_ppo_step(observation, action, reward, done, value, old_pdf,
   clipped_ratio = tf.clip_by_value(ratio, 1 - config.clipping_coef,
                                    1 + config.clipping_coef)
 
-  advantage = calculate_discounted_return(
-      reward, value, done, config.gae_gamma, config.gae_lambda) - value
+  advantage = calculate_generalized_advantage_estimator(
+      reward, value, done, config.gae_gamma, config.gae_lambda)
 
   advantage_mean, advantage_variance = tf.nn.moments(advantage, axes=[0, 1],
                                                      keep_dims=True)
@@ -53,7 +52,11 @@ def define_ppo_step(observation, action, reward, done, value, old_pdf,
 
   total_loss = policy_loss + value_loss + entropy_loss
 
-  optimization_op = config.optimizer(config.learning_rate).minimize(total_loss)
+  optimization_op = tf.contrib.layers.optimize_loss(
+      loss=total_loss,
+      global_step=tf.train.get_or_create_global_step(),
+      optimizer=config.optimizer,
+      learning_rate=config.learning_rate)
 
   with tf.control_dependencies([optimization_op]):
     return [tf.identity(x) for x in (policy_loss, value_loss, entropy_loss)]
@@ -79,12 +82,20 @@ def define_ppo_epoch(memory, policy_factory, config):
       [0., 0., 0.],
       parallel_iterations=1)
 
-  print_losses = tf.group(
-      tf.Print(0, [tf.reduce_mean(policy_loss)], 'policy loss: '),
-      tf.Print(0, [tf.reduce_mean(value_loss)], 'value loss: '),
-      tf.Print(0, [tf.reduce_mean(entropy_loss)], 'entropy loss: '))
+  summaries = [tf.summary.scalar("policy loss", tf.reduce_mean(policy_loss)),
+               tf.summary.scalar("value loss", tf.reduce_mean(value_loss)),
+               tf.summary.scalar("entropy loss", tf.reduce_mean(entropy_loss))]
+
+  losses_summary = tf.summary.merge(summaries)
 
-  return print_losses
+  losses_summary = tf.Print(losses_summary,
+                            [tf.reduce_mean(policy_loss)], "policy loss: ")
+  losses_summary = tf.Print(losses_summary,
+                            [tf.reduce_mean(value_loss)], "value loss: ")
+  losses_summary = tf.Print(losses_summary,
+                            [tf.reduce_mean(entropy_loss)], "entropy loss: ")
+
+  return losses_summary
 
 
 def calculate_discounted_return(reward, value, done, discount, unused_lambda):
@@ -99,4 +110,23 @@ def calculate_discounted_return(reward, value, done, discount, unused_lambda):
       tf.zeros_like(reward[0, :]),  # initializer
       1,
       False), [0])
-  return tf.check_numerics(return_, 'return')
+  return tf.check_numerics(return_, "return")
+
+
+def calculate_generalized_advantage_estimator(
+    reward, value, done, gae_gamma, gae_lambda):
+  """Generalized advantage estimator."""
+  # Below is slight weirdness, we set the last reward to 0.
+  # This makes the adventantage to be 0 in the last timestep.
+  reward = tf.concat([reward[:-1, :], value[-1:, :]], axis=0)
+  next_value = tf.concat([value[1:, :], tf.zeros_like(value[-1:, :])], axis=0)
+  next_not_done = 1 - tf.cast(tf.concat(
+      [done[1:, :], tf.zeros_like(done[-1:, :])], axis=0), tf.float32)
+  delta = reward + gae_gamma * next_value * next_not_done - value
+
+  return_ = tf.reverse(tf.scan(
+      lambda agg, cur: cur[0] + cur[1] * gae_gamma * gae_lambda * agg,
+      [tf.reverse(delta, [0]), tf.reverse(next_not_done, [0])],
+      tf.zeros_like(delta[0, :]),
+      1, False), [0])
+  return tf.check_numerics(tf.stop_gradient(return_), "return")
diff --git a/tensor2tensor/rl/rl_trainer_lib.py b/tensor2tensor/rl/rl_trainer_lib.py
index ee7b2cc83..610607a46 100644
--- a/tensor2tensor/rl/rl_trainer_lib.py
+++ b/tensor2tensor/rl/rl_trainer_lib.py
@@ -13,7 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""Training of RL agent with PPO algorithm."""
+"""Library for training of RL agent with PPO algorithm."""
 
 from __future__ import absolute_import
 
@@ -22,66 +22,50 @@
 # Dependency imports
 
 import gym
+
+
+from tensor2tensor import models  # pylint: disable=unused-import
+from tensor2tensor.models import rl  # pylint: disable=unused-import
 from tensor2tensor.rl import collect
-from tensor2tensor.rl import networks
 from tensor2tensor.rl import ppo
 from tensor2tensor.rl.envs import utils
 
 import tensorflow as tf
 
 
-def define_train(policy_lambda, env_lambda, config):
+def define_train(hparams, environment_name):
   """Define the training setup."""
+  env_lambda = lambda: gym.make(environment_name)
+  policy_lambda = hparams.network
   env = env_lambda()
   action_space = env.action_space
-  observation_space = env.observation_space
 
-  batch_env = utils.define_batch_env(env_lambda, config.num_agents)
+  batch_env = utils.define_batch_env(env_lambda, hparams.num_agents)
 
   policy_factory = tf.make_template(
       "network",
-      functools.partial(policy_lambda, observation_space,
-                        action_space, config))
+      functools.partial(policy_lambda, action_space, hparams))
 
-  (collect_op, memory) = collect.define_collect(
-      policy_factory, batch_env, config)
+  memory, collect_summary = collect.define_collect(policy_factory,
+                                                   batch_env, hparams)
+  ppo_summary = ppo.define_ppo_epoch(memory, policy_factory, hparams)
+  summary = tf.summary.merge([collect_summary, ppo_summary])
 
-  with tf.control_dependencies([collect_op]):
-    ppo_op = ppo.define_ppo_epoch(memory, policy_factory, config)
+  return summary
 
-  return ppo_op
 
+def train(hparams, environment_name, event_dir=None):
+  summary_op = define_train(hparams, environment_name)
 
-def train(params):
-  policy_lambda, env_lambda, config = params
-  ppo_op = define_train(policy_lambda, env_lambda, config)
+  if event_dir:
+    summary_writer = tf.summary.FileWriter(
+        event_dir, graph=tf.get_default_graph(), flush_secs=60)
+  else:
+    summary_writer = None
 
   with tf.Session() as sess:
     sess.run(tf.global_variables_initializer())
-    for _ in range(config.epochs_num):
-      sess.run(ppo_op)
-
-
-def example_params():
-  """Example hyperparameters."""
-  config = tf.contrib.training.HParams(
-      init_mean_factor=0.1,
-      init_logstd=0.1,
-      policy_layers=(100, 100),
-      value_layers=(100, 100),
-      num_agents=30,
-      clipping_coef=0.2,
-      gae_gamma=0.99,
-      gae_lambda=0.95,
-      entropy_loss_coef=0.01,
-      value_loss_coef=1,
-      optimizer=tf.train.AdamOptimizer,
-      learning_rate=1e-4,
-      optimization_epochs=15,
-      epoch_length=200,
-      epochs_num=2000)
-  return networks.feed_forward_gaussian_fun, pendulum_lambda, config
-
-
-def pendulum_lambda():
-  return gym.make("Pendulum-v0")
+    for epoch_index in range(hparams.epochs_num):
+      summary = sess.run(summary_op)
+      if summary_writer:
+        summary_writer.add_summary(summary, epoch_index)
diff --git a/tensor2tensor/rl/rl_trainer_lib_test.py b/tensor2tensor/rl/rl_trainer_lib_test.py
index 0b753a580..9e9b4fd4a 100644
--- a/tensor2tensor/rl/rl_trainer_lib_test.py
+++ b/tensor2tensor/rl/rl_trainer_lib_test.py
@@ -18,6 +18,7 @@
 # Dependency imports
 
 from tensor2tensor.rl import rl_trainer_lib
+from tensor2tensor.utils import trainer_lib
 
 import tensorflow as tf
 
@@ -25,10 +26,13 @@
 class TrainTest(tf.test.TestCase):
 
   def test_no_crash_pendulum(self):
-    params = rl_trainer_lib.example_params()
-    params[2].epochs_num = 10
-    rl_trainer_lib.train(params)
+    hparams = trainer_lib.create_hparams("pendulum_base", "epochs_num=10")
+    rl_trainer_lib.train(hparams, "Pendulum-v0")
 
+  def test_no_crash_cartpole(self):
+    hparams = trainer_lib.create_hparams("cartpole_base", "epochs_num=10")
+    rl_trainer_lib.train(hparams, "CartPole-v0")
 
-if __name__ == '__main__':
+
+if __name__ == "__main__":
   tf.test.main()
diff --git a/tensor2tensor/rl/t2t_rl_trainer.py b/tensor2tensor/rl/t2t_rl_trainer.py
index 6bdbcab06..3abe22723 100644
--- a/tensor2tensor/rl/t2t_rl_trainer.py
+++ b/tensor2tensor/rl/t2t_rl_trainer.py
@@ -18,12 +18,25 @@
 # Dependency imports
 
 from tensor2tensor.rl import rl_trainer_lib
+from tensor2tensor.utils import flags as t2t_flags  # pylint: disable=unused-import
+from tensor2tensor.utils import trainer_lib
 
 import tensorflow as tf
 
+flags = tf.flags
+FLAGS = flags.FLAGS
+
+# To maintain compatibility with some internal libs, we guard against these flag
+# definitions possibly erroring. Apologies for the ugliness.
+try:
+  flags.DEFINE_string("output_dir", "", "Base output directory for run.")
+except:  # pylint: disable=bare-except
+  pass
+
 
 def main(_):
-  rl_trainer_lib.train(rl_trainer_lib.example_params())
+  hparams = trainer_lib.create_hparams(FLAGS.hparams_set, FLAGS.hparams)
+  rl_trainer_lib.train(hparams, FLAGS.problems, FLAGS.output_dir)
 
 
 if __name__ == "__main__":

From 57a27fec1274e8701e7bfdb93695dd99120c3248 Mon Sep 17 00:00:00 2001
From: Lukasz Kaiser <lukaszkaiser@google.com>
Date: Sat, 10 Feb 2018 14:37:09 -0800
Subject: [PATCH 10/14] Increase release number.

PiperOrigin-RevId: 185267399
---
 setup.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/setup.py b/setup.py
index 1153dbba8..12dcafa2f 100644
--- a/setup.py
+++ b/setup.py
@@ -5,7 +5,7 @@
 
 setup(
     name='tensor2tensor',
-    version='1.4.4',
+    version='2.0.0',
     description='Tensor2Tensor',
     author='Google Inc.',
     author_email='no-reply@google.com',

From 70a046477303e78d2cf7f77ffc537ea64511095e Mon Sep 17 00:00:00 2001
From: Lukasz Kaiser <lukaszkaiser@google.com>
Date: Sun, 11 Feb 2018 12:11:19 -0800
Subject: [PATCH 11/14] Move more researchy models under models/research.

PiperOrigin-RevId: 185311067
---
 tensor2tensor/models/__init__.py              | 23 ++++++++++---------
 .../models/{ => research}/aligned.py          |  0
 .../models/{ => research}/attention_lm.py     |  0
 .../models/{ => research}/attention_lm_moe.py |  0
 .../models/{ => research}/cycle_gan.py        |  2 +-
 .../models/{ => research}/gene_expression.py  |  0
 .../{ => research}/gene_expression_test.py    |  2 +-
 .../models/{ => research}/multimodel.py       |  0
 .../models/{ => research}/multimodel_test.py  |  2 +-
 tensor2tensor/models/{ => research}/rl.py     |  0
 .../models/{ => research}/super_lm.py         |  0
 .../models/{ => research}/transformer_moe.py  |  0
 .../{ => research}/transformer_revnet.py      |  0
 .../{ => research}/transformer_revnet_test.py |  2 +-
 .../{ => research}/transformer_sketch.py      |  2 +-
 .../models/{ => research}/transformer_vae.py  |  0
 tensor2tensor/rl/rl_trainer_lib.py            |  2 +-
 17 files changed, 18 insertions(+), 17 deletions(-)
 rename tensor2tensor/models/{ => research}/aligned.py (100%)
 rename tensor2tensor/models/{ => research}/attention_lm.py (100%)
 rename tensor2tensor/models/{ => research}/attention_lm_moe.py (100%)
 rename tensor2tensor/models/{ => research}/cycle_gan.py (98%)
 rename tensor2tensor/models/{ => research}/gene_expression.py (100%)
 rename tensor2tensor/models/{ => research}/gene_expression_test.py (97%)
 rename tensor2tensor/models/{ => research}/multimodel.py (100%)
 rename tensor2tensor/models/{ => research}/multimodel_test.py (97%)
 rename tensor2tensor/models/{ => research}/rl.py (100%)
 rename tensor2tensor/models/{ => research}/super_lm.py (100%)
 rename tensor2tensor/models/{ => research}/transformer_moe.py (100%)
 rename tensor2tensor/models/{ => research}/transformer_revnet.py (100%)
 rename tensor2tensor/models/{ => research}/transformer_revnet_test.py (97%)
 rename tensor2tensor/models/{ => research}/transformer_sketch.py (98%)
 rename tensor2tensor/models/{ => research}/transformer_vae.py (100%)

diff --git a/tensor2tensor/models/__init__.py b/tensor2tensor/models/__init__.py
index 2ed6ca883..58badcb4e 100644
--- a/tensor2tensor/models/__init__.py
+++ b/tensor2tensor/models/__init__.py
@@ -23,26 +23,27 @@
 # pylint: disable=unused-import
 
 from tensor2tensor.layers import modalities
-from tensor2tensor.models import aligned
-from tensor2tensor.models import attention_lm
-from tensor2tensor.models import attention_lm_moe
 from tensor2tensor.models import basic
 from tensor2tensor.models import bytenet
-from tensor2tensor.models import cycle_gan
-from tensor2tensor.models import gene_expression
 from tensor2tensor.models import lstm
-from tensor2tensor.models import multimodel
 from tensor2tensor.models import neural_gpu
 from tensor2tensor.models import resnet
 from tensor2tensor.models import revnet
 from tensor2tensor.models import shake_shake
 from tensor2tensor.models import slicenet
-from tensor2tensor.models import super_lm
 from tensor2tensor.models import transformer
-from tensor2tensor.models import transformer_moe
-from tensor2tensor.models import transformer_revnet
-from tensor2tensor.models import transformer_sketch
-from tensor2tensor.models import transformer_vae
 from tensor2tensor.models import vanilla_gan
 from tensor2tensor.models import xception
+
+from tensor2tensor.models.research import aligned
+from tensor2tensor.models.research import attention_lm
+from tensor2tensor.models.research import attention_lm_moe
+from tensor2tensor.models.research import cycle_gan
+from tensor2tensor.models.research import gene_expression
+from tensor2tensor.models.research import multimodel
+from tensor2tensor.models.research import super_lm
+from tensor2tensor.models.research import transformer_moe
+from tensor2tensor.models.research import transformer_revnet
+from tensor2tensor.models.research import transformer_sketch
+from tensor2tensor.models.research import transformer_vae
 # pylint: enable=unused-import
diff --git a/tensor2tensor/models/aligned.py b/tensor2tensor/models/research/aligned.py
similarity index 100%
rename from tensor2tensor/models/aligned.py
rename to tensor2tensor/models/research/aligned.py
diff --git a/tensor2tensor/models/attention_lm.py b/tensor2tensor/models/research/attention_lm.py
similarity index 100%
rename from tensor2tensor/models/attention_lm.py
rename to tensor2tensor/models/research/attention_lm.py
diff --git a/tensor2tensor/models/attention_lm_moe.py b/tensor2tensor/models/research/attention_lm_moe.py
similarity index 100%
rename from tensor2tensor/models/attention_lm_moe.py
rename to tensor2tensor/models/research/attention_lm_moe.py
diff --git a/tensor2tensor/models/cycle_gan.py b/tensor2tensor/models/research/cycle_gan.py
similarity index 98%
rename from tensor2tensor/models/cycle_gan.py
rename to tensor2tensor/models/research/cycle_gan.py
index 5f964ef6a..55cbe350d 100644
--- a/tensor2tensor/models/cycle_gan.py
+++ b/tensor2tensor/models/research/cycle_gan.py
@@ -22,7 +22,7 @@
 # Dependency imports
 
 from tensor2tensor.layers import common_layers
-from tensor2tensor.models import transformer_vae
+from tensor2tensor.models.research import transformer_vae
 from tensor2tensor.utils import registry
 from tensor2tensor.utils import t2t_model
 
diff --git a/tensor2tensor/models/gene_expression.py b/tensor2tensor/models/research/gene_expression.py
similarity index 100%
rename from tensor2tensor/models/gene_expression.py
rename to tensor2tensor/models/research/gene_expression.py
diff --git a/tensor2tensor/models/gene_expression_test.py b/tensor2tensor/models/research/gene_expression_test.py
similarity index 97%
rename from tensor2tensor/models/gene_expression_test.py
rename to tensor2tensor/models/research/gene_expression_test.py
index ae16b3383..899defadb 100644
--- a/tensor2tensor/models/gene_expression_test.py
+++ b/tensor2tensor/models/research/gene_expression_test.py
@@ -24,7 +24,7 @@
 
 from tensor2tensor.data_generators import gene_expression as gene_data
 from tensor2tensor.layers import modalities  # pylint: disable=unused-import
-from tensor2tensor.models import gene_expression
+from tensor2tensor.models.research import gene_expression
 
 import tensorflow as tf
 
diff --git a/tensor2tensor/models/multimodel.py b/tensor2tensor/models/research/multimodel.py
similarity index 100%
rename from tensor2tensor/models/multimodel.py
rename to tensor2tensor/models/research/multimodel.py
diff --git a/tensor2tensor/models/multimodel_test.py b/tensor2tensor/models/research/multimodel_test.py
similarity index 97%
rename from tensor2tensor/models/multimodel_test.py
rename to tensor2tensor/models/research/multimodel_test.py
index ff233775a..ef8e30138 100644
--- a/tensor2tensor/models/multimodel_test.py
+++ b/tensor2tensor/models/research/multimodel_test.py
@@ -24,7 +24,7 @@
 import numpy as np
 
 from tensor2tensor.data_generators import cifar  # pylint: disable=unused-import
-from tensor2tensor.models import multimodel
+from tensor2tensor.models.research import multimodel
 from tensor2tensor.utils import registry
 
 import tensorflow as tf
diff --git a/tensor2tensor/models/rl.py b/tensor2tensor/models/research/rl.py
similarity index 100%
rename from tensor2tensor/models/rl.py
rename to tensor2tensor/models/research/rl.py
diff --git a/tensor2tensor/models/super_lm.py b/tensor2tensor/models/research/super_lm.py
similarity index 100%
rename from tensor2tensor/models/super_lm.py
rename to tensor2tensor/models/research/super_lm.py
diff --git a/tensor2tensor/models/transformer_moe.py b/tensor2tensor/models/research/transformer_moe.py
similarity index 100%
rename from tensor2tensor/models/transformer_moe.py
rename to tensor2tensor/models/research/transformer_moe.py
diff --git a/tensor2tensor/models/transformer_revnet.py b/tensor2tensor/models/research/transformer_revnet.py
similarity index 100%
rename from tensor2tensor/models/transformer_revnet.py
rename to tensor2tensor/models/research/transformer_revnet.py
diff --git a/tensor2tensor/models/transformer_revnet_test.py b/tensor2tensor/models/research/transformer_revnet_test.py
similarity index 97%
rename from tensor2tensor/models/transformer_revnet_test.py
rename to tensor2tensor/models/research/transformer_revnet_test.py
index 4ea19a0c9..b3d4c0812 100644
--- a/tensor2tensor/models/transformer_revnet_test.py
+++ b/tensor2tensor/models/research/transformer_revnet_test.py
@@ -24,7 +24,7 @@
 import numpy as np
 
 from tensor2tensor.data_generators import problem_hparams
-from tensor2tensor.models import transformer_revnet
+from tensor2tensor.models.research import transformer_revnet
 
 import tensorflow as tf
 
diff --git a/tensor2tensor/models/transformer_sketch.py b/tensor2tensor/models/research/transformer_sketch.py
similarity index 98%
rename from tensor2tensor/models/transformer_sketch.py
rename to tensor2tensor/models/research/transformer_sketch.py
index 0a1442d10..c32a28493 100644
--- a/tensor2tensor/models/transformer_sketch.py
+++ b/tensor2tensor/models/research/transformer_sketch.py
@@ -24,7 +24,7 @@
 
 from tensor2tensor.layers import common_layers
 from tensor2tensor.models import transformer
-from tensor2tensor.models import transformer_vae
+from tensor2tensor.models.research import transformer_vae
 from tensor2tensor.models.transformer import transformer_base
 from tensor2tensor.models.transformer import transformer_small
 from tensor2tensor.utils import registry
diff --git a/tensor2tensor/models/transformer_vae.py b/tensor2tensor/models/research/transformer_vae.py
similarity index 100%
rename from tensor2tensor/models/transformer_vae.py
rename to tensor2tensor/models/research/transformer_vae.py
diff --git a/tensor2tensor/rl/rl_trainer_lib.py b/tensor2tensor/rl/rl_trainer_lib.py
index 610607a46..2cb718771 100644
--- a/tensor2tensor/rl/rl_trainer_lib.py
+++ b/tensor2tensor/rl/rl_trainer_lib.py
@@ -25,7 +25,7 @@
 
 
 from tensor2tensor import models  # pylint: disable=unused-import
-from tensor2tensor.models import rl  # pylint: disable=unused-import
+from tensor2tensor.models.research import rl  # pylint: disable=unused-import
 from tensor2tensor.rl import collect
 from tensor2tensor.rl import ppo
 from tensor2tensor.rl.envs import utils

From 08f0742b06d90662f1d15971dc276e3b3a847862 Mon Sep 17 00:00:00 2001
From: Lukasz Kaiser <lukaszkaiser@google.com>
Date: Sun, 11 Feb 2018 12:40:29 -0800
Subject: [PATCH 12/14] Add a quick MNIST model for README, comments and
 documentation, bump back version.

PiperOrigin-RevId: 185311943
---
 README.md                           | 20 +++++++++++++++++---
 docs/walkthrough.md                 | 20 +++++++++++++++++---
 setup.py                            |  2 +-
 tensor2tensor/bin/t2t-datagen       | 13 ++++++++++++-
 tensor2tensor/bin/t2t-trainer       | 18 +++++++++++++++++-
 tensor2tensor/models/shake_shake.py | 10 ++++++++++
 tensor2tensor/models/transformer.py | 10 +++++++---
 7 files changed, 81 insertions(+), 12 deletions(-)

diff --git a/README.md b/README.md
index de8de0bfe..1f32351ac 100644
--- a/README.md
+++ b/README.md
@@ -19,7 +19,8 @@ or running existing ones on your data. It is actively used and maintained by
 researchers and engineers within
 the [Google Brain team](https://research.google.com/teams/brain/) and was used
 to develop state-of-the-art models for translation (see
-[Attention Is All You Need](https://arxiv.org/abs/1706.03762)), summarization,
+[Attention Is All You Need](https://arxiv.org/abs/1706.03762)),
+[summarization](https://arxiv.org/abs/1801.10198),
 image generation and other tasks. You can read
 more about T2T in the [Google Research Blog post introducing
 it](https://research.googleblog.com/2017/06/accelerating-deep-learning-research.html).
@@ -42,7 +43,20 @@ with T2T announcements.
 browser using a free VM from Google, no installation needed.
 
 Alternatively, here is a one-command version that installs T2T, downloads data,
-trains an English-German translation model, and evaluates it:
+trains an MNIST model and evaluates it:
+
+```
+pip install tensor2tensor && t2t-trainer \
+  --generate_data \
+  --data_dir=~/t2t_data \
+  --problems=image_mnist \
+  --model=shake_shake \
+  --hparams_set=shake_shake_quick \
+  --output_dir=~/t2t_train/mnist1
+```
+
+For a more demanding problem, here is how to train
+an English-German translation model and evaluate it:
 
 ```
 pip install tensor2tensor && t2t-trainer \
@@ -54,7 +68,7 @@ pip install tensor2tensor && t2t-trainer \
   --output_dir=~/t2t_train/base
 ```
 
-You can decode from the model interactively:
+You can decode from the model interactively to get translations:
 
 ```
 t2t-decoder \
diff --git a/docs/walkthrough.md b/docs/walkthrough.md
index de8de0bfe..1f32351ac 100644
--- a/docs/walkthrough.md
+++ b/docs/walkthrough.md
@@ -19,7 +19,8 @@ or running existing ones on your data. It is actively used and maintained by
 researchers and engineers within
 the [Google Brain team](https://research.google.com/teams/brain/) and was used
 to develop state-of-the-art models for translation (see
-[Attention Is All You Need](https://arxiv.org/abs/1706.03762)), summarization,
+[Attention Is All You Need](https://arxiv.org/abs/1706.03762)),
+[summarization](https://arxiv.org/abs/1801.10198),
 image generation and other tasks. You can read
 more about T2T in the [Google Research Blog post introducing
 it](https://research.googleblog.com/2017/06/accelerating-deep-learning-research.html).
@@ -42,7 +43,20 @@ with T2T announcements.
 browser using a free VM from Google, no installation needed.
 
 Alternatively, here is a one-command version that installs T2T, downloads data,
-trains an English-German translation model, and evaluates it:
+trains an MNIST model and evaluates it:
+
+```
+pip install tensor2tensor && t2t-trainer \
+  --generate_data \
+  --data_dir=~/t2t_data \
+  --problems=image_mnist \
+  --model=shake_shake \
+  --hparams_set=shake_shake_quick \
+  --output_dir=~/t2t_train/mnist1
+```
+
+For a more demanding problem, here is how to train
+an English-German translation model and evaluate it:
 
 ```
 pip install tensor2tensor && t2t-trainer \
@@ -54,7 +68,7 @@ pip install tensor2tensor && t2t-trainer \
   --output_dir=~/t2t_train/base
 ```
 
-You can decode from the model interactively:
+You can decode from the model interactively to get translations:
 
 ```
 t2t-decoder \
diff --git a/setup.py b/setup.py
index 12dcafa2f..94f58130c 100644
--- a/setup.py
+++ b/setup.py
@@ -5,7 +5,7 @@
 
 setup(
     name='tensor2tensor',
-    version='2.0.0',
+    version='1.5.0',
     description='Tensor2Tensor',
     author='Google Inc.',
     author_email='no-reply@google.com',
diff --git a/tensor2tensor/bin/t2t-datagen b/tensor2tensor/bin/t2t-datagen
index 4290365b6..332f063ad 100755
--- a/tensor2tensor/bin/t2t-datagen
+++ b/tensor2tensor/bin/t2t-datagen
@@ -1,5 +1,16 @@
 #!/usr/bin/env python
-"""t2t-datagen."""
+"""Data generation for Tensor2Tensor.
+
+This script is used to generate data to train your models
+for a number problems for which open-source data is available.
+
+For example, to generate data for MNIST run this:
+
+t2t-datagen \
+  --problem=image_mnist \
+  --data_dir=~/t2t_data \
+  --tmp_dir=~/t2t_data/tmp
+"""
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
diff --git a/tensor2tensor/bin/t2t-trainer b/tensor2tensor/bin/t2t-trainer
index 77f1ec865..01c2dcb56 100755
--- a/tensor2tensor/bin/t2t-trainer
+++ b/tensor2tensor/bin/t2t-trainer
@@ -1,5 +1,21 @@
 #!/usr/bin/env python
-"""t2t-trainer."""
+"""Trainer for Tensor2Tensor.
+
+This script is used to train your models in Tensor2Tensor.
+
+For example, to train a shake-shake model on MNIST run this:
+
+t2t-trainer \
+  --generate_data \
+  --problems=image_mnist \
+  --data_dir=~/t2t_data \
+  --tmp_dir=~/t2t_data/tmp
+  --model=shake_shake \
+  --hparams_set=shake_shake_quick \
+  --output_dir=~/t2t_train/mnist1 \
+  --train_steps=1000 \
+  --eval_steps=100
+"""
 from __future__ import absolute_import
 from __future__ import division
 from __future__ import print_function
diff --git a/tensor2tensor/models/shake_shake.py b/tensor2tensor/models/shake_shake.py
index 6adbe5b70..c5a6caf04 100644
--- a/tensor2tensor/models/shake_shake.py
+++ b/tensor2tensor/models/shake_shake.py
@@ -185,6 +185,16 @@ def shakeshake_small():
   return hparams
 
 
+@registry.register_hparams
+def shake_shake_quick():
+  hparams = shakeshake_small()
+  hparams.optimizer = "Adam"
+  hparams.learning_rate_cosine_cycle_steps = 1000
+  hparams.learning_rate = 0.5
+  hparams.batch_size = 100
+  return hparams
+
+
 @registry.register_hparams
 def shakeshake_big():
   hparams = shakeshake_small()
diff --git a/tensor2tensor/models/transformer.py b/tensor2tensor/models/transformer.py
index c67a5fb4d..05795323e 100644
--- a/tensor2tensor/models/transformer.py
+++ b/tensor2tensor/models/transformer.py
@@ -13,10 +13,14 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-"""transformer (attention).
+"""Transformer model from "Attention Is All You Need".
 
-encoder: [Self-Attention, Feed-forward] x n
-decoder: [Self-Attention, Source-Target-Attention, Feed-forward] x n
+The Transformer model consists of an encoder and a decoder. Both are stacks
+of self-attention layers followed by feed-forward layers. This model yields
+good results on a number of problems, especially in NLP and machine translation.
+
+See "Attention Is All You Need" (https://arxiv.org/abs/1706.03762) for the full
+description of the model and the results obtained with its early version.
 """
 
 from __future__ import absolute_import

From 30a79e83fd937c47adce98422d98947b3c3be441 Mon Sep 17 00:00:00 2001
From: Ryan Sepassi <rsepassi@google.com>
Date: Sun, 11 Feb 2018 13:40:48 -0800
Subject: [PATCH 13/14] Add models/research/__init__.py

PiperOrigin-RevId: 185314217
---
 tensor2tensor/models/research/__init__.py | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
 create mode 100644 tensor2tensor/models/research/__init__.py

diff --git a/tensor2tensor/models/research/__init__.py b/tensor2tensor/models/research/__init__.py
new file mode 100644
index 000000000..4bd418a74
--- /dev/null
+++ b/tensor2tensor/models/research/__init__.py
@@ -0,0 +1,15 @@
+# coding=utf-8
+# Copyright 2018 The Tensor2Tensor Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+

From 88a3c9b5b3f34d5f1ed43e685ac4ae897000500c Mon Sep 17 00:00:00 2001
From: Lukasz Kaiser <lukaszkaiser@google.com>
Date: Sun, 11 Feb 2018 15:47:10 -0800
Subject: [PATCH 14/14] Extend Cloud TPU docs.

PiperOrigin-RevId: 185318516
---
 README.md           | 245 +++++++++++++++++++++++++-------------------
 docs/cloud_tpu.md   |  54 ++++++++--
 docs/index.md       | 116 ++++++++++++++++++---
 docs/walkthrough.md | 245 +++++++++++++++++++++++++-------------------
 4 files changed, 430 insertions(+), 230 deletions(-)

diff --git a/README.md b/README.md
index 1f32351ac..a569fb80a 100644
--- a/README.md
+++ b/README.md
@@ -12,87 +12,50 @@ welcome](https://img.shields.io/badge/contributions-welcome-brightgreen.svg)](CO
 
 [Tensor2Tensor](https://github.com/tensorflow/tensor2tensor), or
 [T2T](https://github.com/tensorflow/tensor2tensor) for short, is a library
-of deep learning models and datasets. It has binaries to train the models and
-to download and prepare the data for you. T2T is modular and extensible and can
-be used in [notebooks](https://goo.gl/wkHexj) for prototyping your own models
-or running existing ones on your data. It is actively used and maintained by
-researchers and engineers within
-the [Google Brain team](https://research.google.com/teams/brain/) and was used
-to develop state-of-the-art models for translation (see
-[Attention Is All You Need](https://arxiv.org/abs/1706.03762)),
-[summarization](https://arxiv.org/abs/1801.10198),
-image generation and other tasks. You can read
-more about T2T in the [Google Research Blog post introducing
-it](https://research.googleblog.com/2017/06/accelerating-deep-learning-research.html).
-
-We're eager to collaborate with you on extending T2T, so please feel
-free to [open an issue on
-GitHub](https://github.com/tensorflow/tensor2tensor/issues) or
-send along a pull request to add your dataset or model.
-See [our contribution
-doc](CONTRIBUTING.md) for details and our [open
-issues](https://github.com/tensorflow/tensor2tensor/issues).
-You can chat with us and other users on
-[Gitter](https://gitter.im/tensor2tensor/Lobby) and please join our
-[Google Group](https://groups.google.com/forum/#!forum/tensor2tensor) to keep up
-with T2T announcements.
+of deep learning models and datasets designed to [accelerate deep learning
+research](https://research.googleblog.com/2017/06/accelerating-deep-learning-research.html) and make it more accessible.
+
+T2T is actively used and maintained by researchers and engineers within the
+[Google Brain team](https://research.google.com/teams/brain/) and a community
+of users. We're eager to collaborate with you too, so feel free to
+[open an issue on GitHub](https://github.com/tensorflow/tensor2tensor/issues)
+or send along a pull request (see [our contribution doc](CONTRIBUTING.md)).
+You can chat with us on
+[Gitter](https://gitter.im/tensor2tensor/Lobby) and join the
+[T2T Google Group](https://groups.google.com/forum/#!forum/tensor2tensor).
 
 ### Quick Start
 
 [This iPython notebook](https://goo.gl/wkHexj) explains T2T and runs in your
 browser using a free VM from Google, no installation needed.
-
-Alternatively, here is a one-command version that installs T2T, downloads data,
-trains an MNIST model and evaluates it:
+Alternatively, here is a one-command version that installs T2T, downloads MNIST,
+trains a model and evaluates it:
 
 ```
 pip install tensor2tensor && t2t-trainer \
   --generate_data \
   --data_dir=~/t2t_data \
+  --output_dir=~/t2t_train/mnist \
   --problems=image_mnist \
   --model=shake_shake \
   --hparams_set=shake_shake_quick \
-  --output_dir=~/t2t_train/mnist1
-```
-
-For a more demanding problem, here is how to train
-an English-German translation model and evaluate it:
-
-```
-pip install tensor2tensor && t2t-trainer \
-  --generate_data \
-  --data_dir=~/t2t_data \
-  --problems=translate_ende_wmt32k \
-  --model=transformer \
-  --hparams_set=transformer_base_single_gpu \
-  --output_dir=~/t2t_train/base
+  --train_steps=1000 \
+  --eval_steps=100
 ```
 
-You can decode from the model interactively to get translations:
-
-```
-t2t-decoder \
-  --data_dir=~/t2t_data \
-  --problems=translate_ende_wmt32k \
-  --model=transformer \
-  --hparams_set=transformer_base_single_gpu \
-  --output_dir=~/t2t_train/base \
-  --decode_interactive
-```
-
-See the [Walkthrough](#walkthrough) below for more details on each step
-and [Suggested Models](#suggested-models) for well performing models
-on common tasks.
-
 ### Contents
 
-* [Walkthrough](#walkthrough)
-* [Suggested Models](#suggested-models)
-  * [Translation](#translation)
-  * [Summarization](#summarization)
+* [Suggested Datasets and Models](#suggested-datasets-and-models)
   * [Image Classification](#image-classification)
-* [Installation](#installation)
-* [Features](#features)
+  * [Language Modeling](#language-modeling)
+  * [Sentiment Analysis](#sentiment-analysis)
+  * [Speech Recognition](#speech-recognition)
+  * [Summarization](#summarization)
+  * [Translation](#translation)
+* [Basics](#basics)
+  * [Walkthrough](#walkthrough)
+  * [Installation](#installation)
+  * [Features](#features)
 * [T2T Overview](#t2t-overview)
   * [Datasets](#datasets)
   * [Problems and Modalities](#problems-and-modalities)
@@ -101,10 +64,102 @@ on common tasks.
   * [Trainer](#trainer)
 * [Adding your own components](#adding-your-own-components)
 * [Adding a dataset](#adding-a-dataset)
+* [Papers](#papers)
+
+## Suggested Datasets and Models
+
+Below we list a number of tasks that can be solved with T2T when
+you train the appropriate model on the appropriate problem.
+We give the problem and model below and we suggest a setting of
+hyperparameters that we know works well in our setup. We usually
+run either on Cloud TPUs or on 8-GPU machines; you might need
+to modify the hyperparameters if you run on a different setup.
+
+### Image Classification
+
+For image classification, we have a number of standard data-sets:
+* ImageNet (a large data-set): `--problems=image_imagenet`, or one
+   of the re-scaled versions (`image_imagenet224`, `image_imagenet64`,
+   `image_imagenet32`)
+* CIFAR-10: `--problems=image_cifar10` (or
+    `--problems=image_cifar10_plain` to turn off data augmentation)
+* CIFAR-100: `--problems=image_cifar100`
+* MNIST: `--problems=image_mnist`
+
+For ImageNet, we suggest to use the ResNet or Xception, i.e.,
+use `--model=resnet --hparams_set=resnet_50` or
+`--model=xception --hparams_set=xception_base`.
+Resnet should get to above 76% top-1 accuracy on ImageNet.
+
+For CIFAR and MNIST, we suggest to try the shake-shake model:
+`--model=shake_shake --hparams_set=shakeshake_big`.
+This setting trained for `--train_steps=700000` should yield
+close to 97% accuracy on CIFAR-10.
+
+### Language Modeling
+
+For language modeling, we have these data-sets in T2T:
+* PTB (a small data-set): `--problems=languagemodel_ptb10k` for
+    word-level modeling and `--problems=languagemodel_ptb_characters`
+    for character-level modeling.
+* LM1B (a billion-word corpus): `--problems=languagemodel_lm1b32k` for
+    subword-level modeling and `--problems=languagemodel_lm1b_characters`
+    for character-level modeling.
+
+We suggest to start with `--model=transformer` on this task and use
+`--hparams_set=transformer_small` for PTB and
+`--hparams_set=transformer_base` for LM1B.
+
+### Sentiment Analysis
 
----
+For the task of recognizing the sentiment of a sentence, use
+* the IMDB data-set: `--problems=sentiment_imdb`
 
-## Walkthrough
+We suggest to use `--model=transformer_encoder` here and since it is
+a small data-set, try `--hparams_set=transformer_tiny` and train for
+few steps (e.g., `--train_steps=2000`).
+
+### Speech Recognition
+
+For speech-to-text, we have these data-sets in T2T:
+* Librispeech (English speech to text): `--problems=librispeech` for
+    the whole set and `--problems=librispeech_clean` for a smaller
+    but nicely filtered part.
+
+### Summarization
+
+For summarizing longer text into shorter one we have these data-sets:
+* CNN/DailyMail articles summarized into a few sentences:
+  `--problems=summarize_cnn_dailymail32k`
+
+We suggest to use `--model=transformer` and
+`--hparams_set=transformer_prepend` for this task.
+This yields good ROUGE scores.
+
+### Translation
+
+There are a number of translation data-sets in T2T:
+* English-German: `--problems=translate_ende_wmt32k`
+* English-French: `--problems=translate_enfr_wmt32k`
+* English-Czech: `--problems=translate_encs_wmt32k`
+* English-Chinese: `--problems=translate_enzh_wmt32k`
+
+You can get translations in the other direction by appending `_rev` to
+the problem name, e.g., for German-English use
+`--problems=translate_ende_wmt32k_rev`.
+
+For all translation problems, we suggest to try the Transformer model:
+`--model=transformer`. At first it is best to try the base setting,
+`--hparams_set=transformer_base`. When trained on 8 GPUs for 300K steps
+this should reach a BLEU score of about 28 on the English-German data-set,
+which is close to state-of-the art. If training on a single GPU, try the
+`--hparams_set=transformer_base_single_gpu` setting. For very good results
+or larger data-sets (e.g., for English-French)m, try the big model
+with `--hparams_set=transformer_big`.
+
+## Basics
+
+### Walkthrough
 
 Here's a walkthrough training a good English-to-German translation
 model using the Transformer model from [*Attention Is All You
@@ -170,36 +225,8 @@ cat translation.en
 t2t-bleu --translation=translation.en --reference=ref-translation.de
 ```
 
----
-
-## Suggested Models
+### Installation
 
-Here are some combinations of models, hparams and problems that we found
-work well, so we suggest to use them if you're interested in that problem.
-
-### Translation
-
-For translation, esp. English-German and English-French, we suggest to use
-the Transformer model in base or big configurations, i.e.
-for `--problems=translate_ende_wmt32k` use `--model=transformer` and
-`--hparams_set=transformer_base`. When trained on 8 GPUs for 300K steps
-this should reach a BLEU score of about 28.
-
-### Summarization
-
-For summarization suggest to use the Transformer model in prepend mode, i.e.
-for `--problems=summarize_cnn_dailymail32k` use `--model=transformer` and
-`--hparams_set=transformer_prepend`.
-
-### Image Classification
-
-For image classification suggest to use the ResNet or Xception, i.e.
-for `--problems=image_imagenet` use `--model=resnet50` and
-`--hparams_set=resnet_base` or `--model=xception` and
-`--hparams_set=xception_base`.
-
-
-## Installation
 
 ```
 # Assumes tensorflow or tensorflow-gpu installed
@@ -228,9 +255,7 @@ Library usage:
 python -c "from tensor2tensor.models.transformer import Transformer"
 ```
 
----
-
-## Features
+### Features
 
 * Many state of the art and baseline models are built-in and new models can be
   added easily (open an issue or pull request!).
@@ -243,11 +268,10 @@ python -c "from tensor2tensor.models.transformer import Transformer"
   specification.
 * Support for multi-GPU machines and synchronous (1 master, many workers) and
   asynchronous (independent workers synchronizing through a parameter server)
-  [distributed training](https://github.com/tensorflow/tensor2tensor/tree/master/docs/distributed_training.md).
+  [distributed training](https://tensorflow.github.io/tensor2tensor/distributed_training.html).
 * Easily swap amongst datasets and models by command-line flag with the data
   generation script `t2t-datagen` and the training script `t2t-trainer`.
-
----
+* Train on [Google Cloud ML](https://tensorflow.github.io/tensor2tensor/cloud_mlengine.html) and [Cloud TPUs](https://tensorflow.github.io/tensor2tensor/cloud_tpu.html).
 
 ## T2T overview
 
@@ -303,9 +327,7 @@ inference. Users can easily switch between problems, models, and hyperparameter
 sets by using the `--model`, `--problems`, and `--hparams_set` flags. Specific
 hyperparameters can be overridden with the `--hparams` flag. `--schedule` and
 related flags control local and distributed training/evaluation
-([distributed training documentation](https://github.com/tensorflow/tensor2tensor/tree/master/docs/distributed_training.md)).
-
----
+([distributed training documentation](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/g3doc/distributed_training.md)).
 
 ## Adding your own components
 
@@ -331,6 +353,21 @@ for an example.
 Also see the [data generators
 README](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/data_generators/README.md).
 
----
+## Papers
+
+Tensor2Tensor was used to develop a number of state-of-the-art models
+and deep learning methods. Here we list some papers that were based on T2T
+from the start and benefited from its features and architecture in ways
+described in the [Google Research Blog post introducing
+T2T](https://research.googleblog.com/2017/06/accelerating-deep-learning-research.html).
+
+* [Attention Is All You Need](https://arxiv.org/abs/1706.03762)
+* [Depthwise Separable Convolutions for Neural Machine
+   Translation](https://arxiv.org/abs/1706.03059)
+* [One Model To Learn Them All](https://arxiv.org/abs/1706.05137)
+* [Discrete Autoencoders for Sequence Models](https://arxiv.org/abs/1801.09797)
+* [Generating Wikipedia by Summarizing Long
+   Sequences](https://arxiv.org/abs/1801.10198)
+* [Image Transformer](https://openreview.net/forum?id=r16Vyf-0-)
 
 *Note: This is not an official Google product.*
diff --git a/docs/cloud_tpu.md b/docs/cloud_tpu.md
index d375c6741..931f75bbf 100644
--- a/docs/cloud_tpu.md
+++ b/docs/cloud_tpu.md
@@ -1,17 +1,30 @@
 # Running on Cloud TPUs
 
-Tensor2Tensor supports running on Google Cloud Platforms TPUs, chips specialized
-for ML training.
+Tensor2Tensor supports running on Google Cloud Platforms TPUs, chips
+specialized for ML training. See the official tutorial for [running Transfomer
+on Cloud TPUs](https://cloud.google.com/tpu/docs/tutorials/transformer) or
+read on for more T2T models on TPUs.
 
-Models and hparams that are known to work on TPU:
-* `transformer` with `transformer_tpu`
-* `transformer_encoder` with `transformer_tpu`
-* `transformer_decoder` with `transformer_tpu`
-* `resnet50` with `resnet_base`
-* `revnet104` with `revnet_base`
+## Models and hparams for TPU:
 
-TPU access is currently limited, but access will expand soon, so get excited for
-your future ML supercomputers in the cloud.
+Transformer:
+* `transformer` with `transformer_tpu` (or `transformer_packed_tpu`,
+    `transformer_tiny_tpu`, `transformer_big_tpu`)
+* `transformer_encoder` with `transformer_tpu` (and the above ones)
+
+You can run the Transformer model on a number of problems,
+from translation through language modeling to sentiment analysis.
+See the official tutorial for [running Transfomer
+on Cloud TPUs](https://cloud.google.com/tpu/docs/tutorials/transformer)
+for some examples and try out your own problems.
+
+Residual networks:
+* `resnet` with `resnet_50` (or `resnet_18` or `resnet_34`)
+* `revnet` with `revnet_104` (or `revnet_38_cifar`)
+* `shake_shake` with `shakeshake_tpu` (or `shakeshake_small`)
+
+We run residual networks on MNIST, CIFAR and ImageNet, but they should
+work on any image classification data-set.
 
 ## Tutorial: Transformer En-De translation on TPU
 
@@ -75,3 +88,24 @@ switch between hardware at will.
 * `--cloud_tpu_name`: The name of the TPU instance to use or create. If you want
   to launch multiple jobs on TPU, provide different names here for each one.
   Each TPU instance can only be training one model at a time.
+
+## Other T2T models on TPU
+
+To run other models on TPU, proceed exactly as in the tutorial above,
+just with different model, problem and hparams_set (and directories).
+For example, to train a shake-shake model on CIFAR you can run this command.
+```
+t2t-trainer \
+  --model=shake_shake \
+  --hparams_set=shakeshake_tpu \
+  --problems=image_cifar10 \
+  --train_steps=180000 \
+  --eval_steps=9 \
+  --local_eval_frequency=100 \
+  --data_dir=$DATA_DIR \
+  --output_dir=$OUT_DIR \
+  --cloud_tpu \
+  --cloud_delete_on_done
+```
+Note that `eval_steps` should not be too high so as not to run out
+of evaluation data.
diff --git a/docs/index.md b/docs/index.md
index da2446803..060a878a3 100644
--- a/docs/index.md
+++ b/docs/index.md
@@ -1,4 +1,4 @@
-# Tensor2Tensor Docs Index
+# Tensor2Tensor Documentation
 
 [![PyPI
 version](https://badge.fury.io/py/tensor2tensor.svg)](https://badge.fury.io/py/tensor2tensor)
@@ -9,21 +9,113 @@ welcome](https://img.shields.io/badge/contributions-welcome-brightgreen.svg)](CO
 [![Gitter](https://img.shields.io/gitter/room/nwjs/nw.js.svg)](https://gitter.im/tensor2tensor/Lobby)
 [![License](https://img.shields.io/badge/License-Apache%202.0-brightgreen.svg)](https://opensource.org/licenses/Apache-2.0)
 
+[Tensor2Tensor](https://github.com/tensorflow/tensor2tensor), or
+[T2T](https://github.com/tensorflow/tensor2tensor) for short, is a library
+of deep learning models and datasets designed to [accelerate deep learning
+research](https://research.googleblog.com/2017/06/accelerating-deep-learning-research.html) and make it more accessible.
 
-Welcome to Tensor2Tensor!
-
-Tensor2Tensor, or T2T for short, is a library we use to create,
-investigate and deploy deep learning models. This page hosts our
-documentation, from basic tutorials to full code documentation.
 
 ## Basics
 
-* [Walkthrough: Install and Run](walkthrough.md)
-* [Tutorial: Train on Your Data](new_problem.md)
-* [Tutorial: Create Your Own Model](new_model.md)
+* [Walkthrough](walkthrough.md): Install and run.
+* [IPython notebook](https://goo.gl/wkHexj): Get a hands-on experience.
+* [Overview](overview.md): How all parts of T2T code are connected.
+* [New Problem](new_problem.md): Train T2T models on your data.
+* [New Model](new_model.md): Create your own T2T model.
 
-## Deep Dive
+## Training in the cloud
 
-* [System Overview](overview.md): how all parts of T2T are connected and
-  work together
+* [Training on Google Cloud ML](cloud_mlengine.md)
+* [Training on Google Cloud TPUs](cloud_tpu.md)
 * [Distributed Training](distributed_training.md)
+
+## Solving your task
+
+Below we list a number of tasks that can be solved with T2T when
+you train the appropriate model on the appropriate problem.
+We give the problem and model below and we suggest a setting of
+hyperparameters that we know works well in our setup. We usually
+run either on Cloud TPUs or on 8-GPU machines; you might need
+to modify the hyperparameters if you run on a different setup.
+
+### Image Classification
+
+For image classification, we have a number of standard data-sets:
+* ImageNet (a large data-set): `--problems=image_imagenet`, or one
+   of the re-scaled versions (`image_imagenet224`, `image_imagenet64`,
+   `image_imagenet32`)
+* CIFAR-10: `--problems=image_cifar10` (or
+    `--problems=image_cifar10_plain` to turn off data augmentation)
+* CIFAR-100: `--problems=image_cifar100`
+* MNIST: `--problems=image_mnist`
+
+For ImageNet, we suggest to use the ResNet or Xception, i.e.,
+use `--model=resnet --hparams_set=resnet_50` or
+`--model=xception --hparams_set=xception_base`.
+Resnet should get to above 76% top-1 accuracy on ImageNet.
+
+For CIFAR and MNIST, we suggest to try the shake-shake model:
+`--model=shake_shake --hparams_set=shakeshake_big`.
+This setting trained for `--train_steps=700000` should yield
+close to 97% accuracy on CIFAR-10.
+
+### Language Modeling
+
+For language modeling, we have these data-sets in T2T:
+* PTB (a small data-set): `--problems=languagemodel_ptb10k` for
+    word-level modeling and `--problems=languagemodel_ptb_characters`
+    for character-level modeling.
+* LM1B (a billion-word corpus): `--problems=languagemodel_lm1b32k` for
+    subword-level modeling and `--problems=languagemodel_lm1b_characters`
+    for character-level modeling.
+    
+We suggest to start with `--model=transformer` on this task and use
+`--hparams_set=transformer_small` for PTB and
+`--hparams_set=transformer_base` for LM1B.
+
+### Sentiment Analysis
+
+For the task of recognizing the sentiment of a sentence, use
+* the IMDB data-set: `--problems=sentiment_imdb`
+
+We suggest to use `--model=transformer_encoder` here and since it is
+a small data-set, try `--hparams_set=transformer_tiny` and train for
+few steps (e.g., `--train_steps=2000`).
+
+### Speech Recognition
+
+For speech-to-text, we have these data-sets in T2T:
+* Librispeech (English speech to text): `--problems=librispeech` for
+    the whole set and `--problems=librispeech_clean` for a smaller
+    but nicely filtered part.
+
+### Summarization
+
+For summarizing longer text into shorter one we have these data-sets:
+* CNN/DailyMail articles summarized into a few sentences:
+  `--problems=summarize_cnn_dailymail32k`
+  
+We suggest to use `--model=transformer` and
+`--hparams_set=transformer_prepend` for this task.
+This yields good ROUGE scores.
+
+### Translation
+
+There are a number of translation data-sets in T2T:
+* English-German: `--problems=translate_ende_wmt32k`
+* English-French: `--problems=translate_enfr_wmt32k`
+* English-Czech: `--problems=translate_encs_wmt32k`
+* English-Chinese: `--problems=translate_enzh_wmt32k`
+
+You can get translations in the other direction by appending `_rev` to
+the problem name, e.g., for German-English use
+`--problems=translate_ende_wmt32k_rev`.
+
+For all translation problems, we suggest to try the Transformer model:
+`--model=transformer`. At first it is best to try the base setting,
+`--hparams_set=transformer_base`. When trained on 8 GPUs for 300K steps
+this should reach a BLEU score of about 28 on the English-German data-set,
+which is close to state-of-the art. If training on a single GPU, try the
+`--hparams_set=transformer_base_single_gpu` setting. For very good results
+or larger data-sets (e.g., for English-French)m, try the big model
+with `--hparams_set=transformer_big`.
diff --git a/docs/walkthrough.md b/docs/walkthrough.md
index 1f32351ac..a569fb80a 100644
--- a/docs/walkthrough.md
+++ b/docs/walkthrough.md
@@ -12,87 +12,50 @@ welcome](https://img.shields.io/badge/contributions-welcome-brightgreen.svg)](CO
 
 [Tensor2Tensor](https://github.com/tensorflow/tensor2tensor), or
 [T2T](https://github.com/tensorflow/tensor2tensor) for short, is a library
-of deep learning models and datasets. It has binaries to train the models and
-to download and prepare the data for you. T2T is modular and extensible and can
-be used in [notebooks](https://goo.gl/wkHexj) for prototyping your own models
-or running existing ones on your data. It is actively used and maintained by
-researchers and engineers within
-the [Google Brain team](https://research.google.com/teams/brain/) and was used
-to develop state-of-the-art models for translation (see
-[Attention Is All You Need](https://arxiv.org/abs/1706.03762)),
-[summarization](https://arxiv.org/abs/1801.10198),
-image generation and other tasks. You can read
-more about T2T in the [Google Research Blog post introducing
-it](https://research.googleblog.com/2017/06/accelerating-deep-learning-research.html).
-
-We're eager to collaborate with you on extending T2T, so please feel
-free to [open an issue on
-GitHub](https://github.com/tensorflow/tensor2tensor/issues) or
-send along a pull request to add your dataset or model.
-See [our contribution
-doc](CONTRIBUTING.md) for details and our [open
-issues](https://github.com/tensorflow/tensor2tensor/issues).
-You can chat with us and other users on
-[Gitter](https://gitter.im/tensor2tensor/Lobby) and please join our
-[Google Group](https://groups.google.com/forum/#!forum/tensor2tensor) to keep up
-with T2T announcements.
+of deep learning models and datasets designed to [accelerate deep learning
+research](https://research.googleblog.com/2017/06/accelerating-deep-learning-research.html) and make it more accessible.
+
+T2T is actively used and maintained by researchers and engineers within the
+[Google Brain team](https://research.google.com/teams/brain/) and a community
+of users. We're eager to collaborate with you too, so feel free to
+[open an issue on GitHub](https://github.com/tensorflow/tensor2tensor/issues)
+or send along a pull request (see [our contribution doc](CONTRIBUTING.md)).
+You can chat with us on
+[Gitter](https://gitter.im/tensor2tensor/Lobby) and join the
+[T2T Google Group](https://groups.google.com/forum/#!forum/tensor2tensor).
 
 ### Quick Start
 
 [This iPython notebook](https://goo.gl/wkHexj) explains T2T and runs in your
 browser using a free VM from Google, no installation needed.
-
-Alternatively, here is a one-command version that installs T2T, downloads data,
-trains an MNIST model and evaluates it:
+Alternatively, here is a one-command version that installs T2T, downloads MNIST,
+trains a model and evaluates it:
 
 ```
 pip install tensor2tensor && t2t-trainer \
   --generate_data \
   --data_dir=~/t2t_data \
+  --output_dir=~/t2t_train/mnist \
   --problems=image_mnist \
   --model=shake_shake \
   --hparams_set=shake_shake_quick \
-  --output_dir=~/t2t_train/mnist1
-```
-
-For a more demanding problem, here is how to train
-an English-German translation model and evaluate it:
-
-```
-pip install tensor2tensor && t2t-trainer \
-  --generate_data \
-  --data_dir=~/t2t_data \
-  --problems=translate_ende_wmt32k \
-  --model=transformer \
-  --hparams_set=transformer_base_single_gpu \
-  --output_dir=~/t2t_train/base
+  --train_steps=1000 \
+  --eval_steps=100
 ```
 
-You can decode from the model interactively to get translations:
-
-```
-t2t-decoder \
-  --data_dir=~/t2t_data \
-  --problems=translate_ende_wmt32k \
-  --model=transformer \
-  --hparams_set=transformer_base_single_gpu \
-  --output_dir=~/t2t_train/base \
-  --decode_interactive
-```
-
-See the [Walkthrough](#walkthrough) below for more details on each step
-and [Suggested Models](#suggested-models) for well performing models
-on common tasks.
-
 ### Contents
 
-* [Walkthrough](#walkthrough)
-* [Suggested Models](#suggested-models)
-  * [Translation](#translation)
-  * [Summarization](#summarization)
+* [Suggested Datasets and Models](#suggested-datasets-and-models)
   * [Image Classification](#image-classification)
-* [Installation](#installation)
-* [Features](#features)
+  * [Language Modeling](#language-modeling)
+  * [Sentiment Analysis](#sentiment-analysis)
+  * [Speech Recognition](#speech-recognition)
+  * [Summarization](#summarization)
+  * [Translation](#translation)
+* [Basics](#basics)
+  * [Walkthrough](#walkthrough)
+  * [Installation](#installation)
+  * [Features](#features)
 * [T2T Overview](#t2t-overview)
   * [Datasets](#datasets)
   * [Problems and Modalities](#problems-and-modalities)
@@ -101,10 +64,102 @@ on common tasks.
   * [Trainer](#trainer)
 * [Adding your own components](#adding-your-own-components)
 * [Adding a dataset](#adding-a-dataset)
+* [Papers](#papers)
+
+## Suggested Datasets and Models
+
+Below we list a number of tasks that can be solved with T2T when
+you train the appropriate model on the appropriate problem.
+We give the problem and model below and we suggest a setting of
+hyperparameters that we know works well in our setup. We usually
+run either on Cloud TPUs or on 8-GPU machines; you might need
+to modify the hyperparameters if you run on a different setup.
+
+### Image Classification
+
+For image classification, we have a number of standard data-sets:
+* ImageNet (a large data-set): `--problems=image_imagenet`, or one
+   of the re-scaled versions (`image_imagenet224`, `image_imagenet64`,
+   `image_imagenet32`)
+* CIFAR-10: `--problems=image_cifar10` (or
+    `--problems=image_cifar10_plain` to turn off data augmentation)
+* CIFAR-100: `--problems=image_cifar100`
+* MNIST: `--problems=image_mnist`
+
+For ImageNet, we suggest to use the ResNet or Xception, i.e.,
+use `--model=resnet --hparams_set=resnet_50` or
+`--model=xception --hparams_set=xception_base`.
+Resnet should get to above 76% top-1 accuracy on ImageNet.
+
+For CIFAR and MNIST, we suggest to try the shake-shake model:
+`--model=shake_shake --hparams_set=shakeshake_big`.
+This setting trained for `--train_steps=700000` should yield
+close to 97% accuracy on CIFAR-10.
+
+### Language Modeling
+
+For language modeling, we have these data-sets in T2T:
+* PTB (a small data-set): `--problems=languagemodel_ptb10k` for
+    word-level modeling and `--problems=languagemodel_ptb_characters`
+    for character-level modeling.
+* LM1B (a billion-word corpus): `--problems=languagemodel_lm1b32k` for
+    subword-level modeling and `--problems=languagemodel_lm1b_characters`
+    for character-level modeling.
+
+We suggest to start with `--model=transformer` on this task and use
+`--hparams_set=transformer_small` for PTB and
+`--hparams_set=transformer_base` for LM1B.
+
+### Sentiment Analysis
 
----
+For the task of recognizing the sentiment of a sentence, use
+* the IMDB data-set: `--problems=sentiment_imdb`
 
-## Walkthrough
+We suggest to use `--model=transformer_encoder` here and since it is
+a small data-set, try `--hparams_set=transformer_tiny` and train for
+few steps (e.g., `--train_steps=2000`).
+
+### Speech Recognition
+
+For speech-to-text, we have these data-sets in T2T:
+* Librispeech (English speech to text): `--problems=librispeech` for
+    the whole set and `--problems=librispeech_clean` for a smaller
+    but nicely filtered part.
+
+### Summarization
+
+For summarizing longer text into shorter one we have these data-sets:
+* CNN/DailyMail articles summarized into a few sentences:
+  `--problems=summarize_cnn_dailymail32k`
+
+We suggest to use `--model=transformer` and
+`--hparams_set=transformer_prepend` for this task.
+This yields good ROUGE scores.
+
+### Translation
+
+There are a number of translation data-sets in T2T:
+* English-German: `--problems=translate_ende_wmt32k`
+* English-French: `--problems=translate_enfr_wmt32k`
+* English-Czech: `--problems=translate_encs_wmt32k`
+* English-Chinese: `--problems=translate_enzh_wmt32k`
+
+You can get translations in the other direction by appending `_rev` to
+the problem name, e.g., for German-English use
+`--problems=translate_ende_wmt32k_rev`.
+
+For all translation problems, we suggest to try the Transformer model:
+`--model=transformer`. At first it is best to try the base setting,
+`--hparams_set=transformer_base`. When trained on 8 GPUs for 300K steps
+this should reach a BLEU score of about 28 on the English-German data-set,
+which is close to state-of-the art. If training on a single GPU, try the
+`--hparams_set=transformer_base_single_gpu` setting. For very good results
+or larger data-sets (e.g., for English-French)m, try the big model
+with `--hparams_set=transformer_big`.
+
+## Basics
+
+### Walkthrough
 
 Here's a walkthrough training a good English-to-German translation
 model using the Transformer model from [*Attention Is All You
@@ -170,36 +225,8 @@ cat translation.en
 t2t-bleu --translation=translation.en --reference=ref-translation.de
 ```
 
----
-
-## Suggested Models
+### Installation
 
-Here are some combinations of models, hparams and problems that we found
-work well, so we suggest to use them if you're interested in that problem.
-
-### Translation
-
-For translation, esp. English-German and English-French, we suggest to use
-the Transformer model in base or big configurations, i.e.
-for `--problems=translate_ende_wmt32k` use `--model=transformer` and
-`--hparams_set=transformer_base`. When trained on 8 GPUs for 300K steps
-this should reach a BLEU score of about 28.
-
-### Summarization
-
-For summarization suggest to use the Transformer model in prepend mode, i.e.
-for `--problems=summarize_cnn_dailymail32k` use `--model=transformer` and
-`--hparams_set=transformer_prepend`.
-
-### Image Classification
-
-For image classification suggest to use the ResNet or Xception, i.e.
-for `--problems=image_imagenet` use `--model=resnet50` and
-`--hparams_set=resnet_base` or `--model=xception` and
-`--hparams_set=xception_base`.
-
-
-## Installation
 
 ```
 # Assumes tensorflow or tensorflow-gpu installed
@@ -228,9 +255,7 @@ Library usage:
 python -c "from tensor2tensor.models.transformer import Transformer"
 ```
 
----
-
-## Features
+### Features
 
 * Many state of the art and baseline models are built-in and new models can be
   added easily (open an issue or pull request!).
@@ -243,11 +268,10 @@ python -c "from tensor2tensor.models.transformer import Transformer"
   specification.
 * Support for multi-GPU machines and synchronous (1 master, many workers) and
   asynchronous (independent workers synchronizing through a parameter server)
-  [distributed training](https://github.com/tensorflow/tensor2tensor/tree/master/docs/distributed_training.md).
+  [distributed training](https://tensorflow.github.io/tensor2tensor/distributed_training.html).
 * Easily swap amongst datasets and models by command-line flag with the data
   generation script `t2t-datagen` and the training script `t2t-trainer`.
-
----
+* Train on [Google Cloud ML](https://tensorflow.github.io/tensor2tensor/cloud_mlengine.html) and [Cloud TPUs](https://tensorflow.github.io/tensor2tensor/cloud_tpu.html).
 
 ## T2T overview
 
@@ -303,9 +327,7 @@ inference. Users can easily switch between problems, models, and hyperparameter
 sets by using the `--model`, `--problems`, and `--hparams_set` flags. Specific
 hyperparameters can be overridden with the `--hparams` flag. `--schedule` and
 related flags control local and distributed training/evaluation
-([distributed training documentation](https://github.com/tensorflow/tensor2tensor/tree/master/docs/distributed_training.md)).
-
----
+([distributed training documentation](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/g3doc/distributed_training.md)).
 
 ## Adding your own components
 
@@ -331,6 +353,21 @@ for an example.
 Also see the [data generators
 README](https://github.com/tensorflow/tensor2tensor/tree/master/tensor2tensor/data_generators/README.md).
 
----
+## Papers
+
+Tensor2Tensor was used to develop a number of state-of-the-art models
+and deep learning methods. Here we list some papers that were based on T2T
+from the start and benefited from its features and architecture in ways
+described in the [Google Research Blog post introducing
+T2T](https://research.googleblog.com/2017/06/accelerating-deep-learning-research.html).
+
+* [Attention Is All You Need](https://arxiv.org/abs/1706.03762)
+* [Depthwise Separable Convolutions for Neural Machine
+   Translation](https://arxiv.org/abs/1706.03059)
+* [One Model To Learn Them All](https://arxiv.org/abs/1706.05137)
+* [Discrete Autoencoders for Sequence Models](https://arxiv.org/abs/1801.09797)
+* [Generating Wikipedia by Summarizing Long
+   Sequences](https://arxiv.org/abs/1801.10198)
+* [Image Transformer](https://openreview.net/forum?id=r16Vyf-0-)
 
 *Note: This is not an official Google product.*