en-urh/main/English_to_Urhobo_Word-level_notebook.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "starter_notebook.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "toc_visible": true
    },
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.5.6"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "Igc5itf-xMGj"
      },
      "source": [
        "# Masakhane - Machine Translation for African Languages (Using JoeyNMT)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "x4fXCKCf36IK"
      },
      "source": [
        "## Note before beginning:\n",
        "### - The idea is that you should be able to make minimal changes to this in order to get SOME result for your own translation corpus. \n",
        "\n",
        "### - The tl;dr: Go to the **\"TODO\"** comments which will tell you what to update to get up and running\n",
        "\n",
        "### - If you actually want to have a clue what you're doing, read the text and peek at the links\n",
        "\n",
        "### - With 100 epochs, it should take around 7 hours to run in Google Colab\n",
        "\n",
        "### - Once you've gotten a result for your language, please attach and email your notebook that generated it to masakhanetranslation@gmail.com\n",
        "\n",
        "### - If you care enough and get a chance, doing a brief background on your language would be amazing. See examples in  [(Martinus, 2019)](https://arxiv.org/abs/1906.05685)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "l929HimrxS0a"
      },
      "source": [
        "## Retrieve your data & make a parallel corpus\n",
        "\n",
        "If you are wanting to use the JW300 data referenced on the Masakhane website or in our GitHub repo, you can use `opus-tools` to convert the data into a convenient format. `opus_read` from that package provides a convenient tool for reading the native aligned XML files and to convert them to TMX format. The tool can also be used to fetch relevant files from OPUS on the fly and to filter the data as necessary. [Read the documentation](https://pypi.org/project/opustools-pkg/) for more details.\n",
        "\n",
        "Once you have your corpus files in TMX format (an xml structure which will include the sentences in your target language and your source language in a single file), we recommend reading them into a pandas dataframe. Thankfully, Jade wrote a silly `tmx2dataframe` package which converts your tmx file to a pandas dataframe. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "oGRmDELn7Az0",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "b8270ce4-6e57-4ce8-e7d9-46d20290dd69"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount(\"/content/drive\", force_remount=True).\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "Cn3tgQLzUxwn",
        "colab": {}
      },
      "source": [
        "# TODO: Set your source and target languages. Keep in mind, these traditionally use language codes as found here:\n",
        "# These will also become the suffix's of all vocab and corpus files used throughout\n",
        "import os\n",
        "source_language = \"en\"\n",
        "target_language = \"urh\" \n",
        "lc = False  # If True, lowercase the data.\n",
        "seed = 42  # Random seed for shuffling.\n",
        "tag = \"baseline\" # Give a unique name to your folder - this is to ensure you don't rewrite any models you've already submitted\n",
        "\n",
        "os.environ[\"src\"] = source_language # Sets them in bash as well, since we often use bash scripts\n",
        "os.environ[\"tgt\"] = target_language\n",
        "os.environ[\"tag\"] = tag\n",
        "\n",
        "# This will save it to a folder in our gdrive instead!\n",
        "!mkdir -p \"/content/drive/My Drive/masakhane/$src-$tgt-$tag\"\n",
        "os.environ[\"gdrive_path\"] = \"/content/drive/My Drive/masakhane/%s-%s-%s\" % (source_language, target_language, tag)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "kBSgJHEw7Nvx",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "d786cba5-3631-4959-8875-c15ef993c21e"
      },
      "source": [
        "!echo $gdrive_path"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/content/drive/My Drive/masakhane/en-urh-baseline\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "gA75Fs9ys8Y9",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "04370262-f35c-4d29-b251-41d182537b7f"
      },
      "source": [
        "# Install opus-tools\n",
        "! pip install opustools-pkg"
      ],
      "execution_count": 29,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Requirement already satisfied: opustools-pkg in /usr/local/lib/python3.6/dist-packages (0.0.52)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "xq-tDZVks7ZD",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 187
        },
        "outputId": "b62cc8ce-a08b-494f-ea6d-d262f7d2c6f8"
      },
      "source": [
        "# Downloading our corpus\n",
        "! opus_read -d JW300 -s $src -t $tgt -wm moses -w jw300.$src jw300.$tgt -q\n",
        "\n",
        "# extract the corpus file\n",
        "! gunzip JW300_latest_xml_$src-$tgt.xml.gz"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "\n",
            "Alignment file /proj/nlpl/data/OPUS/JW300/latest/xml/en-urh.xml.gz not found. The following files are available for downloading:\n",
            "\n",
            "        ./JW300_latest_xml_en.zip already exists\n",
            "        ./JW300_latest_xml_urh.zip already exists\n",
            " 304 KB https://object.pouta.csc.fi/OPUS-JW300/v1/xml/en-urh.xml.gz\n",
            "\n",
            " 304 KB Total size\n",
            "./JW300_latest_xml_en-urh.xml.gz ... 100% of 304 KB\n",
            "gzip: JW300_latest_xml_en-urh.xml already exists; do you wish to overwrite (y or n)? y\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "n48GDRnP8y2G",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 578
        },
        "outputId": "b50339aa-aae7-4027-a985-f303c50b290b"
      },
      "source": [
        "# Download the global test set.\n",
        "! wget https://raw.githubusercontent.com/juliakreutzer/masakhane/master/jw300_utils/test/test.en-any.en\n",
        "  \n",
        "# And the specific test set for this language pair.\n",
        "os.environ[\"trg\"] = target_language \n",
        "os.environ[\"src\"] = source_language \n",
        "\n",
        "! wget https://raw.githubusercontent.com/juliakreutzer/masakhane/master/jw300_utils/test/test.en-$trg.en \n",
        "! mv test.en-$trg.en test.en\n",
        "! wget https://raw.githubusercontent.com/juliakreutzer/masakhane/master/jw300_utils/test/test.en-$trg.$trg \n",
        "! mv test.en-$trg.$trg test.$trg"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "--2019-12-31 06:41:57--  https://raw.githubusercontent.com/juliakreutzer/masakhane/master/jw300_utils/test/test.en-any.en\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 151.101.0.133, 151.101.64.133, 151.101.128.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|151.101.0.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 277791 (271K) [text/plain]\n",
            "Saving to: ‘test.en-any.en.1’\n",
            "\n",
            "\rtest.en-any.en.1      0%[                    ]       0  --.-KB/s               \rtest.en-any.en.1    100%[===================>] 271.28K  --.-KB/s    in 0.02s   \n",
            "\n",
            "2019-12-31 06:41:57 (16.5 MB/s) - ‘test.en-any.en.1’ saved [277791/277791]\n",
            "\n",
            "--2019-12-31 06:41:59--  https://raw.githubusercontent.com/juliakreutzer/masakhane/master/jw300_utils/test/test.en-urh.en\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 151.101.0.133, 151.101.64.133, 151.101.128.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|151.101.0.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 201504 (197K) [text/plain]\n",
            "Saving to: ‘test.en-urh.en’\n",
            "\n",
            "test.en-urh.en      100%[===================>] 196.78K  --.-KB/s    in 0.01s   \n",
            "\n",
            "2019-12-31 06:41:59 (13.8 MB/s) - ‘test.en-urh.en’ saved [201504/201504]\n",
            "\n",
            "--2019-12-31 06:42:02--  https://raw.githubusercontent.com/juliakreutzer/masakhane/master/jw300_utils/test/test.en-urh.urh\n",
            "Resolving raw.githubusercontent.com (raw.githubusercontent.com)... 151.101.0.133, 151.101.64.133, 151.101.128.133, ...\n",
            "Connecting to raw.githubusercontent.com (raw.githubusercontent.com)|151.101.0.133|:443... connected.\n",
            "HTTP request sent, awaiting response... 200 OK\n",
            "Length: 236859 (231K) [text/plain]\n",
            "Saving to: ‘test.en-urh.urh’\n",
            "\n",
            "test.en-urh.urh     100%[===================>] 231.31K  --.-KB/s    in 0.02s   \n",
            "\n",
            "2019-12-31 06:42:03 (14.5 MB/s) - ‘test.en-urh.urh’ saved [236859/236859]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NqDG-CI28y2L",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "d11c61cb-d165-4fcd-cb20-92015d712ce5"
      },
      "source": [
        "# Read the test data to filter from train and dev splits.\n",
        "# Store english portion in set for quick filtering checks.\n",
        "en_test_sents = set()\n",
        "filter_test_sents = \"test.en-any.en\"\n",
        "j = 0\n",
        "with open(filter_test_sents) as f:\n",
        "  for line in f:\n",
        "    en_test_sents.add(line.strip())\n",
        "    j += 1\n",
        "print('Loaded {} global test sentences to filter from the training/dev data.'.format(j))"
      ],
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loaded 3571 global test sentences to filter from the training/dev data.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "3CNdwLBCfSIl",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 159
        },
        "outputId": "077bb76e-99f2-49e4-f539-1cd7a206eefa"
      },
      "source": [
        "import pandas as pd\n",
        "\n",
        "# TMX file to dataframe\n",
        "source_file = 'jw300.' + source_language\n",
        "target_file = 'jw300.' + target_language\n",
        "\n",
        "source = []\n",
        "target = []\n",
        "skip_lines = []  # Collect the line numbers of the source portion to skip the same lines for the target portion.\n",
        "with open(source_file) as f:\n",
        "    for i, line in enumerate(f):\n",
        "        # Skip sentences that are contained in the test set.\n",
        "        if line.strip() not in en_test_sents:\n",
        "            source.append(line.strip())\n",
        "        else:\n",
        "            skip_lines.append(i)             \n",
        "with open(target_file) as f:\n",
        "    for j, line in enumerate(f):\n",
        "        # Only add to corpus if corresponding source was not skipped.\n",
        "        if j not in skip_lines:\n",
        "            target.append(line.strip())\n",
        "    \n",
        "print('Loaded data and skipped {}/{} lines since contained in test set.'.format(len(skip_lines), i))\n",
        "    \n",
        "df = pd.DataFrame(zip(source, target), columns=['source_sentence', 'target_sentence'])\n",
        "# if you get TypeError: data argument can't be an iterator is because of your zip version run this below\n",
        "#df = pd.DataFrame(list(zip(source, target)), columns=['source_sentence', 'target_sentence'])\n",
        "df.head(3)"
      ],
      "execution_count": 33,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Loaded data and skipped 4050/32709 lines since contained in test set.\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th>source_sentence</th>\n",
              "      <th>target_sentence</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>0</th>\n",
              "      <td>Why It Pays to Be Honest 6</td>\n",
              "      <td>Erere Herọ Ra Vwọ Dia Ohwo rẹ Uyota 5</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>1</th>\n",
              "      <td>The Bible Changes Lives</td>\n",
              "      <td>7 Ovwan “ Jẹn Ẹguọnọ rẹ Iniọvo na Dje Ebuoebuo...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2</th>\n",
              "      <td>Give Me Just One Year of Peace and Happiness 8...</td>\n",
              "      <td>12 Jẹ ‘ Ẹse rẹ Ọghẹnẹ rẹ Unu se Gbe - e na , ’...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                                     source_sentence                                    target_sentence\n",
              "0                         Why It Pays to Be Honest 6              Erere Herọ Ra Vwọ Dia Ohwo rẹ Uyota 5\n",
              "1                            The Bible Changes Lives  7 Ovwan “ Jẹn Ẹguọnọ rẹ Iniọvo na Dje Ebuoebuo...\n",
              "2  Give Me Just One Year of Peace and Happiness 8...  12 Jẹ ‘ Ẹse rẹ Ọghẹnẹ rẹ Unu se Gbe - e na , ’..."
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 33
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "YkuK3B4p2AkN"
      },
      "source": [
        "## Pre-processing and export\n",
        "\n",
        "It is generally a good idea to remove duplicate translations and conflicting translations from the corpus. In practice, these public corpora include some number of these that need to be cleaned.\n",
        "\n",
        "In addition we will split our data into dev/test/train and export to the filesystem."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "M_2ouEOH1_1q",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 187
        },
        "outputId": "72a20551-8b78-4ce6-f227-0a5049476674"
      },
      "source": [
        "# drop duplicate translations\n",
        "df_pp = df.drop_duplicates()\n",
        "\n",
        "# drop conflicting translations\n",
        "# (this is optional and something that you might want to comment out \n",
        "# depending on the size of your corpus)\n",
        "df_pp.drop_duplicates(subset='source_sentence', inplace=True)\n",
        "df_pp.drop_duplicates(subset='target_sentence', inplace=True)\n",
        "\n",
        "# Shuffle the data to remove bias in dev set selection.\n",
        "df_pp = df_pp.sample(frac=1, random_state=seed).reset_index(drop=True)"
      ],
      "execution_count": 34,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:6: SettingWithCopyWarning: \n",
            "A value is trying to be set on a copy of a slice from a DataFrame\n",
            "\n",
            "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
            "  \n",
            "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:7: SettingWithCopyWarning: \n",
            "A value is trying to be set on a copy of a slice from a DataFrame\n",
            "\n",
            "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
            "  import sys\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z_1BwAApEtMk",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 561
        },
        "outputId": "ae7e81b6-ebc5-47d3-c10d-a9dd2b2ced41"
      },
      "source": [
        "# Install fuzzy wuzzy to remove \"almost duplicate\" sentences in the\n",
        "# test and training sets.\n",
        "! pip install fuzzywuzzy\n",
        "! pip install python-Levenshtein\n",
        "import time\n",
        "from fuzzywuzzy import process\n",
        "import numpy as np\n",
        "\n",
        "# reset the index of the training set after previous filtering\n",
        "df_pp.reset_index(drop=False, inplace=True)\n",
        "\n",
        "# Remove samples from the training data set if they \"almost overlap\" with the\n",
        "# samples in the test set.\n",
        "\n",
        "# Filtering function. Adjust pad to narrow down the candidate matches to\n",
        "# within a certain length of characters of the given sample.\n",
        "def fuzzfilter(sample, candidates, pad):\n",
        "  candidates = [x for x in candidates if len(x) <= len(sample)+pad and len(x) >= len(sample)-pad] \n",
        "  if len(candidates) > 0:\n",
        "    return process.extractOne(sample, candidates)[1]\n",
        "  else:\n",
        "    return np.nan\n",
        "\n",
        "# NOTE - This might run slow depending on the size of your training set. We are\n",
        "# printing some information to help you track how long it would take. \n",
        "scores = []\n",
        "start_time = time.time()\n",
        "for idx, row in df_pp.iterrows():\n",
        "  scores.append(fuzzfilter(row['source_sentence'], list(en_test_sents), 5))\n",
        "  if idx % 1000 == 0:\n",
        "    hours, rem = divmod(time.time() - start_time, 3600)\n",
        "    minutes, seconds = divmod(rem, 60)\n",
        "    print(\"{:0>2}:{:0>2}:{:05.2f}\".format(int(hours),int(minutes),seconds), \"%0.2f percent complete\" % (100.0*float(idx)/float(len(df_pp))))\n",
        "\n",
        "# Filter out \"almost overlapping samples\"\n",
        "df_pp['scores'] = scores\n",
        "df_pp = df_pp[df_pp['scores'] < 95]"
      ],
      "execution_count": 35,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Requirement already satisfied: fuzzywuzzy in /usr/local/lib/python3.6/dist-packages (0.17.0)\n",
            "Requirement already satisfied: python-Levenshtein in /usr/local/lib/python3.6/dist-packages (0.12.0)\n",
            "Requirement already satisfied: setuptools in /usr/local/lib/python3.6/dist-packages (from python-Levenshtein) (42.0.2)\n",
            "00:00:00.04 0.00 percent complete\n",
            "00:00:21.42 3.74 percent complete\n",
            "00:00:42.25 7.49 percent complete\n",
            "00:01:03.74 11.23 percent complete\n",
            "00:01:26.69 14.98 percent complete\n",
            "00:01:47.64 18.72 percent complete\n",
            "00:02:08.31 22.46 percent complete\n",
            "00:02:28.50 26.21 percent complete\n",
            "00:02:48.45 29.95 percent complete\n",
            "00:03:08.71 33.70 percent complete\n",
            "00:03:28.66 37.44 percent complete\n",
            "00:03:49.30 41.18 percent complete\n",
            "00:04:09.78 44.93 percent complete\n",
            "00:04:29.06 48.67 percent complete\n",
            "00:04:48.60 52.41 percent complete\n",
            "00:05:08.53 56.16 percent complete\n",
            "00:05:29.01 59.90 percent complete\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:root:Applied processor reduces input query to empty string, all comparisons will have score 0. [Query: '']\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "00:05:48.90 63.65 percent complete\n",
            "00:06:08.27 67.39 percent complete\n",
            "00:06:28.27 71.13 percent complete\n",
            "00:06:49.04 74.88 percent complete\n",
            "00:07:08.41 78.62 percent complete\n",
            "00:07:28.22 82.37 percent complete\n",
            "00:07:47.22 86.11 percent complete\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:root:Applied processor reduces input query to empty string, all comparisons will have score 0. [Query: '*']\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "00:08:06.79 89.85 percent complete\n",
            "00:08:26.41 93.60 percent complete\n",
            "00:08:46.27 97.34 percent complete\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "hxxBOCA-xXhy",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "52e8e2ea-c8f5-4eff-958f-d185ca390e8e"
      },
      "source": [
        "# This section does the split between train/dev for the parallel corpora then saves them as separate files\n",
        "# We use 1000 dev test and the given test set.\n",
        "import csv\n",
        "\n",
        "# Do the split between dev/train and create parallel corpora\n",
        "num_dev_patterns = 1000\n",
        "\n",
        "# Optional: lower case the corpora - this will make it easier to generalize, but without proper casing.\n",
        "if lc:  # Julia: making lowercasing optional\n",
        "    df_pp[\"source_sentence\"] = df_pp[\"source_sentence\"].str.lower()\n",
        "    df_pp[\"target_sentence\"] = df_pp[\"target_sentence\"].str.lower()\n",
        "\n",
        "# Julia: test sets are already generated\n",
        "dev = df_pp.tail(num_dev_patterns) # Herman: Error in original\n",
        "stripped = df_pp.drop(df_pp.tail(num_dev_patterns).index)\n",
        "\n",
        "with open(\"train.\"+source_language, \"w\") as src_file, open(\"train.\"+target_language, \"w\") as trg_file:\n",
        "  for index, row in stripped.iterrows():\n",
        "    src_file.write(row[\"source_sentence\"]+\"\\n\")\n",
        "    trg_file.write(row[\"target_sentence\"]+\"\\n\")\n",
        "    \n",
        "with open(\"dev.\"+source_language, \"w\") as src_file, open(\"dev.\"+target_language, \"w\") as trg_file:\n",
        "  for index, row in dev.iterrows():\n",
        "    src_file.write(row[\"source_sentence\"]+\"\\n\")\n",
        "    trg_file.write(row[\"target_sentence\"]+\"\\n\")\n",
        "\n",
        "#stripped[[\"source_sentence\"]].to_csv(\"train.\"+source_language, header=False, index=False)  # Herman: Added `header=False` everywhere\n",
        "#stripped[[\"target_sentence\"]].to_csv(\"train.\"+target_language, header=False, index=False)  # Julia: Problematic handling of quotation marks.\n",
        "\n",
        "#dev[[\"source_sentence\"]].to_csv(\"dev.\"+source_language, header=False, index=False)\n",
        "#dev[[\"target_sentence\"]].to_csv(\"dev.\"+target_language, header=False, index=False)\n",
        "\n",
        "# Doublecheck the format below. There should be no extra quotation marks or weird characters.\n",
        "! head train.*\n",
        "! head dev.*"
      ],
      "execution_count": 36,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "==> train.bpe.en <==\n",
            "The number of publishers is now about ten times what it was when I began serving here .\n",
            "Sim@@ il@@ arly , elders should not only encourage and con@@ so@@ le their brothers with words but also build them up by showing sin@@ c@@ ere personal interest . ​ — 1 Cor .\n",
            "17 Why We “ Keep B@@ earing M@@ u@@ ch F@@ ru@@ it ”\n",
            "Now I have a j@@ our@@ n@@ al that I keep on my des@@ k to s@@ che@@ du@@ le up@@ coming work , and this helps me to s@@ che@@ du@@ le my@@ self , not lea@@ ving things till the last min@@ ute . ”\n",
            "1 , 2 . ( a ) How do some react to the thought that God has an organization ?\n",
            "We cannot go to the point of dis@@ obey@@ ing God or viol@@ ating our Christian ne@@ u@@ tr@@ ality . ​ — Read 1 Peter 2 : 13 - 17 .\n",
            "Did this mean freedom for every liter@@ al slave ?\n",
            "E@@ ven@@ tually , all my si@@ bl@@ ings did so and became Jehovah’s Witnesses .\n",
            "How plea@@ sed Jehovah will be as he ob@@ ser@@ ves our whole - s@@ ou@@ led efforts to “ keep b@@ earing much fruit ” !\n",
            "Joseph , though , was a disci@@ ple , but he could not bring himself to say so op@@ en@@ ly .\n",
            "\n",
            "==> train.bpe.urh <==\n",
            "I@@ ghwoghwota rehẹ ẹkuotọ na enẹna vwẹ ọh@@ wọ@@ h@@ wọ ihwe vwo bun vrẹ obo rọ hepha ọke me vwọ ga vwẹ oboyin .\n",
            "( 2 Kọr . 12 : 15 ) Vwẹ idjerhe vuọvo na , vwọ vrẹ ota rẹ unu rẹ ekpako cha vwọ bọn iniọvo na gan , o ji fo nẹ ayen ru obo ro che djephia nẹ ayen vwo ọdavwẹ rayen . ​ — 1 Kọr .\n",
            "17 Obo@@ resorọ O Vwo F@@ o N@@ ẹ A “ M@@ ọ I@@ b@@ i Bu@@ eb@@ u ”\n",
            "Asaọkiephana , mi vwo ẹbe rẹ mi si ọrhuẹrẹphiyotọ mẹ phiyọ , ọnana vwẹ ukẹcha kẹ vwẹ vwọ nabọ vwẹrote iruo mẹ , me rha yan@@ jẹ ọvuọvo vwo hẹrhẹ im@@ ib@@ r@@ ẹro ri chekọ bẹ@@ siẹ ọke na vwo re - e . ”\n",
            "1 , 2 . ( a ) Die yen ihwo evo ta siẹrẹ ayen de nyo nẹ Ọghẹnẹ vwo ukoko ?\n",
            "Avwanre cha sa chu@@ rhi rẹ Ọghẹnẹ fikirẹ aye - en yẹrẹ dia ẹbẹre ọvo rẹ akpọ na - a . — Se 1 Pita 2 : 13 - 17 .\n",
            "( Luk 4 : 18 ) Ọnana mudiaphiyọ egbomọphẹ vwọ kẹ ihwo re mu kpo eviẹn ?\n",
            "Ukuotọ rọyen , iniọvo mẹ eje de yono Baibol ji bromaphiyame kerẹ Iseri rẹ Jihova .\n",
            "O muẹro dẹn nẹ oma nabọ vwerhen Jihova kọke kọke rọ da mrẹ oborẹ avwanre davw@@ an te , ra vwọ “ mọ ib@@ i bu@@ eb@@ u ” !\n",
            "Ẹkẹvuọvo , Josẹf ọyen odibo rẹ Jesu ro se dje oma phia vwẹ az@@ a@@ gba - a .\n",
            "\n",
            "==> train.en <==\n",
            "The number of publishers is now about ten times what it was when I began serving here .\n",
            "Similarly , elders should not only encourage and console their brothers with words but also build them up by showing sincere personal interest . ​ — 1 Cor .\n",
            "17 Why We “ Keep Bearing Much Fruit ”\n",
            "Now I have a journal that I keep on my desk to schedule upcoming work , and this helps me to schedule myself , not leaving things till the last minute . ”\n",
            "1 , 2 . ( a ) How do some react to the thought that God has an organization ?\n",
            "We cannot go to the point of disobeying God or violating our Christian neutrality . ​ — Read 1 Peter 2 : 13 - 17 .\n",
            "Did this mean freedom for every literal slave ?\n",
            "Eventually , all my siblings did so and became Jehovah’s Witnesses .\n",
            "How pleased Jehovah will be as he observes our whole - souled efforts to “ keep bearing much fruit ” !\n",
            "Joseph , though , was a disciple , but he could not bring himself to say so openly .\n",
            "\n",
            "==> train.urh <==\n",
            "Ighwoghwota rehẹ ẹkuotọ na enẹna vwẹ ọhwọhwọ ihwe vwo bun vrẹ obo rọ hepha ọke me vwọ ga vwẹ oboyin .\n",
            "( 2 Kọr . 12 : 15 ) Vwẹ idjerhe vuọvo na , vwọ vrẹ ota rẹ unu rẹ ekpako cha vwọ bọn iniọvo na gan , o ji fo nẹ ayen ru obo ro che djephia nẹ ayen vwo ọdavwẹ rayen . ​ — 1 Kọr .\n",
            "17 Oboresorọ O Vwo Fo Nẹ A “ Mọ Ibi Buebu ”\n",
            "Asaọkiephana , mi vwo ẹbe rẹ mi si ọrhuẹrẹphiyotọ mẹ phiyọ , ọnana vwẹ ukẹcha kẹ vwẹ vwọ nabọ vwẹrote iruo mẹ , me rha yanjẹ ọvuọvo vwo hẹrhẹ imibrẹro ri chekọ bẹsiẹ ọke na vwo re - e . ”\n",
            "1 , 2 . ( a ) Die yen ihwo evo ta siẹrẹ ayen de nyo nẹ Ọghẹnẹ vwo ukoko ?\n",
            "Avwanre cha sa churhi rẹ Ọghẹnẹ fikirẹ aye - en yẹrẹ dia ẹbẹre ọvo rẹ akpọ na - a . — Se 1 Pita 2 : 13 - 17 .\n",
            "( Luk 4 : 18 ) Ọnana mudiaphiyọ egbomọphẹ vwọ kẹ ihwo re mu kpo eviẹn ?\n",
            "Ukuotọ rọyen , iniọvo mẹ eje de yono Baibol ji bromaphiyame kerẹ Iseri rẹ Jihova .\n",
            "O muẹro dẹn nẹ oma nabọ vwerhen Jihova kọke kọke rọ da mrẹ oborẹ avwanre davwan te , ra vwọ “ mọ ibi buebu ” !\n",
            "Ẹkẹvuọvo , Josẹf ọyen odibo rẹ Jesu ro se dje oma phia vwẹ azagba - a .\n",
            "==> dev.bpe.en <==\n",
            "These or@@ ch@@ es@@ tra@@ l arrang@@ ements are com@@ po@@ sed in such a way that they will pre@@ p@@ are our heart and mind for the pro@@ gra@@ m to follow .\n",
            "Today he is serving at Bethel .\n",
            "But freedom from what ?\n",
            "A@@ vo@@ id com@@ par@@ ing your new congregation with your pre@@ vi@@ ous one .\n",
            "2 : 16 , 17 .\n",
            "As stated , the v@@ indic@@ ation of Jehovah’s sovereignty is a v@@ ital issue invol@@ ving mankind .\n",
            "That is especially so if our trea@@ ch@@ er@@ ous heart tu@@ g@@ s us in the opp@@ o@@ sit@@ e direction .\n",
            "At times , this resul@@ ted in more money going out than coming in for a peri@@ od of time .\n",
            "How did hope re@@ infor@@ ce No@@ ah’s faith ?\n",
            "What prom@@ p@@ ts a mother to care ten@@ der@@ ly for her new@@ born b@@ ab@@ y ?\n",
            "\n",
            "==> dev.bpe.urh <==\n",
            "E ru u@@ hworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọ@@ rhuẹrẹ@@ phiyọ rẹ ẹdẹ yena .\n",
            "Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "Wọ vwẹ ukoko kpokpọ na vwọ vw@@ an@@ vw@@ en ọ rẹ wo nurhe na - a .\n",
            "2 : 16 , 17 .\n",
            "Kirobo ra t@@ arọ jovwo , eti@@ to rẹ usuon rẹ Jihova , ọyen ota ọghanghanre ro fori nẹ ihworakpọ tẹnrovi .\n",
            "M@@ a rho , udu avwanre rọ vọnre vẹ o@@ phi@@ ẹnvwe na da vuẹ avwanre nẹ e ru obo re chọre .\n",
            "Iruo kpokpọ nana nẹrhẹ a ghwọrọ vrẹ obo re tor@@ ori ọkiọvo .\n",
            "Mavọ yen iphiẹrophiyọ vwọ nẹrhẹ esegbuyota rẹ Noa ganphiyọ ?\n",
            "Die yen mu oni vwọ vwẹrote ọmọ ro ghwe vwiẹ ?\n",
            "\n",
            "==> dev.en <==\n",
            "These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "Today he is serving at Bethel .\n",
            "But freedom from what ?\n",
            "Avoid comparing your new congregation with your previous one .\n",
            "2 : 16 , 17 .\n",
            "As stated , the vindication of Jehovah’s sovereignty is a vital issue involving mankind .\n",
            "That is especially so if our treacherous heart tugs us in the opposite direction .\n",
            "At times , this resulted in more money going out than coming in for a period of time .\n",
            "How did hope reinforce Noah’s faith ?\n",
            "What prompts a mother to care tenderly for her newborn baby ?\n",
            "\n",
            "==> dev.urh <==\n",
            "E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2 : 16 , 17 .\n",
            "Kirobo ra tarọ jovwo , etito rẹ usuon rẹ Jihova , ọyen ota ọghanghanre ro fori nẹ ihworakpọ tẹnrovi .\n",
            "Ma rho , udu avwanre rọ vọnre vẹ ophiẹnvwe na da vuẹ avwanre nẹ e ru obo re chọre .\n",
            "Iruo kpokpọ nana nẹrhẹ a ghwọrọ vrẹ obo re torori ọkiọvo .\n",
            "Mavọ yen iphiẹrophiyọ vwọ nẹrhẹ esegbuyota rẹ Noa ganphiyọ ?\n",
            "Die yen mu oni vwọ vwẹrote ọmọ ro ghwe vwiẹ ?\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "epeCydmCyS8X"
      },
      "source": [
        "\n",
        "\n",
        "---\n",
        "\n",
        "\n",
        "## Installation of JoeyNMT\n",
        "\n",
        "JoeyNMT is a simple, minimalist NMT package which is useful for learning and teaching. Check out the documentation for JoeyNMT [here](https://joeynmt.readthedocs.io)  "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "iBRMm4kMxZ8L",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "0aa61180-5002-4396-b2d2-2e0744e5057b"
      },
      "source": [
        "# Install JoeyNMT\n",
        "! git clone https://github.com/joeynmt/joeynmt.git\n",
        "! cd joeynmt; pip3 install ."
      ],
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "fatal: destination path 'joeynmt' already exists and is not an empty directory.\n",
            "Processing /content/joeynmt\n",
            "Requirement already satisfied: future in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (0.16.0)\n",
            "Requirement already satisfied: pillow in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (4.3.0)\n",
            "Requirement already satisfied: numpy<2.0,>=1.14.5 in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (1.17.4)\n",
            "Requirement already satisfied: setuptools>=41.0.0 in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (42.0.2)\n",
            "Requirement already satisfied: torch>=1.1 in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (1.3.1)\n",
            "Requirement already satisfied: tensorflow>=1.14 in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (1.15.0)\n",
            "Requirement already satisfied: torchtext in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (0.3.1)\n",
            "Requirement already satisfied: sacrebleu>=1.3.6 in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (1.4.3)\n",
            "Requirement already satisfied: subword-nmt in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (0.3.7)\n",
            "Requirement already satisfied: matplotlib in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (3.1.2)\n",
            "Requirement already satisfied: seaborn in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (0.9.0)\n",
            "Requirement already satisfied: pyyaml>=5.1 in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (5.2)\n",
            "Requirement already satisfied: pylint in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (2.4.4)\n",
            "Requirement already satisfied: six==1.12 in /usr/local/lib/python3.6/dist-packages (from joeynmt==0.0.1) (1.12.0)\n",
            "Requirement already satisfied: olefile in /usr/local/lib/python3.6/dist-packages (from pillow->joeynmt==0.0.1) (0.46)\n",
            "Requirement already satisfied: termcolor>=1.1.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (1.1.0)\n",
            "Requirement already satisfied: gast==0.2.2 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (0.2.2)\n",
            "Requirement already satisfied: tensorflow-estimator==1.15.1 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (1.15.1)\n",
            "Requirement already satisfied: opt-einsum>=2.3.2 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (3.1.0)\n",
            "Requirement already satisfied: astor>=0.6.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (0.8.1)\n",
            "Requirement already satisfied: keras-applications>=1.0.8 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (1.0.8)\n",
            "Requirement already satisfied: google-pasta>=0.1.6 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (0.1.8)\n",
            "Requirement already satisfied: keras-preprocessing>=1.0.5 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (1.1.0)\n",
            "Requirement already satisfied: wrapt>=1.11.1 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (1.11.2)\n",
            "Requirement already satisfied: wheel>=0.26 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (0.33.6)\n",
            "Requirement already satisfied: grpcio>=1.8.6 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (1.15.0)\n",
            "Requirement already satisfied: protobuf>=3.6.1 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (3.10.0)\n",
            "Requirement already satisfied: absl-py>=0.7.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (0.8.1)\n",
            "Requirement already satisfied: tensorboard<1.16.0,>=1.15.0 in /usr/local/lib/python3.6/dist-packages (from tensorflow>=1.14->joeynmt==0.0.1) (1.15.0)\n",
            "Requirement already satisfied: tqdm in /usr/local/lib/python3.6/dist-packages (from torchtext->joeynmt==0.0.1) (4.28.1)\n",
            "Requirement already satisfied: requests in /usr/local/lib/python3.6/dist-packages (from torchtext->joeynmt==0.0.1) (2.21.0)\n",
            "Requirement already satisfied: typing in /usr/local/lib/python3.6/dist-packages (from sacrebleu>=1.3.6->joeynmt==0.0.1) (3.6.6)\n",
            "Requirement already satisfied: portalocker in /usr/local/lib/python3.6/dist-packages (from sacrebleu>=1.3.6->joeynmt==0.0.1) (1.5.2)\n",
            "Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in /usr/local/lib/python3.6/dist-packages (from matplotlib->joeynmt==0.0.1) (2.4.5)\n",
            "Requirement already satisfied: kiwisolver>=1.0.1 in /usr/local/lib/python3.6/dist-packages (from matplotlib->joeynmt==0.0.1) (1.1.0)\n",
            "Requirement already satisfied: cycler>=0.10 in /usr/local/lib/python3.6/dist-packages (from matplotlib->joeynmt==0.0.1) (0.10.0)\n",
            "Requirement already satisfied: python-dateutil>=2.1 in /usr/local/lib/python3.6/dist-packages (from matplotlib->joeynmt==0.0.1) (2.6.1)\n",
            "Requirement already satisfied: pandas>=0.15.2 in /usr/local/lib/python3.6/dist-packages (from seaborn->joeynmt==0.0.1) (0.25.3)\n",
            "Requirement already satisfied: scipy>=0.14.0 in /usr/local/lib/python3.6/dist-packages (from seaborn->joeynmt==0.0.1) (1.3.3)\n",
            "Requirement already satisfied: mccabe<0.7,>=0.6 in /usr/local/lib/python3.6/dist-packages (from pylint->joeynmt==0.0.1) (0.6.1)\n",
            "Requirement already satisfied: isort<5,>=4.2.5 in /usr/local/lib/python3.6/dist-packages (from pylint->joeynmt==0.0.1) (4.3.21)\n",
            "Requirement already satisfied: astroid<2.4,>=2.3.0 in /usr/local/lib/python3.6/dist-packages (from pylint->joeynmt==0.0.1) (2.3.3)\n",
            "Requirement already satisfied: h5py in /usr/local/lib/python3.6/dist-packages (from keras-applications>=1.0.8->tensorflow>=1.14->joeynmt==0.0.1) (2.8.0)\n",
            "Requirement already satisfied: werkzeug>=0.11.15 in /usr/local/lib/python3.6/dist-packages (from tensorboard<1.16.0,>=1.15.0->tensorflow>=1.14->joeynmt==0.0.1) (0.16.0)\n",
            "Requirement already satisfied: markdown>=2.6.8 in /usr/local/lib/python3.6/dist-packages (from tensorboard<1.16.0,>=1.15.0->tensorflow>=1.14->joeynmt==0.0.1) (3.1.1)\n",
            "Requirement already satisfied: chardet<3.1.0,>=3.0.2 in /usr/local/lib/python3.6/dist-packages (from requests->torchtext->joeynmt==0.0.1) (3.0.4)\n",
            "Requirement already satisfied: urllib3<1.25,>=1.21.1 in /usr/local/lib/python3.6/dist-packages (from requests->torchtext->joeynmt==0.0.1) (1.24.3)\n",
            "Requirement already satisfied: idna<2.9,>=2.5 in /usr/local/lib/python3.6/dist-packages (from requests->torchtext->joeynmt==0.0.1) (2.8)\n",
            "Requirement already satisfied: certifi>=2017.4.17 in /usr/local/lib/python3.6/dist-packages (from requests->torchtext->joeynmt==0.0.1) (2019.11.28)\n",
            "Requirement already satisfied: pytz>=2017.2 in /usr/local/lib/python3.6/dist-packages (from pandas>=0.15.2->seaborn->joeynmt==0.0.1) (2018.9)\n",
            "Requirement already satisfied: lazy-object-proxy==1.4.* in /usr/local/lib/python3.6/dist-packages (from astroid<2.4,>=2.3.0->pylint->joeynmt==0.0.1) (1.4.3)\n",
            "Requirement already satisfied: typed-ast<1.5,>=1.4.0; implementation_name == \"cpython\" and python_version < \"3.8\" in /usr/local/lib/python3.6/dist-packages (from astroid<2.4,>=2.3.0->pylint->joeynmt==0.0.1) (1.4.0)\n",
            "Building wheels for collected packages: joeynmt\n",
            "  Building wheel for joeynmt (setup.py) ... \u001b[?25l\u001b[?25hdone\n",
            "  Created wheel for joeynmt: filename=joeynmt-0.0.1-cp36-none-any.whl size=72136 sha256=3cc6a4de7274fdcab65f90e3af85b3579607dee1b0258e3b260af28dd3e0bb15\n",
            "  Stored in directory: /tmp/pip-ephem-wheel-cache-dfaqemza/wheels/db/01/db/751cc9f3e7f6faec127c43644ba250a3ea7ad200594aeda70a\n",
            "Successfully built joeynmt\n",
            "Installing collected packages: joeynmt\n",
            "  Found existing installation: joeynmt 0.0.1\n",
            "    Uninstalling joeynmt-0.0.1:\n",
            "      Successfully uninstalled joeynmt-0.0.1\n",
            "Successfully installed joeynmt-0.0.1\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "AaE77Tcppex9"
      },
      "source": [
        "# Preprocessing the Data into Subword BPE Tokens\n",
        "\n",
        "- One of the most powerful improvements for agglutinative languages (a feature of most Bantu languages) is using BPE tokenization [ (Sennrich, 2015) ](https://arxiv.org/abs/1508.07909).\n",
        "\n",
        "- It was also shown that by optimizing the umber of BPE codes we significantly improve results for low-resourced languages [(Sennrich, 2019)](https://www.aclweb.org/anthology/P19-1021) [(Martinus, 2019)](https://arxiv.org/abs/1906.05685)\n",
        "\n",
        "- Below we have the scripts for doing BPE tokenization of our data. We use 4000 tokens as recommended by [(Sennrich, 2019)](https://www.aclweb.org/anthology/P19-1021). You do not need to change anything. Simply running the below will be suitable. "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0DhFg6tlqVW5",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 204
        },
        "outputId": "6cfdc1c4-5eb8-4459-82ef-aee13899953a"
      },
      "source": [
        "!ls drive/'My Drive'/masakhane/en-urh-baseline/models/enurh_transformer"
      ],
      "execution_count": 38,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "00001000.hyps.dev   15000.hyps\t25000.hyps  35000.hyps\t7000.ckpt\n",
            "00001000.hyps.test  16000.hyps\t26000.hyps  36000.hyps\t7000.hyps\n",
            "00007000.hyps.dev   17000.hyps\t27000.hyps  37000.hyps\t8000.hyps\n",
            "00007000.hyps.test  18000.hyps\t28000.hyps  38000.hyps\t9000.hyps\n",
            "10000.hyps\t    19000.hyps\t29000.hyps  39000.hyps\tconfig.yaml\n",
            "1000.ckpt\t    20000.hyps\t30000.hyps  40000.hyps\tsrc_vocab.txt\n",
            "1000.hyps\t    2000.hyps\t3000.hyps   4000.hyps\ttensorboard\n",
            "11000.hyps\t    21000.hyps\t31000.hyps  5000.ckpt\ttrain.log\n",
            "12000.hyps\t    22000.hyps\t32000.hyps  5000.hyps\ttrg_vocab.txt\n",
            "13000.hyps\t    23000.hyps\t33000.hyps  6000.ckpt\tvalidations.txt\n",
            "14000.hyps\t    24000.hyps\t34000.hyps  6000.hyps\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "H-TyjtmXB1mL",
        "colab": {}
      },
      "source": [
        "# ##### IOHAVOC MODIFICATIONS ==>>  WE DO NOT WANT TO DO BPE\n",
        "\n",
        "# # One of the huge boosts in NMT performance was to use a different method of tokenizing. \n",
        "# # Usually, NMT would tokenize by words. However, using a method called BPE gave amazing boosts to performance\n",
        "\n",
        "# # Do subword NMT \n",
        "# from os import path\n",
        "# os.environ[\"src\"] = source_language # Sets them in bash as well, since we often use bash scripts\n",
        "# os.environ[\"tgt\"] = target_language\n",
        "\n",
        "# os.environ[\"data_path\"] = path.join(\"joeynmt\", \"data\", source_language + target_language) # Herman! \n",
        "\n",
        "# # Learn BPEs on the training data.\n",
        "# ! subword-nmt learn-joint-bpe-and-vocab --input train.$src train.$tgt -s 4000 -o bpe.codes.4000 --write-vocabulary vocab.$src vocab.$tgt\n",
        "\n",
        "# # Apply BPE splits to the development and test data.\n",
        "# ! subword-nmt apply-bpe -c bpe.codes.4000 --vocabulary vocab.$src < train.$src > train.bpe.$src\n",
        "# ! subword-nmt apply-bpe -c bpe.codes.4000 --vocabulary vocab.$tgt < train.$tgt > train.bpe.$tgt\n",
        "\n",
        "# ! subword-nmt apply-bpe -c bpe.codes.4000 --vocabulary vocab.$src < dev.$src > dev.bpe.$src\n",
        "# ! subword-nmt apply-bpe -c bpe.codes.4000 --vocabulary vocab.$tgt < dev.$tgt > dev.bpe.$tgt\n",
        "# ! subword-nmt apply-bpe -c bpe.codes.4000 --vocabulary vocab.$src < test.$src > test.bpe.$src\n",
        "# ! subword-nmt apply-bpe -c bpe.codes.4000 --vocabulary vocab.$tgt < test.$tgt > test.bpe.$tgt\n",
        "\n",
        "# # Create directory, move everyone we care about to the correct location\n",
        "# ! mkdir -p $data_path\n",
        "# ! cp train.* $data_path\n",
        "# ! cp test.* $data_path\n",
        "# ! cp dev.* $data_path\n",
        "# ! cp bpe.codes.4000 $data_path\n",
        "# ! ls $data_path\n",
        "\n",
        "# # Also move everything we care about to a mounted location in google drive (relevant if running in colab) at gdrive_path\n",
        "# ! cp train.* \"$gdrive_path\"\n",
        "# ! cp test.* \"$gdrive_path\"\n",
        "# ! cp dev.* \"$gdrive_path\"\n",
        "# ! cp bpe.codes.4000 \"$gdrive_path\"\n",
        "# ! ls \"$gdrive_path\"\n",
        "\n",
        "# # Create that vocab using build_vocab\n",
        "# ! sudo chmod 777 joeynmt/scripts/build_vocab.py\n",
        "# ! joeynmt/scripts/build_vocab.py joeynmt/data/$src$tgt/train.bpe.$src joeynmt/data/$src$tgt/train.bpe.$tgt --output_path joeynmt/data/$src$tgt/vocab.txt\n",
        "\n",
        "# # Some output\n",
        "# ! echo \"BPE Urhobo Sentences\"\n",
        "# ! tail -n 5 test.bpe.$tgt\n",
        "# ! echo \"Combined BPE Vocab\"\n",
        "# ! tail -n 10 joeynmt/data/$src$tgt/vocab.txt  # Herman"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gRiUoc_ryUR8",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 442
        },
        "outputId": "e379ef04-7488-491c-8880-e677943fe724"
      },
      "source": [
        "# ##### IOHAVOC MODIFICATIONS ==>> CREATE THE VOCAB FOR NON-BPE EXPERIMENTS\n",
        "from os import path\n",
        "\n",
        "os.environ[\"src\"] = source_language # Sets them in bash as well, since we often use bash scripts\n",
        "os.environ[\"tgt\"] = target_language\n",
        "os.environ[\"data_path\"] = path.join(\"joeynmt\", \"data\", source_language + target_language) # Herman! \n",
        "\n",
        "# Create directory, move everyone we care about to the correct location\n",
        "! mkdir -p $data_path\n",
        "! cp train.* $data_path\n",
        "! cp test.* $data_path\n",
        "! cp dev.* $data_path\n",
        "! ls $data_path\n",
        "\n",
        "# Also move everything we care about to a mounted location in google drive (relevant if running in colab) at gdrive_path\n",
        "! cp train.* \"$gdrive_path\"\n",
        "! cp test.* \"$gdrive_path\"\n",
        "! cp dev.* \"$gdrive_path\"\n",
        "! ls \"$gdrive_path\"\n",
        "\n",
        "! sudo chmod 777 joeynmt/scripts/build_vocab.py\n",
        "! joeynmt/scripts/build_vocab.py joeynmt/data/$src$tgt/train.$src joeynmt/data/$src$tgt/train.$tgt --output_path joeynmt/data/$src$tgt/vocab-nonBPE.txt\n",
        "\n",
        "# Some output\n",
        "! echo \"Urhobo Sentences\"\n",
        "! tail -n 5 test.$tgt\n",
        "! echo \"Combined Vocab\"\n",
        "! tail -n 10 joeynmt/data/$src$tgt/vocab-nonBPE.txt  # Herman"
      ],
      "execution_count": 40,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "bpe.codes.4000\tdev.urh       test.en-any.en\ttrain.bpe.urh\t  vocab.txt\n",
            "dev.bpe.en\ttest.bpe.en   test.en-any.en.1\ttrain.en\n",
            "dev.bpe.urh\ttest.bpe.urh  test.urh\t\ttrain.urh\n",
            "dev.en\t\ttest.en       train.bpe.en\tvocab-nonBPE.txt\n",
            "bpe.codes.4000\tdev.urh       test.en\t\ttrain.bpe.en   vocab-nonBPE.txt\n",
            "dev.bpe.en\tmodels\t      test.en-any.en\ttrain.bpe.urh\n",
            "dev.bpe.urh\ttest.bpe.en   test.en-any.en.1\ttrain.en\n",
            "dev.en\t\ttest.bpe.urh  test.urh\t\ttrain.urh\n",
            "Urhobo Sentences\n",
            "Diesorọ Hushai vwọ guọnọ uduefiogbere ọ sa vwọ fuevun kẹ Ọghẹnẹ ?\n",
            "Diesorọ ọ vwọ guọnọ uduefiogbere avwanre ke sa fuevun ?\n",
            "Me nẹrhovwo vwọ kẹ uduefiogbere me sa vwọ yọnregan .\n",
            "Enẹna , ẹwẹn rayen kpotọ re , me sa kọn bru ayen ra ọkieje . ” — Se Isẹ 29 : 25 .\n",
            "[ 1 ] ( ẹkorota 7 ) E wene edẹ evo .\n",
            "Combined Vocab\n",
            "devilish\n",
            "mutidia\n",
            "intrusions\n",
            "Motivated\n",
            "slope\n",
            "subtracted\n",
            "concentrations\n",
            "patches\n",
            "blooms\n",
            "ọviẹ\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qdZ_lamIBZva",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!cp joeynmt/data/$src$tgt/vocab-nonBPE.txt \"$gdrive_path\""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "IlMitUHR8Qy-",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 85
        },
        "outputId": "9d926509-f30d-4ccf-98a0-b4eb340974a3"
      },
      "source": [
        "# Also move everything we care about to a mounted location in google drive (relevant if running in colab) at gdrive_path\n",
        "! cp train.* \"$gdrive_path\"\n",
        "! cp test.* \"$gdrive_path\"\n",
        "! cp dev.* \"$gdrive_path\"\n",
        "! cp bpe.codes.4000 \"$gdrive_path\"\n",
        "! ls \"$gdrive_path\""
      ],
      "execution_count": 42,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "bpe.codes.4000\tdev.urh       test.en\t\ttrain.bpe.en   vocab-nonBPE.txt\n",
            "dev.bpe.en\tmodels\t      test.en-any.en\ttrain.bpe.urh\n",
            "dev.bpe.urh\ttest.bpe.en   test.en-any.en.1\ttrain.en\n",
            "dev.en\t\ttest.bpe.urh  test.urh\t\ttrain.urh\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "Ixmzi60WsUZ8"
      },
      "source": [
        "# Creating the JoeyNMT Config\n",
        "\n",
        "JoeyNMT requires a yaml config. We provide a template below. We've also set a number of defaults with it, that you may play with!\n",
        "\n",
        "- We used Transformer architecture \n",
        "- We set our dropout to reasonably high: 0.3 (recommended in  [(Sennrich, 2019)](https://www.aclweb.org/anthology/P19-1021))\n",
        "\n",
        "Things worth playing with:\n",
        "- The batch size (also recommended to change for low-resourced languages)\n",
        "- The number of epochs (we've set it at 30 just so it runs in about an hour, for testing purposes)\n",
        "- The decoder options (beam_size, alpha)\n",
        "- Evaluation metrics (BLEU versus Crhf4)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "PIs1lY2hxMsl",
        "colab": {}
      },
      "source": [
        "# This creates the config file for our JoeyNMT system. It might seem overwhelming so we've provided a couple of useful parameters you'll need to update\n",
        "# (You can of course play with all the parameters if you'd like!)\n",
        "\n",
        "name = '%s%s' % (source_language, target_language)\n",
        "gdrive_path = os.environ[\"gdrive_path\"]\n",
        "\n",
        "# Create the config\n",
        "config = \"\"\"\n",
        "name: \"{name}_transformer\"\n",
        "\n",
        "data:\n",
        "    src: \"{source_language}\"\n",
        "    trg: \"{target_language}\"\n",
        "    train: \"data/{name}/train\"\n",
        "    dev:   \"data/{name}/dev\"\n",
        "    test:  \"data/{name}/test\"\n",
        "    level: \"word\"\n",
        "    lowercase: False\n",
        "    max_sent_length: 100\n",
        "    src_vocab: \"data/{name}/vocab-nonBPE.txt\"\n",
        "    trg_vocab: \"data/{name}/vocab-nonBPE.txt\"\n",
        "\n",
        "testing:\n",
        "    beam_size: 5\n",
        "    alpha: 1.0\n",
        "\n",
        "training:\n",
        "    #load_model: \"{gdrive_path}/models/{name}_transformer/1.ckpt\" # if uncommented, load a pre-trained model from this checkpoint\n",
        "    random_seed: 42\n",
        "    optimizer: \"adam\"\n",
        "    normalization: \"tokens\"\n",
        "    adam_betas: [0.9, 0.999] \n",
        "    scheduling: \"plateau\"           # TODO: try switching from plateau to Noam scheduling\n",
        "    patience: 5                     # For plateau: decrease learning rate by decrease_factor if validation score has not improved for this many validation rounds.\n",
        "    learning_rate_factor: 0.5       # factor for Noam scheduler (used with Transformer)\n",
        "    learning_rate_warmup: 1000      # warmup steps for Noam scheduler (used with Transformer)\n",
        "    decrease_factor: 0.7\n",
        "    loss: \"crossentropy\"\n",
        "    learning_rate: 0.0003\n",
        "    learning_rate_min: 0.00000001\n",
        "    weight_decay: 0.0\n",
        "    label_smoothing: 0.1\n",
        "    batch_size: 4096\n",
        "    batch_type: \"token\"\n",
        "    eval_batch_size: 3600\n",
        "    eval_batch_type: \"token\"\n",
        "    batch_multiplier: 1\n",
        "    early_stopping_metric: \"ppl\"\n",
        "    epochs: 150                     # TODO: Decrease for when playing around and checking of working. Around 30 is sufficient to check if its working at all\n",
        "    validation_freq: 1000           # TODO: Set to at least once per epoch.\n",
        "    logging_freq: 100\n",
        "    eval_metric: \"bleu\"\n",
        "    model_dir: \"models/{name}_transformer\"\n",
        "    overwrite: True                # TODO: Set to True if you want to overwrite possibly existing models. \n",
        "    shuffle: True\n",
        "    use_cuda: True\n",
        "    max_output_length: 100\n",
        "    print_valid_sents: [0, 1, 2, 3]\n",
        "    keep_last_ckpts: 3\n",
        "\n",
        "model:\n",
        "    initializer: \"xavier\"\n",
        "    bias_initializer: \"zeros\"\n",
        "    init_gain: 1.0\n",
        "    embed_initializer: \"xavier\"\n",
        "    embed_init_gain: 1.0\n",
        "    tied_embeddings: True\n",
        "    tied_softmax: True\n",
        "    encoder:\n",
        "        type: \"transformer\"\n",
        "        num_layers: 6\n",
        "        num_heads: 4             # TODO: Increase to 8 for larger data.\n",
        "        embeddings:\n",
        "            embedding_dim: 256   # TODO: Increase to 512 for larger data.\n",
        "            scale: True\n",
        "            dropout: 0.2\n",
        "        # typically ff_size = 4 x hidden_size\n",
        "        hidden_size: 256         # TODO: Increase to 512 for larger data.\n",
        "        ff_size: 1024            # TODO: Increase to 2048 for larger data.\n",
        "        dropout: 0.3\n",
        "    decoder:\n",
        "        type: \"transformer\"\n",
        "        num_layers: 6\n",
        "        num_heads: 4              # TODO: Increase to 8 for larger data.\n",
        "        embeddings:\n",
        "            embedding_dim: 256    # TODO: Increase to 512 for larger data.\n",
        "            scale: True\n",
        "            dropout: 0.2\n",
        "        # typically ff_size = 4 x hidden_size\n",
        "        hidden_size: 256         # TODO: Increase to 512 for larger data.\n",
        "        ff_size: 1024            # TODO: Increase to 2048 for larger data.\n",
        "        dropout: 0.3\n",
        "\"\"\".format(name=name, gdrive_path=os.environ[\"gdrive_path\"], source_language=source_language, target_language=target_language)\n",
        "with open(\"joeynmt/configs/transformer_{name}.yaml\".format(name=name),'w') as f:\n",
        "    f.write(config)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "colab_type": "text",
        "id": "pIifxE3Qzuvs"
      },
      "source": [
        "# Train the Model\n",
        "\n",
        "This single line of joeynmt runs the training using the config we made above"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "6ZBPFwT94WpI",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "06325004-515e-4e59-c1d5-e7298d667d68"
      },
      "source": [
        "# Train the model\n",
        "# You can press Ctrl-C to stop. And then run the next cell to save your checkpoints! \n",
        "!cd joeynmt; python3 -m joeynmt train configs/transformer_$src$tgt.yaml"
      ],
      "execution_count": 44,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "2019-12-31 06:52:11,457 Hello! This is Joey-NMT.\n",
            "2019-12-31 06:52:12,772 Total params: 16802560\n",
            "2019-12-31 06:52:12,773 Trainable parameters: ['decoder.layer_norm.bias', 'decoder.layer_norm.weight', 'decoder.layers.0.dec_layer_norm.bias', 'decoder.layers.0.dec_layer_norm.weight', 'decoder.layers.0.feed_forward.layer_norm.bias', 'decoder.layers.0.feed_forward.layer_norm.weight', 'decoder.layers.0.feed_forward.pwff_layer.0.bias', 'decoder.layers.0.feed_forward.pwff_layer.0.weight', 'decoder.layers.0.feed_forward.pwff_layer.3.bias', 'decoder.layers.0.feed_forward.pwff_layer.3.weight', 'decoder.layers.0.src_trg_att.k_layer.bias', 'decoder.layers.0.src_trg_att.k_layer.weight', 'decoder.layers.0.src_trg_att.output_layer.bias', 'decoder.layers.0.src_trg_att.output_layer.weight', 'decoder.layers.0.src_trg_att.q_layer.bias', 'decoder.layers.0.src_trg_att.q_layer.weight', 'decoder.layers.0.src_trg_att.v_layer.bias', 'decoder.layers.0.src_trg_att.v_layer.weight', 'decoder.layers.0.trg_trg_att.k_layer.bias', 'decoder.layers.0.trg_trg_att.k_layer.weight', 'decoder.layers.0.trg_trg_att.output_layer.bias', 'decoder.layers.0.trg_trg_att.output_layer.weight', 'decoder.layers.0.trg_trg_att.q_layer.bias', 'decoder.layers.0.trg_trg_att.q_layer.weight', 'decoder.layers.0.trg_trg_att.v_layer.bias', 'decoder.layers.0.trg_trg_att.v_layer.weight', 'decoder.layers.0.x_layer_norm.bias', 'decoder.layers.0.x_layer_norm.weight', 'decoder.layers.1.dec_layer_norm.bias', 'decoder.layers.1.dec_layer_norm.weight', 'decoder.layers.1.feed_forward.layer_norm.bias', 'decoder.layers.1.feed_forward.layer_norm.weight', 'decoder.layers.1.feed_forward.pwff_layer.0.bias', 'decoder.layers.1.feed_forward.pwff_layer.0.weight', 'decoder.layers.1.feed_forward.pwff_layer.3.bias', 'decoder.layers.1.feed_forward.pwff_layer.3.weight', 'decoder.layers.1.src_trg_att.k_layer.bias', 'decoder.layers.1.src_trg_att.k_layer.weight', 'decoder.layers.1.src_trg_att.output_layer.bias', 'decoder.layers.1.src_trg_att.output_layer.weight', 'decoder.layers.1.src_trg_att.q_layer.bias', 'decoder.layers.1.src_trg_att.q_layer.weight', 'decoder.layers.1.src_trg_att.v_layer.bias', 'decoder.layers.1.src_trg_att.v_layer.weight', 'decoder.layers.1.trg_trg_att.k_layer.bias', 'decoder.layers.1.trg_trg_att.k_layer.weight', 'decoder.layers.1.trg_trg_att.output_layer.bias', 'decoder.layers.1.trg_trg_att.output_layer.weight', 'decoder.layers.1.trg_trg_att.q_layer.bias', 'decoder.layers.1.trg_trg_att.q_layer.weight', 'decoder.layers.1.trg_trg_att.v_layer.bias', 'decoder.layers.1.trg_trg_att.v_layer.weight', 'decoder.layers.1.x_layer_norm.bias', 'decoder.layers.1.x_layer_norm.weight', 'decoder.layers.2.dec_layer_norm.bias', 'decoder.layers.2.dec_layer_norm.weight', 'decoder.layers.2.feed_forward.layer_norm.bias', 'decoder.layers.2.feed_forward.layer_norm.weight', 'decoder.layers.2.feed_forward.pwff_layer.0.bias', 'decoder.layers.2.feed_forward.pwff_layer.0.weight', 'decoder.layers.2.feed_forward.pwff_layer.3.bias', 'decoder.layers.2.feed_forward.pwff_layer.3.weight', 'decoder.layers.2.src_trg_att.k_layer.bias', 'decoder.layers.2.src_trg_att.k_layer.weight', 'decoder.layers.2.src_trg_att.output_layer.bias', 'decoder.layers.2.src_trg_att.output_layer.weight', 'decoder.layers.2.src_trg_att.q_layer.bias', 'decoder.layers.2.src_trg_att.q_layer.weight', 'decoder.layers.2.src_trg_att.v_layer.bias', 'decoder.layers.2.src_trg_att.v_layer.weight', 'decoder.layers.2.trg_trg_att.k_layer.bias', 'decoder.layers.2.trg_trg_att.k_layer.weight', 'decoder.layers.2.trg_trg_att.output_layer.bias', 'decoder.layers.2.trg_trg_att.output_layer.weight', 'decoder.layers.2.trg_trg_att.q_layer.bias', 'decoder.layers.2.trg_trg_att.q_layer.weight', 'decoder.layers.2.trg_trg_att.v_layer.bias', 'decoder.layers.2.trg_trg_att.v_layer.weight', 'decoder.layers.2.x_layer_norm.bias', 'decoder.layers.2.x_layer_norm.weight', 'decoder.layers.3.dec_layer_norm.bias', 'decoder.layers.3.dec_layer_norm.weight', 'decoder.layers.3.feed_forward.layer_norm.bias', 'decoder.layers.3.feed_forward.layer_norm.weight', 'decoder.layers.3.feed_forward.pwff_layer.0.bias', 'decoder.layers.3.feed_forward.pwff_layer.0.weight', 'decoder.layers.3.feed_forward.pwff_layer.3.bias', 'decoder.layers.3.feed_forward.pwff_layer.3.weight', 'decoder.layers.3.src_trg_att.k_layer.bias', 'decoder.layers.3.src_trg_att.k_layer.weight', 'decoder.layers.3.src_trg_att.output_layer.bias', 'decoder.layers.3.src_trg_att.output_layer.weight', 'decoder.layers.3.src_trg_att.q_layer.bias', 'decoder.layers.3.src_trg_att.q_layer.weight', 'decoder.layers.3.src_trg_att.v_layer.bias', 'decoder.layers.3.src_trg_att.v_layer.weight', 'decoder.layers.3.trg_trg_att.k_layer.bias', 'decoder.layers.3.trg_trg_att.k_layer.weight', 'decoder.layers.3.trg_trg_att.output_layer.bias', 'decoder.layers.3.trg_trg_att.output_layer.weight', 'decoder.layers.3.trg_trg_att.q_layer.bias', 'decoder.layers.3.trg_trg_att.q_layer.weight', 'decoder.layers.3.trg_trg_att.v_layer.bias', 'decoder.layers.3.trg_trg_att.v_layer.weight', 'decoder.layers.3.x_layer_norm.bias', 'decoder.layers.3.x_layer_norm.weight', 'decoder.layers.4.dec_layer_norm.bias', 'decoder.layers.4.dec_layer_norm.weight', 'decoder.layers.4.feed_forward.layer_norm.bias', 'decoder.layers.4.feed_forward.layer_norm.weight', 'decoder.layers.4.feed_forward.pwff_layer.0.bias', 'decoder.layers.4.feed_forward.pwff_layer.0.weight', 'decoder.layers.4.feed_forward.pwff_layer.3.bias', 'decoder.layers.4.feed_forward.pwff_layer.3.weight', 'decoder.layers.4.src_trg_att.k_layer.bias', 'decoder.layers.4.src_trg_att.k_layer.weight', 'decoder.layers.4.src_trg_att.output_layer.bias', 'decoder.layers.4.src_trg_att.output_layer.weight', 'decoder.layers.4.src_trg_att.q_layer.bias', 'decoder.layers.4.src_trg_att.q_layer.weight', 'decoder.layers.4.src_trg_att.v_layer.bias', 'decoder.layers.4.src_trg_att.v_layer.weight', 'decoder.layers.4.trg_trg_att.k_layer.bias', 'decoder.layers.4.trg_trg_att.k_layer.weight', 'decoder.layers.4.trg_trg_att.output_layer.bias', 'decoder.layers.4.trg_trg_att.output_layer.weight', 'decoder.layers.4.trg_trg_att.q_layer.bias', 'decoder.layers.4.trg_trg_att.q_layer.weight', 'decoder.layers.4.trg_trg_att.v_layer.bias', 'decoder.layers.4.trg_trg_att.v_layer.weight', 'decoder.layers.4.x_layer_norm.bias', 'decoder.layers.4.x_layer_norm.weight', 'decoder.layers.5.dec_layer_norm.bias', 'decoder.layers.5.dec_layer_norm.weight', 'decoder.layers.5.feed_forward.layer_norm.bias', 'decoder.layers.5.feed_forward.layer_norm.weight', 'decoder.layers.5.feed_forward.pwff_layer.0.bias', 'decoder.layers.5.feed_forward.pwff_layer.0.weight', 'decoder.layers.5.feed_forward.pwff_layer.3.bias', 'decoder.layers.5.feed_forward.pwff_layer.3.weight', 'decoder.layers.5.src_trg_att.k_layer.bias', 'decoder.layers.5.src_trg_att.k_layer.weight', 'decoder.layers.5.src_trg_att.output_layer.bias', 'decoder.layers.5.src_trg_att.output_layer.weight', 'decoder.layers.5.src_trg_att.q_layer.bias', 'decoder.layers.5.src_trg_att.q_layer.weight', 'decoder.layers.5.src_trg_att.v_layer.bias', 'decoder.layers.5.src_trg_att.v_layer.weight', 'decoder.layers.5.trg_trg_att.k_layer.bias', 'decoder.layers.5.trg_trg_att.k_layer.weight', 'decoder.layers.5.trg_trg_att.output_layer.bias', 'decoder.layers.5.trg_trg_att.output_layer.weight', 'decoder.layers.5.trg_trg_att.q_layer.bias', 'decoder.layers.5.trg_trg_att.q_layer.weight', 'decoder.layers.5.trg_trg_att.v_layer.bias', 'decoder.layers.5.trg_trg_att.v_layer.weight', 'decoder.layers.5.x_layer_norm.bias', 'decoder.layers.5.x_layer_norm.weight', 'encoder.layer_norm.bias', 'encoder.layer_norm.weight', 'encoder.layers.0.feed_forward.layer_norm.bias', 'encoder.layers.0.feed_forward.layer_norm.weight', 'encoder.layers.0.feed_forward.pwff_layer.0.bias', 'encoder.layers.0.feed_forward.pwff_layer.0.weight', 'encoder.layers.0.feed_forward.pwff_layer.3.bias', 'encoder.layers.0.feed_forward.pwff_layer.3.weight', 'encoder.layers.0.layer_norm.bias', 'encoder.layers.0.layer_norm.weight', 'encoder.layers.0.src_src_att.k_layer.bias', 'encoder.layers.0.src_src_att.k_layer.weight', 'encoder.layers.0.src_src_att.output_layer.bias', 'encoder.layers.0.src_src_att.output_layer.weight', 'encoder.layers.0.src_src_att.q_layer.bias', 'encoder.layers.0.src_src_att.q_layer.weight', 'encoder.layers.0.src_src_att.v_layer.bias', 'encoder.layers.0.src_src_att.v_layer.weight', 'encoder.layers.1.feed_forward.layer_norm.bias', 'encoder.layers.1.feed_forward.layer_norm.weight', 'encoder.layers.1.feed_forward.pwff_layer.0.bias', 'encoder.layers.1.feed_forward.pwff_layer.0.weight', 'encoder.layers.1.feed_forward.pwff_layer.3.bias', 'encoder.layers.1.feed_forward.pwff_layer.3.weight', 'encoder.layers.1.layer_norm.bias', 'encoder.layers.1.layer_norm.weight', 'encoder.layers.1.src_src_att.k_layer.bias', 'encoder.layers.1.src_src_att.k_layer.weight', 'encoder.layers.1.src_src_att.output_layer.bias', 'encoder.layers.1.src_src_att.output_layer.weight', 'encoder.layers.1.src_src_att.q_layer.bias', 'encoder.layers.1.src_src_att.q_layer.weight', 'encoder.layers.1.src_src_att.v_layer.bias', 'encoder.layers.1.src_src_att.v_layer.weight', 'encoder.layers.2.feed_forward.layer_norm.bias', 'encoder.layers.2.feed_forward.layer_norm.weight', 'encoder.layers.2.feed_forward.pwff_layer.0.bias', 'encoder.layers.2.feed_forward.pwff_layer.0.weight', 'encoder.layers.2.feed_forward.pwff_layer.3.bias', 'encoder.layers.2.feed_forward.pwff_layer.3.weight', 'encoder.layers.2.layer_norm.bias', 'encoder.layers.2.layer_norm.weight', 'encoder.layers.2.src_src_att.k_layer.bias', 'encoder.layers.2.src_src_att.k_layer.weight', 'encoder.layers.2.src_src_att.output_layer.bias', 'encoder.layers.2.src_src_att.output_layer.weight', 'encoder.layers.2.src_src_att.q_layer.bias', 'encoder.layers.2.src_src_att.q_layer.weight', 'encoder.layers.2.src_src_att.v_layer.bias', 'encoder.layers.2.src_src_att.v_layer.weight', 'encoder.layers.3.feed_forward.layer_norm.bias', 'encoder.layers.3.feed_forward.layer_norm.weight', 'encoder.layers.3.feed_forward.pwff_layer.0.bias', 'encoder.layers.3.feed_forward.pwff_layer.0.weight', 'encoder.layers.3.feed_forward.pwff_layer.3.bias', 'encoder.layers.3.feed_forward.pwff_layer.3.weight', 'encoder.layers.3.layer_norm.bias', 'encoder.layers.3.layer_norm.weight', 'encoder.layers.3.src_src_att.k_layer.bias', 'encoder.layers.3.src_src_att.k_layer.weight', 'encoder.layers.3.src_src_att.output_layer.bias', 'encoder.layers.3.src_src_att.output_layer.weight', 'encoder.layers.3.src_src_att.q_layer.bias', 'encoder.layers.3.src_src_att.q_layer.weight', 'encoder.layers.3.src_src_att.v_layer.bias', 'encoder.layers.3.src_src_att.v_layer.weight', 'encoder.layers.4.feed_forward.layer_norm.bias', 'encoder.layers.4.feed_forward.layer_norm.weight', 'encoder.layers.4.feed_forward.pwff_layer.0.bias', 'encoder.layers.4.feed_forward.pwff_layer.0.weight', 'encoder.layers.4.feed_forward.pwff_layer.3.bias', 'encoder.layers.4.feed_forward.pwff_layer.3.weight', 'encoder.layers.4.layer_norm.bias', 'encoder.layers.4.layer_norm.weight', 'encoder.layers.4.src_src_att.k_layer.bias', 'encoder.layers.4.src_src_att.k_layer.weight', 'encoder.layers.4.src_src_att.output_layer.bias', 'encoder.layers.4.src_src_att.output_layer.weight', 'encoder.layers.4.src_src_att.q_layer.bias', 'encoder.layers.4.src_src_att.q_layer.weight', 'encoder.layers.4.src_src_att.v_layer.bias', 'encoder.layers.4.src_src_att.v_layer.weight', 'encoder.layers.5.feed_forward.layer_norm.bias', 'encoder.layers.5.feed_forward.layer_norm.weight', 'encoder.layers.5.feed_forward.pwff_layer.0.bias', 'encoder.layers.5.feed_forward.pwff_layer.0.weight', 'encoder.layers.5.feed_forward.pwff_layer.3.bias', 'encoder.layers.5.feed_forward.pwff_layer.3.weight', 'encoder.layers.5.layer_norm.bias', 'encoder.layers.5.layer_norm.weight', 'encoder.layers.5.src_src_att.k_layer.bias', 'encoder.layers.5.src_src_att.k_layer.weight', 'encoder.layers.5.src_src_att.output_layer.bias', 'encoder.layers.5.src_src_att.output_layer.weight', 'encoder.layers.5.src_src_att.q_layer.bias', 'encoder.layers.5.src_src_att.q_layer.weight', 'encoder.layers.5.src_src_att.v_layer.bias', 'encoder.layers.5.src_src_att.v_layer.weight', 'src_embed.lut.weight']\n",
            "2019-12-31 06:52:15,793 cfg.name                           : enurh_transformer\n",
            "2019-12-31 06:52:15,793 cfg.data.src                       : en\n",
            "2019-12-31 06:52:15,793 cfg.data.trg                       : urh\n",
            "2019-12-31 06:52:15,793 cfg.data.train                     : data/enurh/train\n",
            "2019-12-31 06:52:15,793 cfg.data.dev                       : data/enurh/dev\n",
            "2019-12-31 06:52:15,793 cfg.data.test                      : data/enurh/test\n",
            "2019-12-31 06:52:15,793 cfg.data.level                     : word\n",
            "2019-12-31 06:52:15,793 cfg.data.lowercase                 : False\n",
            "2019-12-31 06:52:15,793 cfg.data.max_sent_length           : 100\n",
            "2019-12-31 06:52:15,793 cfg.data.src_vocab                 : data/enurh/vocab-nonBPE.txt\n",
            "2019-12-31 06:52:15,793 cfg.data.trg_vocab                 : data/enurh/vocab-nonBPE.txt\n",
            "2019-12-31 06:52:15,793 cfg.testing.beam_size              : 5\n",
            "2019-12-31 06:52:15,793 cfg.testing.alpha                  : 1.0\n",
            "2019-12-31 06:52:15,793 cfg.training.random_seed           : 42\n",
            "2019-12-31 06:52:15,793 cfg.training.optimizer             : adam\n",
            "2019-12-31 06:52:15,793 cfg.training.normalization         : tokens\n",
            "2019-12-31 06:52:15,793 cfg.training.adam_betas            : [0.9, 0.999]\n",
            "2019-12-31 06:52:15,794 cfg.training.scheduling            : plateau\n",
            "2019-12-31 06:52:15,794 cfg.training.patience              : 5\n",
            "2019-12-31 06:52:15,794 cfg.training.learning_rate_factor  : 0.5\n",
            "2019-12-31 06:52:15,794 cfg.training.learning_rate_warmup  : 1000\n",
            "2019-12-31 06:52:15,794 cfg.training.decrease_factor       : 0.7\n",
            "2019-12-31 06:52:15,794 cfg.training.loss                  : crossentropy\n",
            "2019-12-31 06:52:15,794 cfg.training.learning_rate         : 0.0003\n",
            "2019-12-31 06:52:15,794 cfg.training.learning_rate_min     : 1e-08\n",
            "2019-12-31 06:52:15,794 cfg.training.weight_decay          : 0.0\n",
            "2019-12-31 06:52:15,794 cfg.training.label_smoothing       : 0.1\n",
            "2019-12-31 06:52:15,794 cfg.training.batch_size            : 4096\n",
            "2019-12-31 06:52:15,794 cfg.training.batch_type            : token\n",
            "2019-12-31 06:52:15,794 cfg.training.eval_batch_size       : 3600\n",
            "2019-12-31 06:52:15,794 cfg.training.eval_batch_type       : token\n",
            "2019-12-31 06:52:15,794 cfg.training.batch_multiplier      : 1\n",
            "2019-12-31 06:52:15,794 cfg.training.early_stopping_metric : ppl\n",
            "2019-12-31 06:52:15,794 cfg.training.epochs                : 150\n",
            "2019-12-31 06:52:15,794 cfg.training.validation_freq       : 1000\n",
            "2019-12-31 06:52:15,794 cfg.training.logging_freq          : 100\n",
            "2019-12-31 06:52:15,794 cfg.training.eval_metric           : bleu\n",
            "2019-12-31 06:52:15,794 cfg.training.model_dir             : models/enurh_transformer\n",
            "2019-12-31 06:52:15,794 cfg.training.overwrite             : True\n",
            "2019-12-31 06:52:15,794 cfg.training.shuffle               : True\n",
            "2019-12-31 06:52:15,794 cfg.training.use_cuda              : True\n",
            "2019-12-31 06:52:15,795 cfg.training.max_output_length     : 100\n",
            "2019-12-31 06:52:15,795 cfg.training.print_valid_sents     : [0, 1, 2, 3]\n",
            "2019-12-31 06:52:15,795 cfg.training.keep_last_ckpts       : 3\n",
            "2019-12-31 06:52:15,795 cfg.model.initializer              : xavier\n",
            "2019-12-31 06:52:15,795 cfg.model.bias_initializer         : zeros\n",
            "2019-12-31 06:52:15,795 cfg.model.init_gain                : 1.0\n",
            "2019-12-31 06:52:15,795 cfg.model.embed_initializer        : xavier\n",
            "2019-12-31 06:52:15,795 cfg.model.embed_init_gain          : 1.0\n",
            "2019-12-31 06:52:15,795 cfg.model.tied_embeddings          : True\n",
            "2019-12-31 06:52:15,795 cfg.model.tied_softmax             : True\n",
            "2019-12-31 06:52:15,795 cfg.model.encoder.type             : transformer\n",
            "2019-12-31 06:52:15,795 cfg.model.encoder.num_layers       : 6\n",
            "2019-12-31 06:52:15,795 cfg.model.encoder.num_heads        : 4\n",
            "2019-12-31 06:52:15,795 cfg.model.encoder.embeddings.embedding_dim : 256\n",
            "2019-12-31 06:52:15,795 cfg.model.encoder.embeddings.scale : True\n",
            "2019-12-31 06:52:15,795 cfg.model.encoder.embeddings.dropout : 0.2\n",
            "2019-12-31 06:52:15,795 cfg.model.encoder.hidden_size      : 256\n",
            "2019-12-31 06:52:15,795 cfg.model.encoder.ff_size          : 1024\n",
            "2019-12-31 06:52:15,795 cfg.model.encoder.dropout          : 0.3\n",
            "2019-12-31 06:52:15,795 cfg.model.decoder.type             : transformer\n",
            "2019-12-31 06:52:15,795 cfg.model.decoder.num_layers       : 6\n",
            "2019-12-31 06:52:15,795 cfg.model.decoder.num_heads        : 4\n",
            "2019-12-31 06:52:15,795 cfg.model.decoder.embeddings.embedding_dim : 256\n",
            "2019-12-31 06:52:15,796 cfg.model.decoder.embeddings.scale : True\n",
            "2019-12-31 06:52:15,796 cfg.model.decoder.embeddings.dropout : 0.2\n",
            "2019-12-31 06:52:15,796 cfg.model.decoder.hidden_size      : 256\n",
            "2019-12-31 06:52:15,796 cfg.model.decoder.ff_size          : 1024\n",
            "2019-12-31 06:52:15,796 cfg.model.decoder.dropout          : 0.3\n",
            "2019-12-31 06:52:15,796 Data set sizes: \n",
            "\ttrain 25608,\n",
            "\tvalid 1000,\n",
            "\ttest 2652\n",
            "2019-12-31 06:52:15,796 First training example:\n",
            "\t[SRC] The number of publishers is now about ten times what it was when I began serving here .\n",
            "\t[TRG] Ighwoghwota rehẹ ẹkuotọ na enẹna vwẹ ọhwọhwọ ihwe vwo bun vrẹ obo rọ hepha ọke me vwọ ga vwẹ oboyin .\n",
            "2019-12-31 06:52:15,796 First 10 words (src): (0) <unk> (1) <pad> (2) <s> (3) </s> (4) . (5) , (6) rẹ (7) the (8) to (9) na\n",
            "2019-12-31 06:52:15,796 First 10 words (trg): (0) <unk> (1) <pad> (2) <s> (3) </s> (4) . (5) , (6) rẹ (7) the (8) to (9) na\n",
            "2019-12-31 06:52:15,796 Number of Src words (types): 22431\n",
            "2019-12-31 06:52:15,797 Number of Trg words (types): 22431\n",
            "2019-12-31 06:52:15,797 Model(\n",
            "\tencoder=TransformerEncoder(num_layers=6, num_heads=4),\n",
            "\tdecoder=TransformerDecoder(num_layers=6, num_heads=4),\n",
            "\tsrc_embed=Embeddings(embedding_dim=256, vocab_size=22431),\n",
            "\ttrg_embed=Embeddings(embedding_dim=256, vocab_size=22431))\n",
            "2019-12-31 06:52:15,812 EPOCH 1\n",
            "2019-12-31 06:52:30,374 Epoch   1 Step:      100 Batch Loss:     5.427601 Tokens per Sec:    13843, Lr: 0.000300\n",
            "2019-12-31 06:52:45,093 Epoch   1 Step:      200 Batch Loss:     5.082472 Tokens per Sec:    13882, Lr: 0.000300\n",
            "2019-12-31 06:52:55,258 Epoch   1: total training loss 1463.09\n",
            "2019-12-31 06:52:55,258 EPOCH 2\n",
            "2019-12-31 06:52:59,757 Epoch   2 Step:      300 Batch Loss:     4.783172 Tokens per Sec:    13252, Lr: 0.000300\n",
            "2019-12-31 06:53:14,400 Epoch   2 Step:      400 Batch Loss:     4.720972 Tokens per Sec:    13727, Lr: 0.000300\n",
            "2019-12-31 06:53:29,098 Epoch   2 Step:      500 Batch Loss:     4.665186 Tokens per Sec:    13964, Lr: 0.000300\n",
            "2019-12-31 06:53:34,775 Epoch   2: total training loss 1238.50\n",
            "2019-12-31 06:53:34,775 EPOCH 3\n",
            "2019-12-31 06:53:43,729 Epoch   3 Step:      600 Batch Loss:     4.156670 Tokens per Sec:    13992, Lr: 0.000300\n",
            "2019-12-31 06:53:58,306 Epoch   3 Step:      700 Batch Loss:     4.177386 Tokens per Sec:    13452, Lr: 0.000300\n",
            "2019-12-31 06:54:12,836 Epoch   3 Step:      800 Batch Loss:     3.818682 Tokens per Sec:    14428, Lr: 0.000300\n",
            "2019-12-31 06:54:14,015 Epoch   3: total training loss 1105.01\n",
            "2019-12-31 06:54:14,015 EPOCH 4\n",
            "2019-12-31 06:54:27,480 Epoch   4 Step:      900 Batch Loss:     3.945844 Tokens per Sec:    13672, Lr: 0.000300\n",
            "2019-12-31 06:54:42,134 Epoch   4 Step:     1000 Batch Loss:     3.845206 Tokens per Sec:    14279, Lr: 0.000300\n",
            "2019-12-31 06:55:15,730 Hooray! New best validation result [ppl]!\n",
            "2019-12-31 06:55:15,730 Saving new checkpoint.\n",
            "2019-12-31 06:55:16,062 Example #0\n",
            "2019-12-31 06:55:16,063 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 06:55:16,063 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 06:55:16,063 \tHypothesis: ( 1 Jọn 1 : 1 ) Ọ sa dianẹ avwanre vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo ẹguọnọ rẹ avwanre vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo ẹguọnọ rẹ Jihova .\n",
            "2019-12-31 06:55:16,063 Example #1\n",
            "2019-12-31 06:55:16,063 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 06:55:16,063 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 06:55:16,063 \tHypothesis: Ọ da ta : “ Ọ da ta : “ O vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo ẹguọnọ rẹ Ọghẹnẹ .\n",
            "2019-12-31 06:55:16,063 Example #2\n",
            "2019-12-31 06:55:16,063 \tSource:     But freedom from what ?\n",
            "2019-12-31 06:55:16,063 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 06:55:16,063 \tHypothesis: Die yen avwanre vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo ?\n",
            "2019-12-31 06:55:16,063 Example #3\n",
            "2019-12-31 06:55:16,063 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 06:55:16,063 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 06:55:16,063 \tHypothesis: Ọ da dianẹ avwanre vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo vwo ẹguọnọ rẹ Jihova .\n",
            "2019-12-31 06:55:16,063 Validation result (greedy) at epoch   4, step     1000: bleu:   0.44, loss: 78227.9688, ppl:  42.1249, duration: 33.9293s\n",
            "2019-12-31 06:55:27,124 Epoch   4: total training loss 1015.77\n",
            "2019-12-31 06:55:27,124 EPOCH 5\n",
            "2019-12-31 06:55:30,772 Epoch   5 Step:     1100 Batch Loss:     3.604009 Tokens per Sec:    12736, Lr: 0.000300\n",
            "2019-12-31 06:55:45,407 Epoch   5 Step:     1200 Batch Loss:     3.513468 Tokens per Sec:    13516, Lr: 0.000300\n",
            "2019-12-31 06:56:00,296 Epoch   5 Step:     1300 Batch Loss:     3.601948 Tokens per Sec:    14033, Lr: 0.000300\n",
            "2019-12-31 06:56:07,024 Epoch   5: total training loss 969.24\n",
            "2019-12-31 06:56:07,025 EPOCH 6\n",
            "2019-12-31 06:56:15,062 Epoch   6 Step:     1400 Batch Loss:     3.351712 Tokens per Sec:    13493, Lr: 0.000300\n",
            "2019-12-31 06:56:29,594 Epoch   6 Step:     1500 Batch Loss:     3.376288 Tokens per Sec:    13770, Lr: 0.000300\n",
            "2019-12-31 06:56:44,362 Epoch   6 Step:     1600 Batch Loss:     3.233148 Tokens per Sec:    13988, Lr: 0.000300\n",
            "2019-12-31 06:56:46,545 Epoch   6: total training loss 921.14\n",
            "2019-12-31 06:56:46,545 EPOCH 7\n",
            "2019-12-31 06:56:59,105 Epoch   7 Step:     1700 Batch Loss:     3.105795 Tokens per Sec:    13838, Lr: 0.000300\n",
            "2019-12-31 06:57:13,752 Epoch   7 Step:     1800 Batch Loss:     3.416011 Tokens per Sec:    13842, Lr: 0.000300\n",
            "2019-12-31 06:57:26,103 Epoch   7: total training loss 884.22\n",
            "2019-12-31 06:57:26,103 EPOCH 8\n",
            "2019-12-31 06:57:28,276 Epoch   8 Step:     1900 Batch Loss:     3.263082 Tokens per Sec:    12921, Lr: 0.000300\n",
            "2019-12-31 06:57:42,904 Epoch   8 Step:     2000 Batch Loss:     3.043142 Tokens per Sec:    14111, Lr: 0.000300\n",
            "2019-12-31 06:58:16,547 Hooray! New best validation result [ppl]!\n",
            "2019-12-31 06:58:16,547 Saving new checkpoint.\n",
            "2019-12-31 06:58:16,873 Example #0\n",
            "2019-12-31 06:58:16,873 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 06:58:16,873 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 06:58:16,873 \tHypothesis: Ọ da dianẹ a sa mrẹ ukẹcha rẹ ihwo efa , ọ je sa nẹrhẹ a mrẹ ukẹcha rẹ avwanre .\n",
            "2019-12-31 06:58:16,873 Example #1\n",
            "2019-12-31 06:58:16,873 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 06:58:16,874 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 06:58:16,874 \tHypothesis: Ọ da dianẹ ọ dia ọkobaro vwẹ ukpe rẹ ẹkuotọ rẹ ẹkuotọ rẹ Izrẹl .\n",
            "2019-12-31 06:58:16,874 Example #2\n",
            "2019-12-31 06:58:16,874 \tSource:     But freedom from what ?\n",
            "2019-12-31 06:58:16,874 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 06:58:16,874 \tHypothesis: Ẹkẹvuọvo , die yen avwanre vwo ?\n",
            "2019-12-31 06:58:16,874 Example #3\n",
            "2019-12-31 06:58:16,874 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 06:58:16,874 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 06:58:16,874 \tHypothesis: Ọ da dianẹ wọ sa mrẹ ukẹcha kẹ ihwo efa .\n",
            "2019-12-31 06:58:16,874 Validation result (greedy) at epoch   8, step     2000: bleu:   2.83, loss: 67111.7422, ppl:  24.7566, duration: 33.9699s\n",
            "2019-12-31 06:58:31,599 Epoch   8 Step:     2100 Batch Loss:     2.704267 Tokens per Sec:    13466, Lr: 0.000300\n",
            "2019-12-31 06:58:39,953 Epoch   8: total training loss 853.30\n",
            "2019-12-31 06:58:39,953 EPOCH 9\n",
            "2019-12-31 06:58:46,327 Epoch   9 Step:     2200 Batch Loss:     3.021497 Tokens per Sec:    13264, Lr: 0.000300\n",
            "2019-12-31 06:59:01,190 Epoch   9 Step:     2300 Batch Loss:     3.177639 Tokens per Sec:    13941, Lr: 0.000300\n",
            "2019-12-31 06:59:15,820 Epoch   9 Step:     2400 Batch Loss:     2.865735 Tokens per Sec:    13868, Lr: 0.000300\n",
            "2019-12-31 06:59:19,448 Epoch   9: total training loss 805.14\n",
            "2019-12-31 06:59:19,448 EPOCH 10\n",
            "2019-12-31 06:59:30,557 Epoch  10 Step:     2500 Batch Loss:     2.494851 Tokens per Sec:    13864, Lr: 0.000300\n",
            "2019-12-31 06:59:45,149 Epoch  10 Step:     2600 Batch Loss:     3.033088 Tokens per Sec:    13772, Lr: 0.000300\n",
            "2019-12-31 06:59:59,161 Epoch  10: total training loss 779.46\n",
            "2019-12-31 06:59:59,161 EPOCH 11\n",
            "2019-12-31 06:59:59,935 Epoch  11 Step:     2700 Batch Loss:     3.020592 Tokens per Sec:    12892, Lr: 0.000300\n",
            "2019-12-31 07:00:14,664 Epoch  11 Step:     2800 Batch Loss:     2.511718 Tokens per Sec:    13639, Lr: 0.000300\n",
            "2019-12-31 07:00:29,442 Epoch  11 Step:     2900 Batch Loss:     2.980664 Tokens per Sec:    13900, Lr: 0.000300\n",
            "2019-12-31 07:00:38,842 Epoch  11: total training loss 750.48\n",
            "2019-12-31 07:00:38,843 EPOCH 12\n",
            "2019-12-31 07:00:44,065 Epoch  12 Step:     3000 Batch Loss:     2.592532 Tokens per Sec:    13316, Lr: 0.000300\n",
            "2019-12-31 07:01:17,758 Hooray! New best validation result [ppl]!\n",
            "2019-12-31 07:01:17,758 Saving new checkpoint.\n",
            "2019-12-31 07:01:18,076 Example #0\n",
            "2019-12-31 07:01:18,076 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:01:18,077 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:01:18,077 \tHypothesis: Ihwo buebun vwo imuẹro kpahen obo re sa nẹrhẹ ayen riẹn kpahen obo re sa vwọ chọn ayen uko .\n",
            "2019-12-31 07:01:18,077 Example #1\n",
            "2019-12-31 07:01:18,077 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:01:18,077 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:01:18,077 \tHypothesis: ( 1 Kọr . 3 : 1 - 14 ) Ẹkẹvuọvo , o de ji te omarẹ ẹgbukpe ujorin buebun .\n",
            "2019-12-31 07:01:18,077 Example #2\n",
            "2019-12-31 07:01:18,077 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:01:18,077 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:01:18,077 \tHypothesis: Ẹkẹvuọvo , die yen omaevwokpotọ ?\n",
            "2019-12-31 07:01:18,077 Example #3\n",
            "2019-12-31 07:01:18,077 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:01:18,077 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:01:18,077 \tHypothesis: Ọ da dianẹ wọ dia ọtiọyen , kẹ wẹ omavwerhovwẹn .\n",
            "2019-12-31 07:01:18,077 Validation result (greedy) at epoch  12, step     3000: bleu:   4.77, loss: 61367.6328, ppl:  18.8107, duration: 34.0124s\n",
            "2019-12-31 07:01:32,969 Epoch  12 Step:     3100 Batch Loss:     2.875560 Tokens per Sec:    13999, Lr: 0.000300\n",
            "2019-12-31 07:01:47,587 Epoch  12 Step:     3200 Batch Loss:     2.753931 Tokens per Sec:    13652, Lr: 0.000300\n",
            "2019-12-31 07:01:52,566 Epoch  12: total training loss 728.14\n",
            "2019-12-31 07:01:52,566 EPOCH 13\n",
            "2019-12-31 07:02:02,343 Epoch  13 Step:     3300 Batch Loss:     2.643054 Tokens per Sec:    13837, Lr: 0.000300\n",
            "2019-12-31 07:02:17,160 Epoch  13 Step:     3400 Batch Loss:     2.899310 Tokens per Sec:    13726, Lr: 0.000300\n",
            "2019-12-31 07:02:31,803 Epoch  13 Step:     3500 Batch Loss:     2.638120 Tokens per Sec:    13650, Lr: 0.000300\n",
            "2019-12-31 07:02:32,430 Epoch  13: total training loss 705.72\n",
            "2019-12-31 07:02:32,430 EPOCH 14\n",
            "2019-12-31 07:02:46,304 Epoch  14 Step:     3600 Batch Loss:     2.360273 Tokens per Sec:    13483, Lr: 0.000300\n",
            "2019-12-31 07:03:01,038 Epoch  14 Step:     3700 Batch Loss:     2.435451 Tokens per Sec:    13983, Lr: 0.000300\n",
            "2019-12-31 07:03:12,151 Epoch  14: total training loss 680.83\n",
            "2019-12-31 07:03:12,151 EPOCH 15\n",
            "2019-12-31 07:03:15,898 Epoch  15 Step:     3800 Batch Loss:     2.740217 Tokens per Sec:    13625, Lr: 0.000300\n",
            "2019-12-31 07:03:30,757 Epoch  15 Step:     3900 Batch Loss:     2.628906 Tokens per Sec:    13697, Lr: 0.000300\n",
            "2019-12-31 07:03:45,467 Epoch  15 Step:     4000 Batch Loss:     2.365105 Tokens per Sec:    13743, Lr: 0.000300\n",
            "2019-12-31 07:04:19,148 Hooray! New best validation result [ppl]!\n",
            "2019-12-31 07:04:19,148 Saving new checkpoint.\n",
            "2019-12-31 07:04:19,506 Example #0\n",
            "2019-12-31 07:04:19,506 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:04:19,506 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:04:19,506 \tHypothesis: Ihwo buebun vwo omavwerhovwẹn kpahen obo re sa nẹrhẹ ayen se vwo ẹwẹn obrorhiẹn rẹ avwanre , ji vwo ẹwẹn obrorhiẹn rẹ avwanre .\n",
            "2019-12-31 07:04:19,506 Example #1\n",
            "2019-12-31 07:04:19,506 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:04:19,506 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:04:19,506 \tHypothesis: Nonẹna , ọ dia ọkobaro vwẹ Bẹtẹl .\n",
            "2019-12-31 07:04:19,507 Example #2\n",
            "2019-12-31 07:04:19,507 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:04:19,507 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:04:19,507 \tHypothesis: Ẹkẹvuọvo , die yen egbomọphẹ rẹ avwanre ?\n",
            "2019-12-31 07:04:19,507 Example #3\n",
            "2019-12-31 07:04:19,507 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:04:19,507 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:04:19,507 \tHypothesis: Ọ dia ọtiọyen , ọ dia ọtiọyen wo vwo ukoko wẹn - a .\n",
            "2019-12-31 07:04:19,507 Validation result (greedy) at epoch  15, step     4000: bleu:   6.49, loss: 57640.0391, ppl:  15.7396, duration: 34.0399s\n",
            "2019-12-31 07:04:26,260 Epoch  15: total training loss 664.41\n",
            "2019-12-31 07:04:26,260 EPOCH 16\n",
            "2019-12-31 07:04:34,254 Epoch  16 Step:     4100 Batch Loss:     2.644593 Tokens per Sec:    13384, Lr: 0.000300\n",
            "2019-12-31 07:04:49,044 Epoch  16 Step:     4200 Batch Loss:     2.394016 Tokens per Sec:    13601, Lr: 0.000300\n",
            "2019-12-31 07:05:03,978 Epoch  16 Step:     4300 Batch Loss:     2.542410 Tokens per Sec:    13681, Lr: 0.000300\n",
            "2019-12-31 07:05:06,484 Epoch  16: total training loss 644.54\n",
            "2019-12-31 07:05:06,484 EPOCH 17\n",
            "2019-12-31 07:05:18,808 Epoch  17 Step:     4400 Batch Loss:     2.453203 Tokens per Sec:    13454, Lr: 0.000300\n",
            "2019-12-31 07:05:33,650 Epoch  17 Step:     4500 Batch Loss:     2.475983 Tokens per Sec:    13630, Lr: 0.000300\n",
            "2019-12-31 07:05:46,454 Epoch  17: total training loss 623.35\n",
            "2019-12-31 07:05:46,454 EPOCH 18\n",
            "2019-12-31 07:05:48,383 Epoch  18 Step:     4600 Batch Loss:     2.462085 Tokens per Sec:    12807, Lr: 0.000300\n",
            "2019-12-31 07:06:03,125 Epoch  18 Step:     4700 Batch Loss:     2.184745 Tokens per Sec:    13834, Lr: 0.000300\n",
            "2019-12-31 07:06:17,981 Epoch  18 Step:     4800 Batch Loss:     2.145816 Tokens per Sec:    13542, Lr: 0.000300\n",
            "2019-12-31 07:06:26,361 Epoch  18: total training loss 606.98\n",
            "2019-12-31 07:06:26,361 EPOCH 19\n",
            "2019-12-31 07:06:32,781 Epoch  19 Step:     4900 Batch Loss:     2.451422 Tokens per Sec:    13949, Lr: 0.000300\n",
            "2019-12-31 07:06:47,471 Epoch  19 Step:     5000 Batch Loss:     1.994083 Tokens per Sec:    13568, Lr: 0.000300\n",
            "2019-12-31 07:07:21,203 Hooray! New best validation result [ppl]!\n",
            "2019-12-31 07:07:21,203 Saving new checkpoint.\n",
            "2019-12-31 07:07:21,549 Example #0\n",
            "2019-12-31 07:07:21,549 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:07:21,549 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:07:21,549 \tHypothesis: Ọnana yen nẹrhẹ ayen se vwo ẹruọ rẹ obo ra guọnọre , rere ayen se vwo ruiruo .\n",
            "2019-12-31 07:07:21,549 Example #1\n",
            "2019-12-31 07:07:21,549 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:07:21,549 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:07:21,549 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:07:21,549 Example #2\n",
            "2019-12-31 07:07:21,549 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:07:21,549 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:07:21,549 \tHypothesis: Ẹkẹvuọvo , die yen egbomọphẹ ?\n",
            "2019-12-31 07:07:21,549 Example #3\n",
            "2019-12-31 07:07:21,550 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:07:21,550 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:07:21,550 \tHypothesis: Wọ guọnọ ukoko wẹn - a .\n",
            "2019-12-31 07:07:21,550 Validation result (greedy) at epoch  19, step     5000: bleu:   8.95, loss: 55712.8125, ppl:  14.3540, duration: 34.0789s\n",
            "2019-12-31 07:07:36,388 Epoch  19 Step:     5100 Batch Loss:     2.392033 Tokens per Sec:    13726, Lr: 0.000300\n",
            "2019-12-31 07:07:40,376 Epoch  19: total training loss 590.16\n",
            "2019-12-31 07:07:40,377 EPOCH 20\n",
            "2019-12-31 07:07:51,256 Epoch  20 Step:     5200 Batch Loss:     2.342585 Tokens per Sec:    13504, Lr: 0.000300\n",
            "2019-12-31 07:08:05,903 Epoch  20 Step:     5300 Batch Loss:     2.271941 Tokens per Sec:    13812, Lr: 0.000300\n",
            "2019-12-31 07:08:20,102 Epoch  20: total training loss 573.37\n",
            "2019-12-31 07:08:20,102 EPOCH 21\n",
            "2019-12-31 07:08:20,609 Epoch  21 Step:     5400 Batch Loss:     1.367846 Tokens per Sec:    12282, Lr: 0.000300\n",
            "2019-12-31 07:08:35,396 Epoch  21 Step:     5500 Batch Loss:     1.976137 Tokens per Sec:    13481, Lr: 0.000300\n",
            "2019-12-31 07:08:50,187 Epoch  21 Step:     5600 Batch Loss:     1.547307 Tokens per Sec:    13922, Lr: 0.000300\n",
            "2019-12-31 07:09:00,165 Epoch  21: total training loss 560.73\n",
            "2019-12-31 07:09:00,165 EPOCH 22\n",
            "2019-12-31 07:09:04,834 Epoch  22 Step:     5700 Batch Loss:     2.337951 Tokens per Sec:    13558, Lr: 0.000300\n",
            "2019-12-31 07:09:19,484 Epoch  22 Step:     5800 Batch Loss:     2.452579 Tokens per Sec:    13835, Lr: 0.000300\n",
            "2019-12-31 07:09:34,229 Epoch  22 Step:     5900 Batch Loss:     1.569645 Tokens per Sec:    13572, Lr: 0.000300\n",
            "2019-12-31 07:09:39,934 Epoch  22: total training loss 549.55\n",
            "2019-12-31 07:09:39,935 EPOCH 23\n",
            "2019-12-31 07:09:48,881 Epoch  23 Step:     6000 Batch Loss:     2.105916 Tokens per Sec:    13623, Lr: 0.000300\n",
            "2019-12-31 07:10:22,655 Hooray! New best validation result [ppl]!\n",
            "2019-12-31 07:10:22,655 Saving new checkpoint.\n",
            "2019-12-31 07:10:23,013 Example #0\n",
            "2019-12-31 07:10:23,013 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:10:23,013 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:10:23,013 \tHypothesis: Ọnana yen nẹrhẹ ayen vwo oniso rẹ obo re se vwo ru obo re se vwo ru obo re chọre .\n",
            "2019-12-31 07:10:23,013 Example #1\n",
            "2019-12-31 07:10:23,013 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:10:23,013 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:10:23,013 \tHypothesis: Nonẹna , o ji vwo ighwoghwota re ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:10:23,014 Example #2\n",
            "2019-12-31 07:10:23,014 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:10:23,014 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:10:23,014 \tHypothesis: Ẹkẹvuọvo , egbomọphẹ vọ yen e vwo ruiruo ?\n",
            "2019-12-31 07:10:23,014 Example #3\n",
            "2019-12-31 07:10:23,014 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:10:23,014 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:10:23,014 \tHypothesis: Wo jẹ ukoko na rhọnvwe nẹ ukoko wẹn rhe - e .\n",
            "2019-12-31 07:10:23,014 Validation result (greedy) at epoch  23, step     6000: bleu:  10.16, loss: 54999.3242, ppl:  13.8725, duration: 34.1329s\n",
            "2019-12-31 07:10:37,987 Epoch  23 Step:     6100 Batch Loss:     1.944094 Tokens per Sec:    13701, Lr: 0.000300\n",
            "2019-12-31 07:10:52,850 Epoch  23 Step:     6200 Batch Loss:     2.100627 Tokens per Sec:    13753, Lr: 0.000300\n",
            "2019-12-31 07:10:53,985 Epoch  23: total training loss 531.26\n",
            "2019-12-31 07:10:53,985 EPOCH 24\n",
            "2019-12-31 07:11:07,617 Epoch  24 Step:     6300 Batch Loss:     1.290862 Tokens per Sec:    13407, Lr: 0.000300\n",
            "2019-12-31 07:11:22,517 Epoch  24 Step:     6400 Batch Loss:     1.348583 Tokens per Sec:    13468, Lr: 0.000300\n",
            "2019-12-31 07:11:34,352 Epoch  24: total training loss 525.52\n",
            "2019-12-31 07:11:34,352 EPOCH 25\n",
            "2019-12-31 07:11:37,391 Epoch  25 Step:     6500 Batch Loss:     1.742582 Tokens per Sec:    12984, Lr: 0.000300\n",
            "2019-12-31 07:11:52,375 Epoch  25 Step:     6600 Batch Loss:     2.128956 Tokens per Sec:    13696, Lr: 0.000300\n",
            "2019-12-31 07:12:07,241 Epoch  25 Step:     6700 Batch Loss:     1.881225 Tokens per Sec:    13659, Lr: 0.000300\n",
            "2019-12-31 07:12:14,609 Epoch  25: total training loss 507.45\n",
            "2019-12-31 07:12:14,609 EPOCH 26\n",
            "2019-12-31 07:12:22,095 Epoch  26 Step:     6800 Batch Loss:     1.145603 Tokens per Sec:    13430, Lr: 0.000300\n",
            "2019-12-31 07:12:36,854 Epoch  26 Step:     6900 Batch Loss:     1.602573 Tokens per Sec:    13353, Lr: 0.000300\n",
            "2019-12-31 07:12:51,686 Epoch  26 Step:     7000 Batch Loss:     1.932218 Tokens per Sec:    13621, Lr: 0.000300\n",
            "2019-12-31 07:13:25,459 Hooray! New best validation result [ppl]!\n",
            "2019-12-31 07:13:25,460 Saving new checkpoint.\n",
            "2019-12-31 07:13:25,807 Example #0\n",
            "2019-12-31 07:13:25,807 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:13:25,807 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:13:25,807 \tHypothesis: Ọnana nẹrhẹ ayen se vwo ẹwẹn rẹ aghwanre , je davwẹngba vwo nene odjekẹ rẹ Baibol na .\n",
            "2019-12-31 07:13:25,808 Example #1\n",
            "2019-12-31 07:13:25,808 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:13:25,808 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:13:25,808 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:13:25,808 Example #2\n",
            "2019-12-31 07:13:25,808 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:13:25,808 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:13:25,808 \tHypothesis: Ẹkẹvuọvo , die yen egbomọphẹ ?\n",
            "2019-12-31 07:13:25,808 Example #3\n",
            "2019-12-31 07:13:25,808 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:13:25,808 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:13:25,808 \tHypothesis: Wọ rha guọnọ ukoko wẹn - o .\n",
            "2019-12-31 07:13:25,808 Validation result (greedy) at epoch  26, step     7000: bleu:  11.26, loss: 54282.3398, ppl:  13.4050, duration: 34.1218s\n",
            "2019-12-31 07:13:29,124 Epoch  26: total training loss 499.46\n",
            "2019-12-31 07:13:29,125 EPOCH 27\n",
            "2019-12-31 07:13:40,699 Epoch  27 Step:     7100 Batch Loss:     1.918137 Tokens per Sec:    13304, Lr: 0.000300\n",
            "2019-12-31 07:13:55,680 Epoch  27 Step:     7200 Batch Loss:     1.854988 Tokens per Sec:    13690, Lr: 0.000300\n",
            "2019-12-31 07:14:09,467 Epoch  27: total training loss 488.60\n",
            "2019-12-31 07:14:09,467 EPOCH 28\n",
            "2019-12-31 07:14:10,587 Epoch  28 Step:     7300 Batch Loss:     1.792533 Tokens per Sec:    14309, Lr: 0.000300\n",
            "2019-12-31 07:14:25,438 Epoch  28 Step:     7400 Batch Loss:     1.934778 Tokens per Sec:    13459, Lr: 0.000300\n",
            "2019-12-31 07:14:40,228 Epoch  28 Step:     7500 Batch Loss:     1.964069 Tokens per Sec:    13553, Lr: 0.000300\n",
            "2019-12-31 07:14:49,735 Epoch  28: total training loss 476.73\n",
            "2019-12-31 07:14:49,735 EPOCH 29\n",
            "2019-12-31 07:14:55,175 Epoch  29 Step:     7600 Batch Loss:     1.516087 Tokens per Sec:    13747, Lr: 0.000300\n",
            "2019-12-31 07:15:09,971 Epoch  29 Step:     7700 Batch Loss:     1.268625 Tokens per Sec:    13611, Lr: 0.000300\n",
            "2019-12-31 07:15:24,918 Epoch  29 Step:     7800 Batch Loss:     2.163529 Tokens per Sec:    13709, Lr: 0.000300\n",
            "2019-12-31 07:15:29,970 Epoch  29: total training loss 465.72\n",
            "2019-12-31 07:15:29,970 EPOCH 30\n",
            "2019-12-31 07:15:39,892 Epoch  30 Step:     7900 Batch Loss:     1.518623 Tokens per Sec:    13499, Lr: 0.000300\n",
            "2019-12-31 07:15:54,837 Epoch  30 Step:     8000 Batch Loss:     1.001237 Tokens per Sec:    13340, Lr: 0.000300\n",
            "2019-12-31 07:16:28,566 Example #0\n",
            "2019-12-31 07:16:28,566 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:16:28,566 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:16:28,566 \tHypothesis: Ọnana yen nẹrhẹ e se vwo ẹwẹn rẹ doe , je nabọ muegbe rẹ iroro rẹ avwanre vwo nene .\n",
            "2019-12-31 07:16:28,566 Example #1\n",
            "2019-12-31 07:16:28,566 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:16:28,566 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:16:28,566 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:16:28,567 Example #2\n",
            "2019-12-31 07:16:28,567 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:16:28,567 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:16:28,567 \tHypothesis: Ẹkẹvuọvo , egbomọphẹ vọ yen a vwọ kẹ egbomọphẹ ?\n",
            "2019-12-31 07:16:28,567 Example #3\n",
            "2019-12-31 07:16:28,567 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:16:28,567 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:16:28,567 \tHypothesis: Wọ riẹnre nẹ ukoko wẹn vẹ ukoko wẹn ọvo yen wọ hepha na - a .\n",
            "2019-12-31 07:16:28,567 Validation result (greedy) at epoch  30, step     8000: bleu:  12.18, loss: 54547.9141, ppl:  13.5763, duration: 33.7300s\n",
            "2019-12-31 07:16:43,485 Epoch  30 Step:     8100 Batch Loss:     1.709848 Tokens per Sec:    13856, Lr: 0.000300\n",
            "2019-12-31 07:16:43,933 Epoch  30: total training loss 454.59\n",
            "2019-12-31 07:16:43,933 EPOCH 31\n",
            "2019-12-31 07:16:58,426 Epoch  31 Step:     8200 Batch Loss:     1.812082 Tokens per Sec:    13612, Lr: 0.000300\n",
            "2019-12-31 07:17:13,237 Epoch  31 Step:     8300 Batch Loss:     1.758263 Tokens per Sec:    13320, Lr: 0.000300\n",
            "2019-12-31 07:17:24,266 Epoch  31: total training loss 450.16\n",
            "2019-12-31 07:17:24,267 EPOCH 32\n",
            "2019-12-31 07:17:28,183 Epoch  32 Step:     8400 Batch Loss:     1.760935 Tokens per Sec:    13715, Lr: 0.000300\n",
            "2019-12-31 07:17:42,936 Epoch  32 Step:     8500 Batch Loss:     1.677956 Tokens per Sec:    13580, Lr: 0.000300\n",
            "2019-12-31 07:17:57,806 Epoch  32 Step:     8600 Batch Loss:     1.764362 Tokens per Sec:    13357, Lr: 0.000300\n",
            "2019-12-31 07:18:04,582 Epoch  32: total training loss 441.58\n",
            "2019-12-31 07:18:04,583 EPOCH 33\n",
            "2019-12-31 07:18:12,782 Epoch  33 Step:     8700 Batch Loss:     1.516737 Tokens per Sec:    13426, Lr: 0.000300\n",
            "2019-12-31 07:18:27,660 Epoch  33 Step:     8800 Batch Loss:     1.292231 Tokens per Sec:    13503, Lr: 0.000300\n",
            "2019-12-31 07:18:42,588 Epoch  33 Step:     8900 Batch Loss:     1.753464 Tokens per Sec:    13668, Lr: 0.000300\n",
            "2019-12-31 07:18:44,874 Epoch  33: total training loss 431.91\n",
            "2019-12-31 07:18:44,874 EPOCH 34\n",
            "2019-12-31 07:18:57,558 Epoch  34 Step:     9000 Batch Loss:     1.862409 Tokens per Sec:    13725, Lr: 0.000300\n",
            "2019-12-31 07:19:31,389 Example #0\n",
            "2019-12-31 07:19:31,389 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:19:31,390 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:19:31,390 \tHypothesis: Ọnana nẹrhẹ ayen se vwo oniso rẹ oborẹ ubiudu avwanre se vwo ruiruo wan .\n",
            "2019-12-31 07:19:31,390 Example #1\n",
            "2019-12-31 07:19:31,390 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:19:31,390 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:19:31,390 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:19:31,390 Example #2\n",
            "2019-12-31 07:19:31,390 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:19:31,390 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:19:31,390 \tHypothesis: Ẹkẹvuọvo , egbomọphẹ vọ yen a vwọ kẹ ?\n",
            "2019-12-31 07:19:31,390 Example #3\n",
            "2019-12-31 07:19:31,390 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:19:31,390 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:19:31,390 \tHypothesis: Wọ vwẹroso ukoko wẹn jovwo .\n",
            "2019-12-31 07:19:31,390 Validation result (greedy) at epoch  34, step     9000: bleu:  12.61, loss: 54701.2695, ppl:  13.6762, duration: 33.8314s\n",
            "2019-12-31 07:19:46,345 Epoch  34 Step:     9100 Batch Loss:     1.435172 Tokens per Sec:    13780, Lr: 0.000300\n",
            "2019-12-31 07:19:58,647 Epoch  34: total training loss 417.15\n",
            "2019-12-31 07:19:58,647 EPOCH 35\n",
            "2019-12-31 07:20:01,300 Epoch  35 Step:     9200 Batch Loss:     1.500320 Tokens per Sec:    14582, Lr: 0.000300\n",
            "2019-12-31 07:20:16,040 Epoch  35 Step:     9300 Batch Loss:     1.891159 Tokens per Sec:    13347, Lr: 0.000300\n",
            "2019-12-31 07:20:30,961 Epoch  35 Step:     9400 Batch Loss:     1.892118 Tokens per Sec:    13556, Lr: 0.000300\n",
            "2019-12-31 07:20:38,832 Epoch  35: total training loss 416.28\n",
            "2019-12-31 07:20:38,832 EPOCH 36\n",
            "2019-12-31 07:20:45,805 Epoch  36 Step:     9500 Batch Loss:     1.718353 Tokens per Sec:    13421, Lr: 0.000300\n",
            "2019-12-31 07:21:00,800 Epoch  36 Step:     9600 Batch Loss:     1.592865 Tokens per Sec:    13581, Lr: 0.000300\n",
            "2019-12-31 07:21:15,653 Epoch  36 Step:     9700 Batch Loss:     1.648238 Tokens per Sec:    13649, Lr: 0.000300\n",
            "2019-12-31 07:21:19,154 Epoch  36: total training loss 408.22\n",
            "2019-12-31 07:21:19,154 EPOCH 37\n",
            "2019-12-31 07:21:30,550 Epoch  37 Step:     9800 Batch Loss:     1.706467 Tokens per Sec:    13861, Lr: 0.000300\n",
            "2019-12-31 07:21:45,351 Epoch  37 Step:     9900 Batch Loss:     1.681366 Tokens per Sec:    13437, Lr: 0.000300\n",
            "2019-12-31 07:21:59,221 Epoch  37: total training loss 401.36\n",
            "2019-12-31 07:21:59,221 EPOCH 38\n",
            "2019-12-31 07:22:00,182 Epoch  38 Step:    10000 Batch Loss:     1.660343 Tokens per Sec:    13737, Lr: 0.000300\n",
            "2019-12-31 07:22:33,954 Example #0\n",
            "2019-12-31 07:22:33,955 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:22:33,955 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:22:33,955 \tHypothesis: Enana nẹrhẹ ayen se vwo ẹwẹn rẹ avwanre vwo muegbe rẹ iroro rẹ avwanre , ji nene odjekẹ rẹ Baibol na .\n",
            "2019-12-31 07:22:33,955 Example #1\n",
            "2019-12-31 07:22:33,955 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:22:33,955 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:22:33,955 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:22:33,955 Example #2\n",
            "2019-12-31 07:22:33,955 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:22:33,955 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:22:33,955 \tHypothesis: Ẹkẹvuọvo , egbomọphẹ vọ yen a vwọ kẹ ?\n",
            "2019-12-31 07:22:33,955 Example #3\n",
            "2019-12-31 07:22:33,955 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:22:33,955 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:22:33,955 \tHypothesis: Wọ rha riẹn nẹ ukoko wẹn vẹ ukoko ọfa - a .\n",
            "2019-12-31 07:22:33,955 Validation result (greedy) at epoch  38, step    10000: bleu:  12.95, loss: 55179.7461, ppl:  13.9927, duration: 33.7736s\n",
            "2019-12-31 07:22:48,834 Epoch  38 Step:    10100 Batch Loss:     1.684994 Tokens per Sec:    13447, Lr: 0.000300\n",
            "2019-12-31 07:23:03,751 Epoch  38 Step:    10200 Batch Loss:     1.395089 Tokens per Sec:    13622, Lr: 0.000300\n",
            "2019-12-31 07:23:13,392 Epoch  38: total training loss 395.41\n",
            "2019-12-31 07:23:13,392 EPOCH 39\n",
            "2019-12-31 07:23:18,659 Epoch  39 Step:    10300 Batch Loss:     1.058967 Tokens per Sec:    13694, Lr: 0.000300\n",
            "2019-12-31 07:23:33,611 Epoch  39 Step:    10400 Batch Loss:     1.043420 Tokens per Sec:    13491, Lr: 0.000300\n",
            "2019-12-31 07:23:48,601 Epoch  39 Step:    10500 Batch Loss:     1.775222 Tokens per Sec:    13584, Lr: 0.000300\n",
            "2019-12-31 07:23:53,704 Epoch  39: total training loss 383.54\n",
            "2019-12-31 07:23:53,704 EPOCH 40\n",
            "2019-12-31 07:24:03,512 Epoch  40 Step:    10600 Batch Loss:     1.408048 Tokens per Sec:    13430, Lr: 0.000300\n",
            "2019-12-31 07:24:18,440 Epoch  40 Step:    10700 Batch Loss:     1.531781 Tokens per Sec:    13340, Lr: 0.000300\n",
            "2019-12-31 07:24:33,265 Epoch  40 Step:    10800 Batch Loss:     1.343017 Tokens per Sec:    13682, Lr: 0.000300\n",
            "2019-12-31 07:24:34,159 Epoch  40: total training loss 382.81\n",
            "2019-12-31 07:24:34,159 EPOCH 41\n",
            "2019-12-31 07:24:48,186 Epoch  41 Step:    10900 Batch Loss:     0.993885 Tokens per Sec:    13652, Lr: 0.000300\n",
            "2019-12-31 07:25:03,010 Epoch  41 Step:    11000 Batch Loss:     0.823253 Tokens per Sec:    13285, Lr: 0.000300\n",
            "2019-12-31 07:25:36,790 Example #0\n",
            "2019-12-31 07:25:36,790 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:25:36,791 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:25:36,791 \tHypothesis: Ọnana nẹrhẹ ayen se vwo oniso rẹ oborẹ ubiudu avwanre se vwo ruiruo .\n",
            "2019-12-31 07:25:36,791 Example #1\n",
            "2019-12-31 07:25:36,791 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:25:36,791 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:25:36,791 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:25:36,791 Example #2\n",
            "2019-12-31 07:25:36,791 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:25:36,791 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:25:36,791 \tHypothesis: Ẹkẹvuọvo , egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 07:25:36,791 Example #3\n",
            "2019-12-31 07:25:36,791 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:25:36,792 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:25:36,792 \tHypothesis: Wọ vwẹroso ukoko wẹn jovwo re .\n",
            "2019-12-31 07:25:36,792 Validation result (greedy) at epoch  41, step    11000: bleu:  13.17, loss: 55769.6367, ppl:  14.3930, duration: 33.7811s\n",
            "2019-12-31 07:25:48,155 Epoch  41: total training loss 373.73\n",
            "2019-12-31 07:25:48,155 EPOCH 42\n",
            "2019-12-31 07:25:51,671 Epoch  42 Step:    11100 Batch Loss:     1.485085 Tokens per Sec:    14135, Lr: 0.000300\n",
            "2019-12-31 07:26:06,501 Epoch  42 Step:    11200 Batch Loss:     1.645910 Tokens per Sec:    13656, Lr: 0.000300\n",
            "2019-12-31 07:26:21,306 Epoch  42 Step:    11300 Batch Loss:     1.245075 Tokens per Sec:    13573, Lr: 0.000300\n",
            "2019-12-31 07:26:28,026 Epoch  42: total training loss 363.26\n",
            "2019-12-31 07:26:28,026 EPOCH 43\n",
            "2019-12-31 07:26:36,205 Epoch  43 Step:    11400 Batch Loss:     1.492405 Tokens per Sec:    13992, Lr: 0.000300\n",
            "2019-12-31 07:26:51,050 Epoch  43 Step:    11500 Batch Loss:     1.488233 Tokens per Sec:    13812, Lr: 0.000300\n",
            "2019-12-31 07:27:05,757 Epoch  43 Step:    11600 Batch Loss:     1.194742 Tokens per Sec:    13588, Lr: 0.000300\n",
            "2019-12-31 07:27:07,815 Epoch  43: total training loss 358.63\n",
            "2019-12-31 07:27:07,815 EPOCH 44\n",
            "2019-12-31 07:27:20,483 Epoch  44 Step:    11700 Batch Loss:     0.656036 Tokens per Sec:    13722, Lr: 0.000300\n",
            "2019-12-31 07:27:35,368 Epoch  44 Step:    11800 Batch Loss:     1.543444 Tokens per Sec:    13849, Lr: 0.000300\n",
            "2019-12-31 07:27:47,541 Epoch  44: total training loss 354.16\n",
            "2019-12-31 07:27:47,541 EPOCH 45\n",
            "2019-12-31 07:27:50,119 Epoch  45 Step:    11900 Batch Loss:     1.403820 Tokens per Sec:    13083, Lr: 0.000300\n",
            "2019-12-31 07:28:04,991 Epoch  45 Step:    12000 Batch Loss:     1.411539 Tokens per Sec:    13758, Lr: 0.000300\n",
            "2019-12-31 07:28:38,668 Example #0\n",
            "2019-12-31 07:28:38,668 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:28:38,668 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:28:38,668 \tHypothesis: Ẹnwan nana nẹrhẹ ihwo buebun se muegbe rẹ ayen vwo muegbe rẹ iroro rẹ avwanre , je reyọ ayen vwo ruiruo .\n",
            "2019-12-31 07:28:38,668 Example #1\n",
            "2019-12-31 07:28:38,668 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:28:38,668 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:28:38,668 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:28:38,668 Example #2\n",
            "2019-12-31 07:28:38,669 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:28:38,669 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:28:38,669 \tHypothesis: Kẹ egbomọphẹ vwo ?\n",
            "2019-12-31 07:28:38,669 Example #3\n",
            "2019-12-31 07:28:38,669 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:28:38,669 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:28:38,669 \tHypothesis: Wọ riẹnre nẹ ukoko wẹn yovwin nọ wẹ - ẹ .\n",
            "2019-12-31 07:28:38,669 Validation result (greedy) at epoch  45, step    12000: bleu:  13.49, loss: 56564.6719, ppl:  14.9507, duration: 33.6780s\n",
            "2019-12-31 07:28:53,376 Epoch  45 Step:    12100 Batch Loss:     1.306053 Tokens per Sec:    13489, Lr: 0.000300\n",
            "2019-12-31 07:29:01,360 Epoch  45: total training loss 351.32\n",
            "2019-12-31 07:29:01,360 EPOCH 46\n",
            "2019-12-31 07:29:08,301 Epoch  46 Step:    12200 Batch Loss:     1.005195 Tokens per Sec:    14128, Lr: 0.000300\n",
            "2019-12-31 07:29:23,083 Epoch  46 Step:    12300 Batch Loss:     1.091875 Tokens per Sec:    13500, Lr: 0.000300\n",
            "2019-12-31 07:29:37,956 Epoch  46 Step:    12400 Batch Loss:     0.687656 Tokens per Sec:    13488, Lr: 0.000300\n",
            "2019-12-31 07:29:41,380 Epoch  46: total training loss 343.18\n",
            "2019-12-31 07:29:41,380 EPOCH 47\n",
            "2019-12-31 07:29:52,949 Epoch  47 Step:    12500 Batch Loss:     1.045718 Tokens per Sec:    13965, Lr: 0.000300\n",
            "2019-12-31 07:30:07,637 Epoch  47 Step:    12600 Batch Loss:     1.468696 Tokens per Sec:    13524, Lr: 0.000300\n",
            "2019-12-31 07:30:21,283 Epoch  47: total training loss 338.36\n",
            "2019-12-31 07:30:21,283 EPOCH 48\n",
            "2019-12-31 07:30:22,330 Epoch  48 Step:    12700 Batch Loss:     1.017751 Tokens per Sec:    12864, Lr: 0.000300\n",
            "2019-12-31 07:30:37,170 Epoch  48 Step:    12800 Batch Loss:     1.525858 Tokens per Sec:    13621, Lr: 0.000300\n",
            "2019-12-31 07:30:51,918 Epoch  48 Step:    12900 Batch Loss:     1.321356 Tokens per Sec:    13467, Lr: 0.000300\n",
            "2019-12-31 07:31:01,576 Epoch  48: total training loss 335.82\n",
            "2019-12-31 07:31:01,577 EPOCH 49\n",
            "2019-12-31 07:31:06,772 Epoch  49 Step:    13000 Batch Loss:     1.303249 Tokens per Sec:    13987, Lr: 0.000300\n",
            "2019-12-31 07:31:40,484 Example #0\n",
            "2019-12-31 07:31:40,484 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:31:40,484 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:31:40,484 \tHypothesis: A mrẹ ọnana vwẹ idjerhe tiọna , kidie ayen muegbe rẹ ayen vwo nene odjekẹ rẹ Jihova .\n",
            "2019-12-31 07:31:40,484 Example #1\n",
            "2019-12-31 07:31:40,484 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:31:40,484 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:31:40,484 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:31:40,485 Example #2\n",
            "2019-12-31 07:31:40,485 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:31:40,485 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:31:40,485 \tHypothesis: Kẹ egbomọphẹ vọ yen a vwọ kẹ ayen ?\n",
            "2019-12-31 07:31:40,485 Example #3\n",
            "2019-12-31 07:31:40,485 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:31:40,485 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:31:40,485 \tHypothesis: Wọ vwẹroso ukoko wẹn jovwo re .\n",
            "2019-12-31 07:31:40,485 Validation result (greedy) at epoch  49, step    13000: bleu:  13.26, loss: 56796.0586, ppl:  15.1171, duration: 33.7130s\n",
            "2019-12-31 07:31:55,255 Epoch  49 Step:    13100 Batch Loss:     1.389380 Tokens per Sec:    13469, Lr: 0.000210\n",
            "2019-12-31 07:32:10,009 Epoch  49 Step:    13200 Batch Loss:     1.064081 Tokens per Sec:    13524, Lr: 0.000210\n",
            "2019-12-31 07:32:15,387 Epoch  49: total training loss 323.91\n",
            "2019-12-31 07:32:15,387 EPOCH 50\n",
            "2019-12-31 07:32:24,759 Epoch  50 Step:    13300 Batch Loss:     0.791446 Tokens per Sec:    13819, Lr: 0.000210\n",
            "2019-12-31 07:32:39,696 Epoch  50 Step:    13400 Batch Loss:     1.359972 Tokens per Sec:    13756, Lr: 0.000210\n",
            "2019-12-31 07:32:54,375 Epoch  50 Step:    13500 Batch Loss:     1.257862 Tokens per Sec:    13295, Lr: 0.000210\n",
            "2019-12-31 07:32:55,532 Epoch  50: total training loss 316.78\n",
            "2019-12-31 07:32:55,532 EPOCH 51\n",
            "2019-12-31 07:33:09,273 Epoch  51 Step:    13600 Batch Loss:     1.332123 Tokens per Sec:    13406, Lr: 0.000210\n",
            "2019-12-31 07:33:23,968 Epoch  51 Step:    13700 Batch Loss:     1.320551 Tokens per Sec:    13654, Lr: 0.000210\n",
            "2019-12-31 07:33:35,646 Epoch  51: total training loss 310.94\n",
            "2019-12-31 07:33:35,646 EPOCH 52\n",
            "2019-12-31 07:33:38,791 Epoch  52 Step:    13800 Batch Loss:     1.326518 Tokens per Sec:    13637, Lr: 0.000210\n",
            "2019-12-31 07:33:53,612 Epoch  52 Step:    13900 Batch Loss:     0.786145 Tokens per Sec:    13392, Lr: 0.000210\n",
            "2019-12-31 07:34:08,496 Epoch  52 Step:    14000 Batch Loss:     1.086468 Tokens per Sec:    13691, Lr: 0.000210\n",
            "2019-12-31 07:34:42,326 Example #0\n",
            "2019-12-31 07:34:42,326 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:34:42,326 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:34:42,326 \tHypothesis: Enana nẹrhẹ ayen muegbe phiyotọ rere ayen se vwo muegbe rẹ iroro rẹ avwanre , ji nene odjekẹ rẹ Jihova .\n",
            "2019-12-31 07:34:42,326 Example #1\n",
            "2019-12-31 07:34:42,326 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:34:42,326 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:34:42,326 \tHypothesis: Enẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:34:42,326 Example #2\n",
            "2019-12-31 07:34:42,327 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:34:42,327 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:34:42,327 \tHypothesis: Ẹkẹvuọvo , egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 07:34:42,327 Example #3\n",
            "2019-12-31 07:34:42,327 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:34:42,327 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:34:42,327 \tHypothesis: Wọ vwẹroso ukoko wẹn jovwo re .\n",
            "2019-12-31 07:34:42,327 Validation result (greedy) at epoch  52, step    14000: bleu:  13.94, loss: 57613.8438, ppl:  15.7199, duration: 33.8313s\n",
            "2019-12-31 07:34:49,612 Epoch  52: total training loss 306.30\n",
            "2019-12-31 07:34:49,613 EPOCH 53\n",
            "2019-12-31 07:34:57,183 Epoch  53 Step:    14100 Batch Loss:     1.331935 Tokens per Sec:    13640, Lr: 0.000210\n",
            "2019-12-31 07:35:12,018 Epoch  53 Step:    14200 Batch Loss:     0.785916 Tokens per Sec:    13685, Lr: 0.000210\n",
            "2019-12-31 07:35:26,934 Epoch  53 Step:    14300 Batch Loss:     1.049746 Tokens per Sec:    13668, Lr: 0.000210\n",
            "2019-12-31 07:35:29,572 Epoch  53: total training loss 302.14\n",
            "2019-12-31 07:35:29,572 EPOCH 54\n",
            "2019-12-31 07:35:41,774 Epoch  54 Step:    14400 Batch Loss:     1.237039 Tokens per Sec:    13656, Lr: 0.000210\n",
            "2019-12-31 07:35:56,680 Epoch  54 Step:    14500 Batch Loss:     1.054131 Tokens per Sec:    13559, Lr: 0.000210\n",
            "2019-12-31 07:36:09,812 Epoch  54: total training loss 300.16\n",
            "2019-12-31 07:36:09,813 EPOCH 55\n",
            "2019-12-31 07:36:11,531 Epoch  55 Step:    14600 Batch Loss:     1.191510 Tokens per Sec:    13756, Lr: 0.000210\n",
            "2019-12-31 07:36:26,345 Epoch  55 Step:    14700 Batch Loss:     1.276201 Tokens per Sec:    13744, Lr: 0.000210\n",
            "2019-12-31 07:36:41,171 Epoch  55 Step:    14800 Batch Loss:     1.238366 Tokens per Sec:    13368, Lr: 0.000210\n",
            "2019-12-31 07:36:50,167 Epoch  55: total training loss 297.80\n",
            "2019-12-31 07:36:50,167 EPOCH 56\n",
            "2019-12-31 07:36:56,088 Epoch  56 Step:    14900 Batch Loss:     0.694806 Tokens per Sec:    13567, Lr: 0.000210\n",
            "2019-12-31 07:37:10,996 Epoch  56 Step:    15000 Batch Loss:     0.932301 Tokens per Sec:    13844, Lr: 0.000210\n",
            "2019-12-31 07:37:44,803 Example #0\n",
            "2019-12-31 07:37:44,804 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:37:44,804 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:37:44,804 \tHypothesis: Enana nẹrhẹ ihwo buebun se muegbe rẹ iroro rayen , ayen me je nabọ nene odjekẹ rẹ ẹwẹn na .\n",
            "2019-12-31 07:37:44,804 Example #1\n",
            "2019-12-31 07:37:44,804 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:37:44,804 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:37:44,804 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:37:44,804 Example #2\n",
            "2019-12-31 07:37:44,804 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:37:44,804 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:37:44,805 \tHypothesis: Ẹkẹvuọvo , die yen egbomọphẹ ?\n",
            "2019-12-31 07:37:44,805 Example #3\n",
            "2019-12-31 07:37:44,805 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:37:44,805 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:37:44,805 \tHypothesis: Wọ je vwẹroso ukoko wẹn jovwo re .\n",
            "2019-12-31 07:37:44,805 Validation result (greedy) at epoch  56, step    15000: bleu:  13.85, loss: 58165.7656, ppl:  16.1403, duration: 33.8089s\n",
            "2019-12-31 07:37:59,785 Epoch  56 Step:    15100 Batch Loss:     1.242236 Tokens per Sec:    13386, Lr: 0.000210\n",
            "2019-12-31 07:38:04,108 Epoch  56: total training loss 289.53\n",
            "2019-12-31 07:38:04,108 EPOCH 57\n",
            "2019-12-31 07:38:14,736 Epoch  57 Step:    15200 Batch Loss:     1.003102 Tokens per Sec:    13638, Lr: 0.000210\n",
            "2019-12-31 07:38:29,607 Epoch  57 Step:    15300 Batch Loss:     1.058721 Tokens per Sec:    13589, Lr: 0.000210\n",
            "2019-12-31 07:38:44,320 Epoch  57: total training loss 288.69\n",
            "2019-12-31 07:38:44,320 EPOCH 58\n",
            "2019-12-31 07:38:44,516 Epoch  58 Step:    15400 Batch Loss:     1.080109 Tokens per Sec:    12250, Lr: 0.000210\n",
            "2019-12-31 07:38:59,397 Epoch  58 Step:    15500 Batch Loss:     1.339564 Tokens per Sec:    13319, Lr: 0.000210\n",
            "2019-12-31 07:39:14,251 Epoch  58 Step:    15600 Batch Loss:     1.139009 Tokens per Sec:    13703, Lr: 0.000210\n",
            "2019-12-31 07:39:24,965 Epoch  58: total training loss 287.31\n",
            "2019-12-31 07:39:24,965 EPOCH 59\n",
            "2019-12-31 07:39:29,128 Epoch  59 Step:    15700 Batch Loss:     1.161043 Tokens per Sec:    12641, Lr: 0.000210\n",
            "2019-12-31 07:39:44,108 Epoch  59 Step:    15800 Batch Loss:     1.135937 Tokens per Sec:    13509, Lr: 0.000210\n",
            "2019-12-31 07:39:58,904 Epoch  59 Step:    15900 Batch Loss:     0.949926 Tokens per Sec:    13314, Lr: 0.000210\n",
            "2019-12-31 07:40:05,583 Epoch  59: total training loss 286.00\n",
            "2019-12-31 07:40:05,584 EPOCH 60\n",
            "2019-12-31 07:40:13,830 Epoch  60 Step:    16000 Batch Loss:     1.321432 Tokens per Sec:    13721, Lr: 0.000210\n",
            "2019-12-31 07:40:47,531 Example #0\n",
            "2019-12-31 07:40:47,531 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:40:47,531 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:40:47,531 \tHypothesis: Osichọ nana nẹrhẹ ihwo efa se muegbe rẹ ubiudu avwanre , ayen me je nabọ nene odjekẹ rẹ avwanre .\n",
            "2019-12-31 07:40:47,531 Example #1\n",
            "2019-12-31 07:40:47,531 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:40:47,531 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:40:47,532 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:40:47,532 Example #2\n",
            "2019-12-31 07:40:47,532 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:40:47,532 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:40:47,532 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 07:40:47,532 Example #3\n",
            "2019-12-31 07:40:47,532 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:40:47,532 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:40:47,532 \tHypothesis: Wọ vwẹroso ukoko wẹn jovwo re .\n",
            "2019-12-31 07:40:47,532 Validation result (greedy) at epoch  60, step    16000: bleu:  14.43, loss: 58579.1992, ppl:  16.4626, duration: 33.7021s\n",
            "2019-12-31 07:41:02,310 Epoch  60 Step:    16100 Batch Loss:     0.775197 Tokens per Sec:    13268, Lr: 0.000210\n",
            "2019-12-31 07:41:17,105 Epoch  60 Step:    16200 Batch Loss:     1.118505 Tokens per Sec:    13730, Lr: 0.000210\n",
            "2019-12-31 07:41:19,602 Epoch  60: total training loss 281.15\n",
            "2019-12-31 07:41:19,602 EPOCH 61\n",
            "2019-12-31 07:41:31,965 Epoch  61 Step:    16300 Batch Loss:     1.093107 Tokens per Sec:    13611, Lr: 0.000210\n",
            "2019-12-31 07:41:46,584 Epoch  61 Step:    16400 Batch Loss:     0.663107 Tokens per Sec:    13746, Lr: 0.000210\n",
            "2019-12-31 07:41:59,520 Epoch  61: total training loss 277.93\n",
            "2019-12-31 07:41:59,520 EPOCH 62\n",
            "2019-12-31 07:42:01,338 Epoch  62 Step:    16500 Batch Loss:     0.502820 Tokens per Sec:    13691, Lr: 0.000210\n",
            "2019-12-31 07:42:16,126 Epoch  62 Step:    16600 Batch Loss:     1.289075 Tokens per Sec:    13521, Lr: 0.000210\n",
            "2019-12-31 07:42:31,028 Epoch  62 Step:    16700 Batch Loss:     1.139260 Tokens per Sec:    13888, Lr: 0.000210\n",
            "2019-12-31 07:42:39,467 Epoch  62: total training loss 271.96\n",
            "2019-12-31 07:42:39,467 EPOCH 63\n",
            "2019-12-31 07:42:45,900 Epoch  63 Step:    16800 Batch Loss:     1.126073 Tokens per Sec:    13210, Lr: 0.000210\n",
            "2019-12-31 07:43:00,664 Epoch  63 Step:    16900 Batch Loss:     0.861230 Tokens per Sec:    13775, Lr: 0.000210\n",
            "2019-12-31 07:43:15,617 Epoch  63 Step:    17000 Batch Loss:     1.107252 Tokens per Sec:    13746, Lr: 0.000210\n",
            "2019-12-31 07:43:49,350 Example #0\n",
            "2019-12-31 07:43:49,351 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:43:49,351 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:43:49,351 \tHypothesis: Ebẹnbẹn nana nẹrhẹ ayen se muegbe rẹ ubiudu rayen eje , ji muegbe rẹ oborẹ avwanre seri .\n",
            "2019-12-31 07:43:49,351 Example #1\n",
            "2019-12-31 07:43:49,351 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:43:49,351 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:43:49,351 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:43:49,351 Example #2\n",
            "2019-12-31 07:43:49,351 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:43:49,351 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:43:49,351 \tHypothesis: Ẹkẹvuọvo , die yen egbomọphẹ ?\n",
            "2019-12-31 07:43:49,352 Example #3\n",
            "2019-12-31 07:43:49,352 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:43:49,352 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:43:49,352 \tHypothesis: Wọ je guọnọ yan nene ukoko wẹn ọfa jovwo re .\n",
            "2019-12-31 07:43:49,352 Validation result (greedy) at epoch  63, step    17000: bleu:  14.35, loss: 59104.8047, ppl:  16.8816, duration: 33.7348s\n",
            "2019-12-31 07:43:53,277 Epoch  63: total training loss 270.66\n",
            "2019-12-31 07:43:53,277 EPOCH 64\n",
            "2019-12-31 07:44:04,195 Epoch  64 Step:    17100 Batch Loss:     0.870494 Tokens per Sec:    13742, Lr: 0.000210\n",
            "2019-12-31 07:44:18,938 Epoch  64 Step:    17200 Batch Loss:     0.613872 Tokens per Sec:    13650, Lr: 0.000210\n",
            "2019-12-31 07:44:33,162 Epoch  64: total training loss 267.55\n",
            "2019-12-31 07:44:33,162 EPOCH 65\n",
            "2019-12-31 07:44:33,629 Epoch  65 Step:    17300 Batch Loss:     0.944022 Tokens per Sec:    11031, Lr: 0.000210\n",
            "2019-12-31 07:44:48,454 Epoch  65 Step:    17400 Batch Loss:     1.272358 Tokens per Sec:    13888, Lr: 0.000210\n",
            "2019-12-31 07:45:03,180 Epoch  65 Step:    17500 Batch Loss:     1.089397 Tokens per Sec:    13364, Lr: 0.000210\n",
            "2019-12-31 07:45:13,125 Epoch  65: total training loss 264.64\n",
            "2019-12-31 07:45:13,125 EPOCH 66\n",
            "2019-12-31 07:45:18,055 Epoch  66 Step:    17600 Batch Loss:     1.153565 Tokens per Sec:    13187, Lr: 0.000210\n",
            "2019-12-31 07:45:32,883 Epoch  66 Step:    17700 Batch Loss:     1.080082 Tokens per Sec:    13863, Lr: 0.000210\n",
            "2019-12-31 07:45:47,669 Epoch  66 Step:    17800 Batch Loss:     1.123865 Tokens per Sec:    13598, Lr: 0.000210\n",
            "2019-12-31 07:45:53,157 Epoch  66: total training loss 263.13\n",
            "2019-12-31 07:45:53,157 EPOCH 67\n",
            "2019-12-31 07:46:02,589 Epoch  67 Step:    17900 Batch Loss:     0.950385 Tokens per Sec:    13734, Lr: 0.000210\n",
            "2019-12-31 07:46:17,427 Epoch  67 Step:    18000 Batch Loss:     0.714562 Tokens per Sec:    13507, Lr: 0.000210\n",
            "2019-12-31 07:46:51,215 Example #0\n",
            "2019-12-31 07:46:51,216 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:46:51,216 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:46:51,216 \tHypothesis: Oyan nana nẹrhẹ ayen se muegbe rẹ iroro vẹ iruo rẹ avwanre , ayen me je nabọ nene odjekẹ na .\n",
            "2019-12-31 07:46:51,216 Example #1\n",
            "2019-12-31 07:46:51,216 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:46:51,216 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:46:51,216 \tHypothesis: Enẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:46:51,216 Example #2\n",
            "2019-12-31 07:46:51,216 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:46:51,216 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:46:51,216 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 07:46:51,216 Example #3\n",
            "2019-12-31 07:46:51,217 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:46:51,217 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:46:51,217 \tHypothesis: Wọ je vwẹroso ukoko wẹn jovwo re .\n",
            "2019-12-31 07:46:51,217 Validation result (greedy) at epoch  67, step    18000: bleu:  13.90, loss: 59971.5781, ppl:  17.5960, duration: 33.7899s\n",
            "2019-12-31 07:47:06,104 Epoch  67 Step:    18100 Batch Loss:     0.912966 Tokens per Sec:    13543, Lr: 0.000210\n",
            "2019-12-31 07:47:07,160 Epoch  67: total training loss 260.01\n",
            "2019-12-31 07:47:07,160 EPOCH 68\n",
            "2019-12-31 07:47:21,057 Epoch  68 Step:    18200 Batch Loss:     1.043700 Tokens per Sec:    13453, Lr: 0.000210\n",
            "2019-12-31 07:47:36,048 Epoch  68 Step:    18300 Batch Loss:     0.670840 Tokens per Sec:    13918, Lr: 0.000210\n",
            "2019-12-31 07:47:47,234 Epoch  68: total training loss 255.27\n",
            "2019-12-31 07:47:47,234 EPOCH 69\n",
            "2019-12-31 07:47:51,020 Epoch  69 Step:    18400 Batch Loss:     1.011232 Tokens per Sec:    13661, Lr: 0.000210\n",
            "2019-12-31 07:48:05,847 Epoch  69 Step:    18500 Batch Loss:     1.036424 Tokens per Sec:    13554, Lr: 0.000210\n",
            "2019-12-31 07:48:20,864 Epoch  69 Step:    18600 Batch Loss:     1.124365 Tokens per Sec:    13382, Lr: 0.000210\n",
            "2019-12-31 07:48:27,453 Epoch  69: total training loss 253.82\n",
            "2019-12-31 07:48:27,453 EPOCH 70\n",
            "2019-12-31 07:48:35,817 Epoch  70 Step:    18700 Batch Loss:     0.772242 Tokens per Sec:    13317, Lr: 0.000210\n",
            "2019-12-31 07:48:50,724 Epoch  70 Step:    18800 Batch Loss:     1.204297 Tokens per Sec:    13848, Lr: 0.000210\n",
            "2019-12-31 07:49:05,593 Epoch  70 Step:    18900 Batch Loss:     1.094892 Tokens per Sec:    13427, Lr: 0.000210\n",
            "2019-12-31 07:49:07,563 Epoch  70: total training loss 251.40\n",
            "2019-12-31 07:49:07,564 EPOCH 71\n",
            "2019-12-31 07:49:20,587 Epoch  71 Step:    19000 Batch Loss:     1.145924 Tokens per Sec:    13584, Lr: 0.000210\n",
            "2019-12-31 07:49:54,405 Example #0\n",
            "2019-12-31 07:49:54,405 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:49:54,405 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:49:54,405 \tHypothesis: Enana nẹrhẹ ihwo efa muegbe rẹ ayen vwo muegbe rẹ obo rehẹ ubiudu avwanre , je nene odjekẹ na .\n",
            "2019-12-31 07:49:54,405 Example #1\n",
            "2019-12-31 07:49:54,405 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:49:54,405 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:49:54,405 \tHypothesis: Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:49:54,405 Example #2\n",
            "2019-12-31 07:49:54,406 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:49:54,406 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:49:54,406 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 07:49:54,406 Example #3\n",
            "2019-12-31 07:49:54,406 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:49:54,406 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:49:54,406 \tHypothesis: Wọ je guọnọ ukoko wẹn ọfa .\n",
            "2019-12-31 07:49:54,406 Validation result (greedy) at epoch  71, step    19000: bleu:  14.10, loss: 60328.6914, ppl:  17.8990, duration: 33.8183s\n",
            "2019-12-31 07:50:09,371 Epoch  71 Step:    19100 Batch Loss:     1.050977 Tokens per Sec:    13614, Lr: 0.000147\n",
            "2019-12-31 07:50:21,724 Epoch  71: total training loss 246.91\n",
            "2019-12-31 07:50:21,724 EPOCH 72\n",
            "2019-12-31 07:50:24,262 Epoch  72 Step:    19200 Batch Loss:     0.450549 Tokens per Sec:    12690, Lr: 0.000147\n",
            "2019-12-31 07:50:39,234 Epoch  72 Step:    19300 Batch Loss:     0.853629 Tokens per Sec:    13687, Lr: 0.000147\n",
            "2019-12-31 07:50:54,168 Epoch  72 Step:    19400 Batch Loss:     0.891482 Tokens per Sec:    13935, Lr: 0.000147\n",
            "2019-12-31 07:51:01,721 Epoch  72: total training loss 239.45\n",
            "2019-12-31 07:51:01,721 EPOCH 73\n",
            "2019-12-31 07:51:08,985 Epoch  73 Step:    19500 Batch Loss:     1.079561 Tokens per Sec:    13156, Lr: 0.000147\n",
            "2019-12-31 07:51:23,955 Epoch  73 Step:    19600 Batch Loss:     1.002316 Tokens per Sec:    13846, Lr: 0.000147\n",
            "2019-12-31 07:51:38,831 Epoch  73 Step:    19700 Batch Loss:     0.988723 Tokens per Sec:    13756, Lr: 0.000147\n",
            "2019-12-31 07:51:41,769 Epoch  73: total training loss 239.19\n",
            "2019-12-31 07:51:41,769 EPOCH 74\n",
            "2019-12-31 07:51:53,585 Epoch  74 Step:    19800 Batch Loss:     0.640235 Tokens per Sec:    13263, Lr: 0.000147\n",
            "2019-12-31 07:52:08,590 Epoch  74 Step:    19900 Batch Loss:     1.022838 Tokens per Sec:    13644, Lr: 0.000147\n",
            "2019-12-31 07:52:22,155 Epoch  74: total training loss 238.37\n",
            "2019-12-31 07:52:22,156 EPOCH 75\n",
            "2019-12-31 07:52:23,374 Epoch  75 Step:    20000 Batch Loss:     0.906945 Tokens per Sec:    12966, Lr: 0.000147\n",
            "2019-12-31 07:52:57,184 Example #0\n",
            "2019-12-31 07:52:57,184 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:52:57,184 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:52:57,184 \tHypothesis: Enana nẹrhẹ ihwo efa se muegbe rẹ ayen vwo muegbe rẹ obo rehẹ ubiudu avwanre , ji nene odjekẹ na .\n",
            "2019-12-31 07:52:57,185 Example #1\n",
            "2019-12-31 07:52:57,185 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:52:57,185 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:52:57,185 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:52:57,185 Example #2\n",
            "2019-12-31 07:52:57,185 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:52:57,185 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:52:57,185 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 07:52:57,185 Example #3\n",
            "2019-12-31 07:52:57,185 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:52:57,185 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:52:57,185 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn vẹ ukoko wẹn ro jovwo na .\n",
            "2019-12-31 07:52:57,186 Validation result (greedy) at epoch  75, step    20000: bleu:  14.55, loss: 60525.4727, ppl:  18.0682, duration: 33.8115s\n",
            "2019-12-31 07:53:12,078 Epoch  75 Step:    20100 Batch Loss:     1.026942 Tokens per Sec:    13675, Lr: 0.000147\n",
            "2019-12-31 07:53:26,775 Epoch  75 Step:    20200 Batch Loss:     1.028158 Tokens per Sec:    13415, Lr: 0.000147\n",
            "2019-12-31 07:53:36,291 Epoch  75: total training loss 235.79\n",
            "2019-12-31 07:53:36,291 EPOCH 76\n",
            "2019-12-31 07:53:41,639 Epoch  76 Step:    20300 Batch Loss:     0.872331 Tokens per Sec:    13056, Lr: 0.000147\n",
            "2019-12-31 07:53:56,459 Epoch  76 Step:    20400 Batch Loss:     0.974461 Tokens per Sec:    13431, Lr: 0.000147\n",
            "2019-12-31 07:54:11,488 Epoch  76 Step:    20500 Batch Loss:     0.923049 Tokens per Sec:    13733, Lr: 0.000147\n",
            "2019-12-31 07:54:16,788 Epoch  76: total training loss 234.84\n",
            "2019-12-31 07:54:16,788 EPOCH 77\n",
            "2019-12-31 07:54:26,426 Epoch  77 Step:    20600 Batch Loss:     0.581441 Tokens per Sec:    13485, Lr: 0.000147\n",
            "2019-12-31 07:54:41,404 Epoch  77 Step:    20700 Batch Loss:     0.839941 Tokens per Sec:    13588, Lr: 0.000147\n",
            "2019-12-31 07:54:56,243 Epoch  77 Step:    20800 Batch Loss:     0.593163 Tokens per Sec:    13439, Lr: 0.000147\n",
            "2019-12-31 07:54:57,168 Epoch  77: total training loss 230.64\n",
            "2019-12-31 07:54:57,168 EPOCH 78\n",
            "2019-12-31 07:55:11,224 Epoch  78 Step:    20900 Batch Loss:     0.635312 Tokens per Sec:    13392, Lr: 0.000147\n",
            "2019-12-31 07:55:26,134 Epoch  78 Step:    21000 Batch Loss:     0.955511 Tokens per Sec:    13504, Lr: 0.000147\n",
            "2019-12-31 07:55:59,934 Example #0\n",
            "2019-12-31 07:55:59,935 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:55:59,935 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:55:59,935 \tHypothesis: Ihoho nana nẹrhẹ ayen se muegbe rẹ ubiudu rayen , ayen me je nabọ muegbe rẹ oborẹ avwanre yonori .\n",
            "2019-12-31 07:55:59,935 Example #1\n",
            "2019-12-31 07:55:59,935 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:55:59,935 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:55:59,935 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:55:59,935 Example #2\n",
            "2019-12-31 07:55:59,935 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:55:59,935 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:55:59,936 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 07:55:59,936 Example #3\n",
            "2019-12-31 07:55:59,936 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:55:59,936 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:55:59,936 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn yovwin nọ wẹ nure .\n",
            "2019-12-31 07:55:59,936 Validation result (greedy) at epoch  78, step    21000: bleu:  14.48, loss: 61023.7109, ppl:  18.5039, duration: 33.8013s\n",
            "2019-12-31 07:56:11,388 Epoch  78: total training loss 230.17\n",
            "2019-12-31 07:56:11,388 EPOCH 79\n",
            "2019-12-31 07:56:14,832 Epoch  79 Step:    21100 Batch Loss:     0.634400 Tokens per Sec:    13166, Lr: 0.000147\n",
            "2019-12-31 07:56:29,734 Epoch  79 Step:    21200 Batch Loss:     0.793637 Tokens per Sec:    13534, Lr: 0.000147\n",
            "2019-12-31 07:56:44,726 Epoch  79 Step:    21300 Batch Loss:     0.754804 Tokens per Sec:    13533, Lr: 0.000147\n",
            "2019-12-31 07:56:51,738 Epoch  79: total training loss 227.47\n",
            "2019-12-31 07:56:51,738 EPOCH 80\n",
            "2019-12-31 07:56:59,622 Epoch  80 Step:    21400 Batch Loss:     0.845221 Tokens per Sec:    13202, Lr: 0.000147\n",
            "2019-12-31 07:57:14,470 Epoch  80 Step:    21500 Batch Loss:     0.930473 Tokens per Sec:    13504, Lr: 0.000147\n",
            "2019-12-31 07:57:29,341 Epoch  80 Step:    21600 Batch Loss:     0.399067 Tokens per Sec:    13554, Lr: 0.000147\n",
            "2019-12-31 07:57:32,338 Epoch  80: total training loss 227.87\n",
            "2019-12-31 07:57:32,339 EPOCH 81\n",
            "2019-12-31 07:57:44,336 Epoch  81 Step:    21700 Batch Loss:     1.001035 Tokens per Sec:    13749, Lr: 0.000147\n",
            "2019-12-31 07:57:59,179 Epoch  81 Step:    21800 Batch Loss:     0.795596 Tokens per Sec:    13464, Lr: 0.000147\n",
            "2019-12-31 07:58:12,574 Epoch  81: total training loss 224.77\n",
            "2019-12-31 07:58:12,574 EPOCH 82\n",
            "2019-12-31 07:58:14,125 Epoch  82 Step:    21900 Batch Loss:     0.872754 Tokens per Sec:    13425, Lr: 0.000147\n",
            "2019-12-31 07:58:29,065 Epoch  82 Step:    22000 Batch Loss:     0.965728 Tokens per Sec:    13633, Lr: 0.000147\n",
            "2019-12-31 07:59:02,907 Example #0\n",
            "2019-12-31 07:59:02,907 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 07:59:02,907 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 07:59:02,907 \tHypothesis: Ihoho nana nẹrhẹ ayen se muegbe rẹ ubiudu avwanre , ji nene odjekẹ rẹ avwanre .\n",
            "2019-12-31 07:59:02,907 Example #1\n",
            "2019-12-31 07:59:02,907 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 07:59:02,907 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:59:02,907 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 07:59:02,908 Example #2\n",
            "2019-12-31 07:59:02,908 \tSource:     But freedom from what ?\n",
            "2019-12-31 07:59:02,908 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 07:59:02,908 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 07:59:02,908 Example #3\n",
            "2019-12-31 07:59:02,908 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 07:59:02,908 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 07:59:02,908 \tHypothesis: Wọ je vwẹroso ukoko wẹn jovwo re .\n",
            "2019-12-31 07:59:02,908 Validation result (greedy) at epoch  82, step    22000: bleu:  14.41, loss: 61665.9570, ppl:  19.0809, duration: 33.8427s\n",
            "2019-12-31 07:59:17,755 Epoch  82 Step:    22100 Batch Loss:     1.046957 Tokens per Sec:    13367, Lr: 0.000147\n",
            "2019-12-31 07:59:26,828 Epoch  82: total training loss 223.08\n",
            "2019-12-31 07:59:26,828 EPOCH 83\n",
            "2019-12-31 07:59:32,698 Epoch  83 Step:    22200 Batch Loss:     0.395410 Tokens per Sec:    13809, Lr: 0.000147\n",
            "2019-12-31 07:59:47,606 Epoch  83 Step:    22300 Batch Loss:     0.757746 Tokens per Sec:    13413, Lr: 0.000147\n",
            "2019-12-31 08:00:02,660 Epoch  83 Step:    22400 Batch Loss:     0.406432 Tokens per Sec:    13740, Lr: 0.000147\n",
            "2019-12-31 08:00:07,195 Epoch  83: total training loss 222.05\n",
            "2019-12-31 08:00:07,195 EPOCH 84\n",
            "2019-12-31 08:00:17,540 Epoch  84 Step:    22500 Batch Loss:     0.903960 Tokens per Sec:    13877, Lr: 0.000147\n",
            "2019-12-31 08:00:32,503 Epoch  84 Step:    22600 Batch Loss:     0.983404 Tokens per Sec:    13303, Lr: 0.000147\n",
            "2019-12-31 08:00:47,450 Epoch  84 Step:    22700 Batch Loss:     0.904634 Tokens per Sec:    13579, Lr: 0.000147\n",
            "2019-12-31 08:00:47,451 Epoch  84: total training loss 218.51\n",
            "2019-12-31 08:00:47,451 EPOCH 85\n",
            "2019-12-31 08:01:02,503 Epoch  85 Step:    22800 Batch Loss:     0.816896 Tokens per Sec:    13549, Lr: 0.000147\n",
            "2019-12-31 08:01:17,350 Epoch  85 Step:    22900 Batch Loss:     0.859011 Tokens per Sec:    13531, Lr: 0.000147\n",
            "2019-12-31 08:01:27,849 Epoch  85: total training loss 218.53\n",
            "2019-12-31 08:01:27,849 EPOCH 86\n",
            "2019-12-31 08:01:32,229 Epoch  86 Step:    23000 Batch Loss:     0.690743 Tokens per Sec:    13111, Lr: 0.000147\n",
            "2019-12-31 08:02:06,068 Example #0\n",
            "2019-12-31 08:02:06,068 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:02:06,068 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:02:06,068 \tHypothesis: Enana nẹrhẹ ihwo ni opharo rẹ avwanre , ayen muegbe rẹ ayen vwo nene odjekẹ rẹ avwanre .\n",
            "2019-12-31 08:02:06,068 Example #1\n",
            "2019-12-31 08:02:06,068 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:02:06,068 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:02:06,068 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:02:06,068 Example #2\n",
            "2019-12-31 08:02:06,068 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:02:06,069 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:02:06,069 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 08:02:06,069 Example #3\n",
            "2019-12-31 08:02:06,069 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:02:06,069 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:02:06,069 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn yovwin nọ wẹ nure .\n",
            "2019-12-31 08:02:06,069 Validation result (greedy) at epoch  86, step    23000: bleu:  14.13, loss: 61697.9805, ppl:  19.1102, duration: 33.8394s\n",
            "2019-12-31 08:02:20,911 Epoch  86 Step:    23100 Batch Loss:     0.723428 Tokens per Sec:    13765, Lr: 0.000147\n",
            "2019-12-31 08:02:35,758 Epoch  86 Step:    23200 Batch Loss:     0.777151 Tokens per Sec:    13440, Lr: 0.000147\n",
            "2019-12-31 08:02:41,741 Epoch  86: total training loss 216.06\n",
            "2019-12-31 08:02:41,741 EPOCH 87\n",
            "2019-12-31 08:02:50,768 Epoch  87 Step:    23300 Batch Loss:     0.904117 Tokens per Sec:    13317, Lr: 0.000147\n",
            "2019-12-31 08:03:05,718 Epoch  87 Step:    23400 Batch Loss:     0.684742 Tokens per Sec:    13432, Lr: 0.000147\n",
            "2019-12-31 08:03:20,612 Epoch  87 Step:    23500 Batch Loss:     0.683818 Tokens per Sec:    13709, Lr: 0.000147\n",
            "2019-12-31 08:03:22,095 Epoch  87: total training loss 215.53\n",
            "2019-12-31 08:03:22,095 EPOCH 88\n",
            "2019-12-31 08:03:35,457 Epoch  88 Step:    23600 Batch Loss:     0.964557 Tokens per Sec:    13396, Lr: 0.000147\n",
            "2019-12-31 08:03:50,390 Epoch  88 Step:    23700 Batch Loss:     0.859948 Tokens per Sec:    13891, Lr: 0.000147\n",
            "2019-12-31 08:04:02,077 Epoch  88: total training loss 212.63\n",
            "2019-12-31 08:04:02,077 EPOCH 89\n",
            "2019-12-31 08:04:05,457 Epoch  89 Step:    23800 Batch Loss:     0.803128 Tokens per Sec:    13798, Lr: 0.000147\n",
            "2019-12-31 08:04:20,364 Epoch  89 Step:    23900 Batch Loss:     0.681111 Tokens per Sec:    13352, Lr: 0.000147\n",
            "2019-12-31 08:04:35,226 Epoch  89 Step:    24000 Batch Loss:     1.011402 Tokens per Sec:    13921, Lr: 0.000147\n",
            "2019-12-31 08:05:09,041 Example #0\n",
            "2019-12-31 08:05:09,041 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:05:09,042 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:05:09,042 \tHypothesis: Omaẹkparọ nẹrhẹ ihwo efa muegbe rẹ ayen vwo muegbe rẹ obo rehẹ ubiudu avwanre , ji nene odjekẹ na .\n",
            "2019-12-31 08:05:09,042 Example #1\n",
            "2019-12-31 08:05:09,042 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:05:09,042 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:05:09,042 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:05:09,042 Example #2\n",
            "2019-12-31 08:05:09,042 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:05:09,042 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:05:09,042 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 08:05:09,042 Example #3\n",
            "2019-12-31 08:05:09,043 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:05:09,043 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:05:09,043 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn yovwin nọ wẹ nure .\n",
            "2019-12-31 08:05:09,043 Validation result (greedy) at epoch  89, step    24000: bleu:  14.51, loss: 62261.5195, ppl:  19.6321, duration: 33.8165s\n",
            "2019-12-31 08:05:16,275 Epoch  89: total training loss 212.92\n",
            "2019-12-31 08:05:16,275 EPOCH 90\n",
            "2019-12-31 08:05:23,913 Epoch  90 Step:    24100 Batch Loss:     0.698223 Tokens per Sec:    13657, Lr: 0.000147\n",
            "2019-12-31 08:05:38,865 Epoch  90 Step:    24200 Batch Loss:     0.823427 Tokens per Sec:    13556, Lr: 0.000147\n",
            "2019-12-31 08:05:53,847 Epoch  90 Step:    24300 Batch Loss:     0.494558 Tokens per Sec:    13554, Lr: 0.000147\n",
            "2019-12-31 08:05:56,501 Epoch  90: total training loss 210.07\n",
            "2019-12-31 08:05:56,501 EPOCH 91\n",
            "2019-12-31 08:06:08,772 Epoch  91 Step:    24400 Batch Loss:     0.809604 Tokens per Sec:    13435, Lr: 0.000147\n",
            "2019-12-31 08:06:23,745 Epoch  91 Step:    24500 Batch Loss:     0.923940 Tokens per Sec:    13606, Lr: 0.000147\n",
            "2019-12-31 08:06:36,915 Epoch  91: total training loss 210.43\n",
            "2019-12-31 08:06:36,915 EPOCH 92\n",
            "2019-12-31 08:06:38,647 Epoch  92 Step:    24600 Batch Loss:     0.761108 Tokens per Sec:    14282, Lr: 0.000147\n",
            "2019-12-31 08:06:53,544 Epoch  92 Step:    24700 Batch Loss:     0.954293 Tokens per Sec:    13616, Lr: 0.000147\n",
            "2019-12-31 08:07:08,481 Epoch  92 Step:    24800 Batch Loss:     0.939290 Tokens per Sec:    13480, Lr: 0.000147\n",
            "2019-12-31 08:07:17,137 Epoch  92: total training loss 207.64\n",
            "2019-12-31 08:07:17,137 EPOCH 93\n",
            "2019-12-31 08:07:23,413 Epoch  93 Step:    24900 Batch Loss:     0.868608 Tokens per Sec:    13154, Lr: 0.000147\n",
            "2019-12-31 08:07:38,392 Epoch  93 Step:    25000 Batch Loss:     0.356983 Tokens per Sec:    13460, Lr: 0.000147\n",
            "2019-12-31 08:08:12,162 Example #0\n",
            "2019-12-31 08:08:12,162 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:08:12,162 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:08:12,162 \tHypothesis: Ihoho nana nẹrhẹ ayen se muegbe rẹ ubiudu rayen phiyọ , ayen me je nabọ muegbe rẹ oborẹ avwanre yonori .\n",
            "2019-12-31 08:08:12,162 Example #1\n",
            "2019-12-31 08:08:12,163 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:08:12,163 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:08:12,163 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:08:12,163 Example #2\n",
            "2019-12-31 08:08:12,163 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:08:12,163 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:08:12,163 \tHypothesis: Die yen egbomọphẹ ?\n",
            "2019-12-31 08:08:12,163 Example #3\n",
            "2019-12-31 08:08:12,163 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:08:12,163 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:08:12,163 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn yovwin nọ wẹ nure .\n",
            "2019-12-31 08:08:12,163 Validation result (greedy) at epoch  93, step    25000: bleu:  14.49, loss: 62652.5859, ppl:  20.0027, duration: 33.7710s\n",
            "2019-12-31 08:08:27,031 Epoch  93 Step:    25100 Batch Loss:     0.833798 Tokens per Sec:    13794, Lr: 0.000103\n",
            "2019-12-31 08:08:31,179 Epoch  93: total training loss 206.05\n",
            "2019-12-31 08:08:31,180 EPOCH 94\n",
            "2019-12-31 08:08:41,985 Epoch  94 Step:    25200 Batch Loss:     0.896664 Tokens per Sec:    13432, Lr: 0.000103\n",
            "2019-12-31 08:08:56,905 Epoch  94 Step:    25300 Batch Loss:     0.770320 Tokens per Sec:    13398, Lr: 0.000103\n",
            "2019-12-31 08:09:11,453 Epoch  94: total training loss 201.40\n",
            "2019-12-31 08:09:11,453 EPOCH 95\n",
            "2019-12-31 08:09:11,794 Epoch  95 Step:    25400 Batch Loss:     0.753056 Tokens per Sec:    11766, Lr: 0.000103\n",
            "2019-12-31 08:09:26,696 Epoch  95 Step:    25500 Batch Loss:     0.646330 Tokens per Sec:    13487, Lr: 0.000103\n",
            "2019-12-31 08:09:41,555 Epoch  95 Step:    25600 Batch Loss:     0.880025 Tokens per Sec:    13754, Lr: 0.000103\n",
            "2019-12-31 08:09:51,630 Epoch  95: total training loss 200.13\n",
            "2019-12-31 08:09:51,631 EPOCH 96\n",
            "2019-12-31 08:09:56,507 Epoch  96 Step:    25700 Batch Loss:     0.580037 Tokens per Sec:    13308, Lr: 0.000103\n",
            "2019-12-31 08:10:11,289 Epoch  96 Step:    25800 Batch Loss:     0.790462 Tokens per Sec:    13633, Lr: 0.000103\n",
            "2019-12-31 08:10:26,149 Epoch  96 Step:    25900 Batch Loss:     0.689909 Tokens per Sec:    13501, Lr: 0.000103\n",
            "2019-12-31 08:10:31,984 Epoch  96: total training loss 199.20\n",
            "2019-12-31 08:10:31,984 EPOCH 97\n",
            "2019-12-31 08:10:41,191 Epoch  97 Step:    26000 Batch Loss:     0.800894 Tokens per Sec:    13551, Lr: 0.000103\n",
            "2019-12-31 08:11:15,027 Example #0\n",
            "2019-12-31 08:11:15,028 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:11:15,028 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:11:15,028 \tHypothesis: Ihoho nana churobọ si ubiudu rayen , ayen me je nabọ muegbe rẹ oborẹ avwanre yonori .\n",
            "2019-12-31 08:11:15,028 Example #1\n",
            "2019-12-31 08:11:15,028 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:11:15,028 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:11:15,028 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:11:15,028 Example #2\n",
            "2019-12-31 08:11:15,028 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:11:15,028 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:11:15,028 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 08:11:15,028 Example #3\n",
            "2019-12-31 08:11:15,028 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:11:15,028 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:11:15,028 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn yovwin nọ wẹ nure .\n",
            "2019-12-31 08:11:15,029 Validation result (greedy) at epoch  97, step    26000: bleu:  14.73, loss: 62752.0742, ppl:  20.0981, duration: 33.8377s\n",
            "2019-12-31 08:11:29,947 Epoch  97 Step:    26100 Batch Loss:     0.860063 Tokens per Sec:    13616, Lr: 0.000103\n",
            "2019-12-31 08:11:44,672 Epoch  97 Step:    26200 Batch Loss:     0.839487 Tokens per Sec:    13409, Lr: 0.000103\n",
            "2019-12-31 08:11:46,150 Epoch  97: total training loss 198.43\n",
            "2019-12-31 08:11:46,151 EPOCH 98\n",
            "2019-12-31 08:11:59,666 Epoch  98 Step:    26300 Batch Loss:     0.830938 Tokens per Sec:    13622, Lr: 0.000103\n",
            "2019-12-31 08:12:14,541 Epoch  98 Step:    26400 Batch Loss:     0.935780 Tokens per Sec:    13437, Lr: 0.000103\n",
            "2019-12-31 08:12:26,447 Epoch  98: total training loss 197.23\n",
            "2019-12-31 08:12:26,447 EPOCH 99\n",
            "2019-12-31 08:12:29,317 Epoch  99 Step:    26500 Batch Loss:     0.875571 Tokens per Sec:    13825, Lr: 0.000103\n",
            "2019-12-31 08:12:44,262 Epoch  99 Step:    26600 Batch Loss:     0.352891 Tokens per Sec:    13702, Lr: 0.000103\n",
            "2019-12-31 08:12:59,066 Epoch  99 Step:    26700 Batch Loss:     0.833609 Tokens per Sec:    13488, Lr: 0.000103\n",
            "2019-12-31 08:13:06,533 Epoch  99: total training loss 194.85\n",
            "2019-12-31 08:13:06,533 EPOCH 100\n",
            "2019-12-31 08:13:14,083 Epoch 100 Step:    26800 Batch Loss:     0.637767 Tokens per Sec:    13720, Lr: 0.000103\n",
            "2019-12-31 08:13:28,823 Epoch 100 Step:    26900 Batch Loss:     0.532974 Tokens per Sec:    13263, Lr: 0.000103\n",
            "2019-12-31 08:13:43,810 Epoch 100 Step:    27000 Batch Loss:     0.383115 Tokens per Sec:    13737, Lr: 0.000103\n",
            "2019-12-31 08:14:17,614 Example #0\n",
            "2019-12-31 08:14:17,614 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:14:17,614 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:14:17,614 \tHypothesis: Ihoho nana nẹrhẹ ayen se muegbe rẹ ubiudu rayen , ayen me je nabọ muegbe rẹ oborẹ avwanre che ru .\n",
            "2019-12-31 08:14:17,614 Example #1\n",
            "2019-12-31 08:14:17,614 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:14:17,614 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:14:17,614 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:14:17,614 Example #2\n",
            "2019-12-31 08:14:17,615 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:14:17,615 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:14:17,615 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 08:14:17,615 Example #3\n",
            "2019-12-31 08:14:17,615 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:14:17,615 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:14:17,615 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn yovwin nọ wẹ nure .\n",
            "2019-12-31 08:14:17,615 Validation result (greedy) at epoch 100, step    27000: bleu:  15.06, loss: 62960.0586, ppl:  20.2990, duration: 33.8048s\n",
            "2019-12-31 08:14:20,675 Epoch 100: total training loss 195.52\n",
            "2019-12-31 08:14:20,675 EPOCH 101\n",
            "2019-12-31 08:14:32,553 Epoch 101 Step:    27100 Batch Loss:     0.880407 Tokens per Sec:    13567, Lr: 0.000103\n",
            "2019-12-31 08:14:47,508 Epoch 101 Step:    27200 Batch Loss:     0.542491 Tokens per Sec:    13770, Lr: 0.000103\n",
            "2019-12-31 08:15:01,034 Epoch 101: total training loss 193.58\n",
            "2019-12-31 08:15:01,034 EPOCH 102\n",
            "2019-12-31 08:15:02,403 Epoch 102 Step:    27300 Batch Loss:     0.871403 Tokens per Sec:    13309, Lr: 0.000103\n",
            "2019-12-31 08:15:17,348 Epoch 102 Step:    27400 Batch Loss:     0.793331 Tokens per Sec:    13614, Lr: 0.000103\n",
            "2019-12-31 08:15:32,176 Epoch 102 Step:    27500 Batch Loss:     0.909883 Tokens per Sec:    13253, Lr: 0.000103\n",
            "2019-12-31 08:15:41,366 Epoch 102: total training loss 192.21\n",
            "2019-12-31 08:15:41,366 EPOCH 103\n",
            "2019-12-31 08:15:47,148 Epoch 103 Step:    27600 Batch Loss:     0.856236 Tokens per Sec:    13411, Lr: 0.000103\n",
            "2019-12-31 08:16:02,063 Epoch 103 Step:    27700 Batch Loss:     0.708610 Tokens per Sec:    13737, Lr: 0.000103\n",
            "2019-12-31 08:16:16,996 Epoch 103 Step:    27800 Batch Loss:     0.633424 Tokens per Sec:    13478, Lr: 0.000103\n",
            "2019-12-31 08:16:21,740 Epoch 103: total training loss 191.63\n",
            "2019-12-31 08:16:21,741 EPOCH 104\n",
            "2019-12-31 08:16:31,880 Epoch 104 Step:    27900 Batch Loss:     0.491745 Tokens per Sec:    13300, Lr: 0.000103\n",
            "2019-12-31 08:16:46,706 Epoch 104 Step:    28000 Batch Loss:     0.372040 Tokens per Sec:    13779, Lr: 0.000103\n",
            "2019-12-31 08:17:20,419 Example #0\n",
            "2019-12-31 08:17:20,419 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:17:20,420 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:17:20,420 \tHypothesis: Ihoho nana nẹrhẹ ayen se muegbe rẹ ubiudu rayen , ayen me je nene odjekẹ rẹ avwanre .\n",
            "2019-12-31 08:17:20,420 Example #1\n",
            "2019-12-31 08:17:20,420 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:17:20,420 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:17:20,420 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:17:20,420 Example #2\n",
            "2019-12-31 08:17:20,420 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:17:20,420 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:17:20,420 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 08:17:20,420 Example #3\n",
            "2019-12-31 08:17:20,420 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:17:20,420 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:17:20,420 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn yovwin nọ wẹ - ẹ .\n",
            "2019-12-31 08:17:20,420 Validation result (greedy) at epoch 104, step    28000: bleu:  15.01, loss: 63579.7070, ppl:  20.9094, duration: 33.7139s\n",
            "2019-12-31 08:17:35,252 Epoch 104 Step:    28100 Batch Loss:     0.861210 Tokens per Sec:    13773, Lr: 0.000103\n",
            "2019-12-31 08:17:35,401 Epoch 104: total training loss 189.86\n",
            "2019-12-31 08:17:35,402 EPOCH 105\n",
            "2019-12-31 08:17:50,225 Epoch 105 Step:    28200 Batch Loss:     0.928001 Tokens per Sec:    13623, Lr: 0.000103\n",
            "2019-12-31 08:18:05,015 Epoch 105 Step:    28300 Batch Loss:     0.688904 Tokens per Sec:    13628, Lr: 0.000103\n",
            "2019-12-31 08:18:15,385 Epoch 105: total training loss 189.11\n",
            "2019-12-31 08:18:15,385 EPOCH 106\n",
            "2019-12-31 08:18:19,859 Epoch 106 Step:    28400 Batch Loss:     0.750112 Tokens per Sec:    12831, Lr: 0.000103\n",
            "2019-12-31 08:18:34,721 Epoch 106 Step:    28500 Batch Loss:     0.744182 Tokens per Sec:    13755, Lr: 0.000103\n",
            "2019-12-31 08:18:49,336 Epoch 106 Step:    28600 Batch Loss:     0.835842 Tokens per Sec:    13869, Lr: 0.000103\n",
            "2019-12-31 08:18:55,260 Epoch 106: total training loss 188.94\n",
            "2019-12-31 08:18:55,261 EPOCH 107\n",
            "2019-12-31 08:19:04,177 Epoch 107 Step:    28700 Batch Loss:     0.718878 Tokens per Sec:    13587, Lr: 0.000103\n",
            "2019-12-31 08:19:18,887 Epoch 107 Step:    28800 Batch Loss:     0.802731 Tokens per Sec:    13650, Lr: 0.000103\n",
            "2019-12-31 08:19:33,537 Epoch 107 Step:    28900 Batch Loss:     0.671088 Tokens per Sec:    13860, Lr: 0.000103\n",
            "2019-12-31 08:19:35,021 Epoch 107: total training loss 187.64\n",
            "2019-12-31 08:19:35,021 EPOCH 108\n",
            "2019-12-31 08:19:48,220 Epoch 108 Step:    29000 Batch Loss:     0.744159 Tokens per Sec:    14015, Lr: 0.000103\n",
            "2019-12-31 08:20:21,944 Example #0\n",
            "2019-12-31 08:20:21,945 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:20:21,945 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:20:21,945 \tHypothesis: Ihoho nana nẹrhẹ ihwo roro nẹ ayen che muegbe rẹ ubiudu avwanre , je nene odjekẹ rẹ avwanre .\n",
            "2019-12-31 08:20:21,945 Example #1\n",
            "2019-12-31 08:20:21,945 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:20:21,945 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:20:21,945 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:20:21,945 Example #2\n",
            "2019-12-31 08:20:21,945 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:20:21,945 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:20:21,946 \tHypothesis: Die kọyen egbomọphẹ ?\n",
            "2019-12-31 08:20:21,946 Example #3\n",
            "2019-12-31 08:20:21,946 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:20:21,946 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:20:21,946 \tHypothesis: Ọ je hepha kẹ ukoko wẹn vẹ ukoko wẹn ro vwo oka rẹ ovwan na ọfa - a .\n",
            "2019-12-31 08:20:21,946 Validation result (greedy) at epoch 108, step    29000: bleu:  14.89, loss: 63843.8672, ppl:  21.1752, duration: 33.7256s\n",
            "2019-12-31 08:20:36,682 Epoch 108 Step:    29100 Batch Loss:     0.891153 Tokens per Sec:    13388, Lr: 0.000103\n",
            "2019-12-31 08:20:48,392 Epoch 108: total training loss 187.12\n",
            "2019-12-31 08:20:48,392 EPOCH 109\n",
            "2019-12-31 08:20:51,382 Epoch 109 Step:    29200 Batch Loss:     0.728129 Tokens per Sec:    14000, Lr: 0.000103\n",
            "2019-12-31 08:21:06,115 Epoch 109 Step:    29300 Batch Loss:     0.619693 Tokens per Sec:    13408, Lr: 0.000103\n",
            "2019-12-31 08:21:20,918 Epoch 109 Step:    29400 Batch Loss:     0.911039 Tokens per Sec:    13772, Lr: 0.000103\n",
            "2019-12-31 08:21:28,466 Epoch 109: total training loss 187.66\n",
            "2019-12-31 08:21:28,466 EPOCH 110\n",
            "2019-12-31 08:21:35,511 Epoch 110 Step:    29500 Batch Loss:     0.824383 Tokens per Sec:    13307, Lr: 0.000103\n",
            "2019-12-31 08:21:50,191 Epoch 110 Step:    29600 Batch Loss:     0.877905 Tokens per Sec:    14070, Lr: 0.000103\n",
            "2019-12-31 08:22:04,899 Epoch 110 Step:    29700 Batch Loss:     0.511252 Tokens per Sec:    13970, Lr: 0.000103\n",
            "2019-12-31 08:22:08,037 Epoch 110: total training loss 185.65\n",
            "2019-12-31 08:22:08,037 EPOCH 111\n",
            "2019-12-31 08:22:19,345 Epoch 111 Step:    29800 Batch Loss:     0.817970 Tokens per Sec:    13773, Lr: 0.000103\n",
            "2019-12-31 08:22:33,970 Epoch 111 Step:    29900 Batch Loss:     0.541562 Tokens per Sec:    13603, Lr: 0.000103\n",
            "2019-12-31 08:22:47,925 Epoch 111: total training loss 186.77\n",
            "2019-12-31 08:22:47,925 EPOCH 112\n",
            "2019-12-31 08:22:48,548 Epoch 112 Step:    30000 Batch Loss:     0.565645 Tokens per Sec:    11169, Lr: 0.000103\n",
            "2019-12-31 08:23:22,160 Example #0\n",
            "2019-12-31 08:23:22,161 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:23:22,161 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:23:22,161 \tHypothesis: Ihoho nana nẹrhẹ e se muegbe rẹ ubiudu avwanre , ayen me je nabọ muegbe rẹ oborẹ avwanre che ru .\n",
            "2019-12-31 08:23:22,161 Example #1\n",
            "2019-12-31 08:23:22,161 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:23:22,161 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:23:22,161 \tHypothesis: Enẹna , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:23:22,161 Example #2\n",
            "2019-12-31 08:23:22,161 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:23:22,161 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:23:22,161 \tHypothesis: Die yen egbomọphẹ ?\n",
            "2019-12-31 08:23:22,161 Example #3\n",
            "2019-12-31 08:23:22,161 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:23:22,161 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:23:22,162 \tHypothesis: Ọ je dianẹ ovwan je yan obaro vwẹ ukoko wẹn ọfa - a .\n",
            "2019-12-31 08:23:22,162 Validation result (greedy) at epoch 112, step    30000: bleu:  14.17, loss: 63885.1172, ppl:  21.2170, duration: 33.6139s\n",
            "2019-12-31 08:23:36,850 Epoch 112 Step:    30100 Batch Loss:     0.451706 Tokens per Sec:    13599, Lr: 0.000103\n",
            "2019-12-31 08:23:51,510 Epoch 112 Step:    30200 Batch Loss:     0.474423 Tokens per Sec:    14008, Lr: 0.000103\n",
            "2019-12-31 08:24:01,170 Epoch 112: total training loss 184.29\n",
            "2019-12-31 08:24:01,170 EPOCH 113\n",
            "2019-12-31 08:24:06,296 Epoch 113 Step:    30300 Batch Loss:     0.697271 Tokens per Sec:    13793, Lr: 0.000103\n",
            "2019-12-31 08:24:20,758 Epoch 113 Step:    30400 Batch Loss:     0.743529 Tokens per Sec:    13803, Lr: 0.000103\n",
            "2019-12-31 08:24:35,331 Epoch 113 Step:    30500 Batch Loss:     0.725941 Tokens per Sec:    13680, Lr: 0.000103\n",
            "2019-12-31 08:24:40,937 Epoch 113: total training loss 183.82\n",
            "2019-12-31 08:24:40,937 EPOCH 114\n",
            "2019-12-31 08:24:50,100 Epoch 114 Step:    30600 Batch Loss:     0.692801 Tokens per Sec:    14110, Lr: 0.000103\n",
            "2019-12-31 08:25:04,741 Epoch 114 Step:    30700 Batch Loss:     0.356915 Tokens per Sec:    13379, Lr: 0.000103\n",
            "2019-12-31 08:25:19,367 Epoch 114 Step:    30800 Batch Loss:     0.715890 Tokens per Sec:    13855, Lr: 0.000103\n",
            "2019-12-31 08:25:20,565 Epoch 114: total training loss 181.90\n",
            "2019-12-31 08:25:20,565 EPOCH 115\n",
            "2019-12-31 08:25:33,996 Epoch 115 Step:    30900 Batch Loss:     0.760353 Tokens per Sec:    13800, Lr: 0.000103\n",
            "2019-12-31 08:25:48,510 Epoch 115 Step:    31000 Batch Loss:     0.846082 Tokens per Sec:    13814, Lr: 0.000103\n",
            "2019-12-31 08:26:22,152 Example #0\n",
            "2019-12-31 08:26:22,153 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:26:22,153 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:26:22,153 \tHypothesis: Ihoho nana nẹrhẹ ihwo roro nẹ ayen che muegbe rẹ ubiudu avwanre , je nene odjekẹ rẹ avwanre .\n",
            "2019-12-31 08:26:22,153 Example #1\n",
            "2019-12-31 08:26:22,153 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:26:22,153 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:26:22,153 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:26:22,153 Example #2\n",
            "2019-12-31 08:26:22,153 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:26:22,153 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:26:22,153 \tHypothesis: Die yen egbomọphẹ ?\n",
            "2019-12-31 08:26:22,153 Example #3\n",
            "2019-12-31 08:26:22,154 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:26:22,154 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:26:22,154 \tHypothesis: Ọ je hepha kẹ ukoko wẹn jovwo - o .\n",
            "2019-12-31 08:26:22,154 Validation result (greedy) at epoch 115, step    31000: bleu:  14.80, loss: 64090.2539, ppl:  21.4262, duration: 33.6435s\n",
            "2019-12-31 08:26:33,461 Epoch 115: total training loss 179.71\n",
            "2019-12-31 08:26:33,461 EPOCH 116\n",
            "2019-12-31 08:26:36,880 Epoch 116 Step:    31100 Batch Loss:     0.547474 Tokens per Sec:    13448, Lr: 0.000072\n",
            "2019-12-31 08:26:51,474 Epoch 116 Step:    31200 Batch Loss:     0.625631 Tokens per Sec:    13932, Lr: 0.000072\n",
            "2019-12-31 08:27:06,060 Epoch 116 Step:    31300 Batch Loss:     0.768800 Tokens per Sec:    13774, Lr: 0.000072\n",
            "2019-12-31 08:27:12,989 Epoch 116: total training loss 178.01\n",
            "2019-12-31 08:27:12,989 EPOCH 117\n",
            "2019-12-31 08:27:20,635 Epoch 117 Step:    31400 Batch Loss:     0.576995 Tokens per Sec:    13612, Lr: 0.000072\n",
            "2019-12-31 08:27:35,302 Epoch 117 Step:    31500 Batch Loss:     0.813073 Tokens per Sec:    13943, Lr: 0.000072\n",
            "2019-12-31 08:27:49,812 Epoch 117 Step:    31600 Batch Loss:     0.714269 Tokens per Sec:    13863, Lr: 0.000072\n",
            "2019-12-31 08:27:52,404 Epoch 117: total training loss 176.84\n",
            "2019-12-31 08:27:52,404 EPOCH 118\n",
            "2019-12-31 08:28:04,401 Epoch 118 Step:    31700 Batch Loss:     0.583710 Tokens per Sec:    13717, Lr: 0.000072\n",
            "2019-12-31 08:28:18,978 Epoch 118 Step:    31800 Batch Loss:     0.643689 Tokens per Sec:    13814, Lr: 0.000072\n",
            "2019-12-31 08:28:31,886 Epoch 118: total training loss 176.23\n",
            "2019-12-31 08:28:31,886 EPOCH 119\n",
            "2019-12-31 08:28:33,650 Epoch 119 Step:    31900 Batch Loss:     0.445811 Tokens per Sec:    13144, Lr: 0.000072\n",
            "2019-12-31 08:28:48,137 Epoch 119 Step:    32000 Batch Loss:     0.812937 Tokens per Sec:    13998, Lr: 0.000072\n",
            "2019-12-31 08:29:21,645 Example #0\n",
            "2019-12-31 08:29:21,645 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:29:21,645 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:29:21,645 \tHypothesis: Ihoho nana toroba oborẹ ubiudu avwanre se vwo muegbe rẹ ayen che vwo ru nene .\n",
            "2019-12-31 08:29:21,645 Example #1\n",
            "2019-12-31 08:29:21,646 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:29:21,646 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:29:21,646 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:29:21,646 Example #2\n",
            "2019-12-31 08:29:21,646 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:29:21,646 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:29:21,646 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ?\n",
            "2019-12-31 08:29:21,646 Example #3\n",
            "2019-12-31 08:29:21,646 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:29:21,646 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:29:21,646 \tHypothesis: Ọ je hepha kẹ ukoko wẹn ri jovwo na , wo ji no ọtiọye - en .\n",
            "2019-12-31 08:29:21,646 Validation result (greedy) at epoch 119, step    32000: bleu:  15.04, loss: 64068.0273, ppl:  21.4034, duration: 33.5093s\n",
            "2019-12-31 08:29:36,081 Epoch 119 Step:    32100 Batch Loss:     0.580997 Tokens per Sec:    13755, Lr: 0.000072\n",
            "2019-12-31 08:29:44,696 Epoch 119: total training loss 176.20\n",
            "2019-12-31 08:29:44,696 EPOCH 120\n",
            "2019-12-31 08:29:50,684 Epoch 120 Step:    32200 Batch Loss:     0.782154 Tokens per Sec:    13827, Lr: 0.000072\n",
            "2019-12-31 08:30:05,160 Epoch 120 Step:    32300 Batch Loss:     0.705442 Tokens per Sec:    13865, Lr: 0.000072\n",
            "2019-12-31 08:30:19,643 Epoch 120 Step:    32400 Batch Loss:     0.641414 Tokens per Sec:    13829, Lr: 0.000072\n",
            "2019-12-31 08:30:24,000 Epoch 120: total training loss 175.48\n",
            "2019-12-31 08:30:24,000 EPOCH 121\n",
            "2019-12-31 08:30:34,179 Epoch 121 Step:    32500 Batch Loss:     0.456334 Tokens per Sec:    13758, Lr: 0.000072\n",
            "2019-12-31 08:30:48,627 Epoch 121 Step:    32600 Batch Loss:     0.797962 Tokens per Sec:    14015, Lr: 0.000072\n",
            "2019-12-31 08:31:03,008 Epoch 121 Step:    32700 Batch Loss:     0.701287 Tokens per Sec:    13838, Lr: 0.000072\n",
            "2019-12-31 08:31:03,303 Epoch 121: total training loss 174.83\n",
            "2019-12-31 08:31:03,303 EPOCH 122\n",
            "2019-12-31 08:31:17,798 Epoch 122 Step:    32800 Batch Loss:     0.788418 Tokens per Sec:    13909, Lr: 0.000072\n",
            "2019-12-31 08:31:32,423 Epoch 122 Step:    32900 Batch Loss:     0.700216 Tokens per Sec:    13811, Lr: 0.000072\n",
            "2019-12-31 08:31:42,642 Epoch 122: total training loss 172.71\n",
            "2019-12-31 08:31:42,642 EPOCH 123\n",
            "2019-12-31 08:31:46,826 Epoch 123 Step:    33000 Batch Loss:     0.682884 Tokens per Sec:    14435, Lr: 0.000072\n",
            "2019-12-31 08:32:20,349 Example #0\n",
            "2019-12-31 08:32:20,350 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:32:20,350 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:32:20,350 \tHypothesis: Ihoho nana toroba obo rehẹ ubiudu rẹ avwanre , kidie ayen muegbe rẹ ayen vwo nene ọrhuẹrẹphiyotọ na .\n",
            "2019-12-31 08:32:20,350 Example #1\n",
            "2019-12-31 08:32:20,350 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:32:20,350 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:32:20,350 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:32:20,350 Example #2\n",
            "2019-12-31 08:32:20,350 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:32:20,350 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:32:20,350 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ?\n",
            "2019-12-31 08:32:20,350 Example #3\n",
            "2019-12-31 08:32:20,350 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:32:20,350 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:32:20,350 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn ro vwo oka rẹ ovwan na ọfa - a .\n",
            "2019-12-31 08:32:20,350 Validation result (greedy) at epoch 123, step    33000: bleu:  15.01, loss: 64300.8086, ppl:  21.6430, duration: 33.5238s\n",
            "2019-12-31 08:32:34,814 Epoch 123 Step:    33100 Batch Loss:     0.708807 Tokens per Sec:    13836, Lr: 0.000072\n",
            "2019-12-31 08:32:49,398 Epoch 123 Step:    33200 Batch Loss:     0.482665 Tokens per Sec:    14072, Lr: 0.000072\n",
            "2019-12-31 08:32:55,234 Epoch 123: total training loss 172.43\n",
            "2019-12-31 08:32:55,234 EPOCH 124\n",
            "2019-12-31 08:33:03,927 Epoch 124 Step:    33300 Batch Loss:     0.745029 Tokens per Sec:    14101, Lr: 0.000072\n",
            "2019-12-31 08:33:18,480 Epoch 124 Step:    33400 Batch Loss:     0.749112 Tokens per Sec:    13505, Lr: 0.000072\n",
            "2019-12-31 08:33:33,032 Epoch 124 Step:    33500 Batch Loss:     0.645195 Tokens per Sec:    13862, Lr: 0.000072\n",
            "2019-12-31 08:33:34,660 Epoch 124: total training loss 172.04\n",
            "2019-12-31 08:33:34,660 EPOCH 125\n",
            "2019-12-31 08:33:47,673 Epoch 125 Step:    33600 Batch Loss:     0.725643 Tokens per Sec:    13701, Lr: 0.000072\n",
            "2019-12-31 08:34:02,083 Epoch 125 Step:    33700 Batch Loss:     0.329218 Tokens per Sec:    13977, Lr: 0.000072\n",
            "2019-12-31 08:34:13,869 Epoch 125: total training loss 172.38\n",
            "2019-12-31 08:34:13,869 EPOCH 126\n",
            "2019-12-31 08:34:16,570 Epoch 126 Step:    33800 Batch Loss:     0.455275 Tokens per Sec:    14614, Lr: 0.000072\n",
            "2019-12-31 08:34:30,970 Epoch 126 Step:    33900 Batch Loss:     0.687133 Tokens per Sec:    13738, Lr: 0.000072\n",
            "2019-12-31 08:34:45,472 Epoch 126 Step:    34000 Batch Loss:     0.614258 Tokens per Sec:    14191, Lr: 0.000072\n",
            "2019-12-31 08:35:19,016 Example #0\n",
            "2019-12-31 08:35:19,017 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:35:19,017 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:35:19,017 \tHypothesis: Ihoho nana nẹrhẹ ihwo roro nẹ ayen che muegbe rẹ ubiudu avwanre , je nene odjekẹ rẹ avwanre .\n",
            "2019-12-31 08:35:19,017 Example #1\n",
            "2019-12-31 08:35:19,017 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:35:19,017 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:35:19,017 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:35:19,017 Example #2\n",
            "2019-12-31 08:35:19,017 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:35:19,017 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:35:19,018 \tHypothesis: Die kọyen egbomọphẹ ?\n",
            "2019-12-31 08:35:19,018 Example #3\n",
            "2019-12-31 08:35:19,018 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:35:19,018 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:35:19,018 \tHypothesis: Ọ je hepha kẹ ukoko wẹn jovwo , wo ji no nẹ ukoko wẹn yovwin nọ - ọ .\n",
            "2019-12-31 08:35:19,018 Validation result (greedy) at epoch 126, step    34000: bleu:  14.89, loss: 64409.6836, ppl:  21.7559, duration: 33.5455s\n",
            "2019-12-31 08:35:26,814 Epoch 126: total training loss 171.41\n",
            "2019-12-31 08:35:26,814 EPOCH 127\n",
            "2019-12-31 08:35:33,465 Epoch 127 Step:    34100 Batch Loss:     0.470043 Tokens per Sec:    13709, Lr: 0.000072\n",
            "2019-12-31 08:35:47,978 Epoch 127 Step:    34200 Batch Loss:     0.549079 Tokens per Sec:    13924, Lr: 0.000072\n",
            "2019-12-31 08:36:02,424 Epoch 127 Step:    34300 Batch Loss:     0.335187 Tokens per Sec:    14004, Lr: 0.000072\n",
            "2019-12-31 08:36:06,125 Epoch 127: total training loss 171.47\n",
            "2019-12-31 08:36:06,126 EPOCH 128\n",
            "2019-12-31 08:36:16,903 Epoch 128 Step:    34400 Batch Loss:     0.479019 Tokens per Sec:    13803, Lr: 0.000072\n",
            "2019-12-31 08:36:31,386 Epoch 128 Step:    34500 Batch Loss:     0.529685 Tokens per Sec:    13881, Lr: 0.000072\n",
            "2019-12-31 08:36:45,419 Epoch 128: total training loss 170.12\n",
            "2019-12-31 08:36:45,420 EPOCH 129\n",
            "2019-12-31 08:36:45,897 Epoch 129 Step:    34600 Batch Loss:     0.621966 Tokens per Sec:    10921, Lr: 0.000072\n",
            "2019-12-31 08:37:00,417 Epoch 129 Step:    34700 Batch Loss:     0.296525 Tokens per Sec:    13966, Lr: 0.000072\n",
            "2019-12-31 08:37:14,951 Epoch 129 Step:    34800 Batch Loss:     0.709406 Tokens per Sec:    14071, Lr: 0.000072\n",
            "2019-12-31 08:37:24,555 Epoch 129: total training loss 169.07\n",
            "2019-12-31 08:37:24,555 EPOCH 130\n",
            "2019-12-31 08:37:29,425 Epoch 130 Step:    34900 Batch Loss:     0.679545 Tokens per Sec:    14058, Lr: 0.000072\n",
            "2019-12-31 08:37:43,824 Epoch 130 Step:    35000 Batch Loss:     0.619538 Tokens per Sec:    13891, Lr: 0.000072\n",
            "2019-12-31 08:38:17,317 Example #0\n",
            "2019-12-31 08:38:17,318 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:38:17,318 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:38:17,318 \tHypothesis: Ihoho nana toroba obo rehẹ ubiudu rayen , kidie ayen muegbe rẹ ayen vwo nene odjekẹ na .\n",
            "2019-12-31 08:38:17,318 Example #1\n",
            "2019-12-31 08:38:17,318 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:38:17,318 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:38:17,318 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:38:17,318 Example #2\n",
            "2019-12-31 08:38:17,318 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:38:17,318 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:38:17,318 \tHypothesis: Die kọyen egbomọphẹ ?\n",
            "2019-12-31 08:38:17,318 Example #3\n",
            "2019-12-31 08:38:17,318 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:38:17,318 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:38:17,319 \tHypothesis: Ọ je hepha kẹ ukoko wẹn jovwo , wo ji no ọtiọye - en .\n",
            "2019-12-31 08:38:17,319 Validation result (greedy) at epoch 130, step    35000: bleu:  14.92, loss: 64785.3125, ppl:  22.1502, duration: 33.4939s\n",
            "2019-12-31 08:38:31,732 Epoch 130 Step:    35100 Batch Loss:     0.425837 Tokens per Sec:    14068, Lr: 0.000072\n",
            "2019-12-31 08:38:37,186 Epoch 130: total training loss 168.85\n",
            "2019-12-31 08:38:37,186 EPOCH 131\n",
            "2019-12-31 08:38:46,263 Epoch 131 Step:    35200 Batch Loss:     0.675949 Tokens per Sec:    13969, Lr: 0.000072\n",
            "2019-12-31 08:39:00,906 Epoch 131 Step:    35300 Batch Loss:     0.569329 Tokens per Sec:    13871, Lr: 0.000072\n",
            "2019-12-31 08:39:15,401 Epoch 131 Step:    35400 Batch Loss:     0.749621 Tokens per Sec:    13825, Lr: 0.000072\n",
            "2019-12-31 08:39:16,440 Epoch 131: total training loss 167.70\n",
            "2019-12-31 08:39:16,441 EPOCH 132\n",
            "2019-12-31 08:39:29,939 Epoch 132 Step:    35500 Batch Loss:     0.570535 Tokens per Sec:    13893, Lr: 0.000072\n",
            "2019-12-31 08:39:44,448 Epoch 132 Step:    35600 Batch Loss:     0.694527 Tokens per Sec:    14036, Lr: 0.000072\n",
            "2019-12-31 08:39:55,453 Epoch 132: total training loss 167.57\n",
            "2019-12-31 08:39:55,453 EPOCH 133\n",
            "2019-12-31 08:39:58,948 Epoch 133 Step:    35700 Batch Loss:     0.670402 Tokens per Sec:    13051, Lr: 0.000072\n",
            "2019-12-31 08:40:13,536 Epoch 133 Step:    35800 Batch Loss:     0.434567 Tokens per Sec:    14156, Lr: 0.000072\n",
            "2019-12-31 08:40:28,058 Epoch 133 Step:    35900 Batch Loss:     0.658459 Tokens per Sec:    14040, Lr: 0.000072\n",
            "2019-12-31 08:40:34,659 Epoch 133: total training loss 167.31\n",
            "2019-12-31 08:40:34,659 EPOCH 134\n",
            "2019-12-31 08:40:42,449 Epoch 134 Step:    36000 Batch Loss:     0.556847 Tokens per Sec:    13505, Lr: 0.000072\n",
            "2019-12-31 08:41:15,959 Example #0\n",
            "2019-12-31 08:41:15,960 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:41:15,960 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:41:15,960 \tHypothesis: Omaẹkparọ yen nẹrhẹ ihwo efa muegbe rẹ ayen vwo muegbe rẹ obo rehẹ ubiudu avwanre , je nene odjekẹ na .\n",
            "2019-12-31 08:41:15,960 Example #1\n",
            "2019-12-31 08:41:15,960 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:41:15,960 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:41:15,960 \tHypothesis: Enẹna , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:41:15,960 Example #2\n",
            "2019-12-31 08:41:15,960 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:41:15,960 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:41:15,960 \tHypothesis: Die kọyen egbomọphẹ ?\n",
            "2019-12-31 08:41:15,960 Example #3\n",
            "2019-12-31 08:41:15,961 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:41:15,961 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:41:15,961 \tHypothesis: Wọ je guọnọ nẹ ukoko wẹn ro vwo oka rẹ ovwan na ọfa - a .\n",
            "2019-12-31 08:41:15,961 Validation result (greedy) at epoch 134, step    36000: bleu:  14.50, loss: 64826.3672, ppl:  22.1938, duration: 33.5113s\n",
            "2019-12-31 08:41:30,509 Epoch 134 Step:    36100 Batch Loss:     0.788069 Tokens per Sec:    14193, Lr: 0.000072\n",
            "2019-12-31 08:41:44,908 Epoch 134 Step:    36200 Batch Loss:     0.550192 Tokens per Sec:    13956, Lr: 0.000072\n",
            "2019-12-31 08:41:47,342 Epoch 134: total training loss 167.40\n",
            "2019-12-31 08:41:47,342 EPOCH 135\n",
            "2019-12-31 08:41:59,600 Epoch 135 Step:    36300 Batch Loss:     0.454963 Tokens per Sec:    13954, Lr: 0.000072\n",
            "2019-12-31 08:42:14,033 Epoch 135 Step:    36400 Batch Loss:     0.555912 Tokens per Sec:    13706, Lr: 0.000072\n",
            "2019-12-31 08:42:26,802 Epoch 135: total training loss 166.62\n",
            "2019-12-31 08:42:26,803 EPOCH 136\n",
            "2019-12-31 08:42:28,590 Epoch 136 Step:    36500 Batch Loss:     0.545697 Tokens per Sec:    13688, Lr: 0.000072\n",
            "2019-12-31 08:42:43,132 Epoch 136 Step:    36600 Batch Loss:     0.414271 Tokens per Sec:    13848, Lr: 0.000072\n",
            "2019-12-31 08:42:57,747 Epoch 136 Step:    36700 Batch Loss:     0.489893 Tokens per Sec:    14059, Lr: 0.000072\n",
            "2019-12-31 08:43:06,055 Epoch 136: total training loss 164.71\n",
            "2019-12-31 08:43:06,055 EPOCH 137\n",
            "2019-12-31 08:43:12,283 Epoch 137 Step:    36800 Batch Loss:     0.737937 Tokens per Sec:    13703, Lr: 0.000072\n",
            "2019-12-31 08:43:26,841 Epoch 137 Step:    36900 Batch Loss:     0.663248 Tokens per Sec:    14081, Lr: 0.000072\n",
            "2019-12-31 08:43:41,102 Epoch 137 Step:    37000 Batch Loss:     0.484364 Tokens per Sec:    13674, Lr: 0.000072\n",
            "2019-12-31 08:44:14,580 Example #0\n",
            "2019-12-31 08:44:14,580 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:44:14,581 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:44:14,581 \tHypothesis: Omaẹkparọ yen nẹrhẹ ihwo ni ubiudu rayen ghanghanre , ayen me je nabọ muegbe rẹ oborẹ avwanre che nene ọrhuẹrẹphiyotọ na .\n",
            "2019-12-31 08:44:14,581 Example #1\n",
            "2019-12-31 08:44:14,581 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:44:14,581 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:44:14,581 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:44:14,581 Example #2\n",
            "2019-12-31 08:44:14,581 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:44:14,581 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:44:14,581 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 08:44:14,581 Example #3\n",
            "2019-12-31 08:44:14,581 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:44:14,581 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:44:14,581 \tHypothesis: Ọ je dianẹ ovwan vẹ ukoko wẹn ro vwo ẹghwọ na - a .\n",
            "2019-12-31 08:44:14,581 Validation result (greedy) at epoch 137, step    37000: bleu:  14.71, loss: 65046.7500, ppl:  22.4289, duration: 33.4795s\n",
            "2019-12-31 08:44:18,653 Epoch 137: total training loss 165.17\n",
            "2019-12-31 08:44:18,653 EPOCH 138\n",
            "2019-12-31 08:44:29,148 Epoch 138 Step:    37100 Batch Loss:     0.660452 Tokens per Sec:    13988, Lr: 0.000050\n",
            "2019-12-31 08:44:43,642 Epoch 138 Step:    37200 Batch Loss:     0.658041 Tokens per Sec:    14156, Lr: 0.000050\n",
            "2019-12-31 08:44:57,747 Epoch 138: total training loss 162.98\n",
            "2019-12-31 08:44:57,747 EPOCH 139\n",
            "2019-12-31 08:44:58,080 Epoch 139 Step:    37300 Batch Loss:     0.659146 Tokens per Sec:    13309, Lr: 0.000050\n",
            "2019-12-31 08:45:12,701 Epoch 139 Step:    37400 Batch Loss:     0.622420 Tokens per Sec:    13942, Lr: 0.000050\n",
            "2019-12-31 08:45:27,176 Epoch 139 Step:    37500 Batch Loss:     0.640145 Tokens per Sec:    13825, Lr: 0.000050\n",
            "2019-12-31 08:45:37,097 Epoch 139: total training loss 161.53\n",
            "2019-12-31 08:45:37,098 EPOCH 140\n",
            "2019-12-31 08:45:41,881 Epoch 140 Step:    37600 Batch Loss:     0.494648 Tokens per Sec:    14162, Lr: 0.000050\n",
            "2019-12-31 08:45:56,479 Epoch 140 Step:    37700 Batch Loss:     0.602178 Tokens per Sec:    14065, Lr: 0.000050\n",
            "2019-12-31 08:46:11,004 Epoch 140 Step:    37800 Batch Loss:     0.734182 Tokens per Sec:    13795, Lr: 0.000050\n",
            "2019-12-31 08:46:16,137 Epoch 140: total training loss 160.72\n",
            "2019-12-31 08:46:16,137 EPOCH 141\n",
            "2019-12-31 08:46:25,679 Epoch 141 Step:    37900 Batch Loss:     0.720358 Tokens per Sec:    14316, Lr: 0.000050\n",
            "2019-12-31 08:46:40,202 Epoch 141 Step:    38000 Batch Loss:     0.777925 Tokens per Sec:    13369, Lr: 0.000050\n",
            "2019-12-31 08:47:13,748 Example #0\n",
            "2019-12-31 08:47:13,748 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:47:13,748 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:47:13,749 \tHypothesis: Ihoho nana pha ghanghanre vwẹ idjerhe tiọna , ayen muegbe rẹ ayen vwo nene odjekẹ rẹ avwanre .\n",
            "2019-12-31 08:47:13,749 Example #1\n",
            "2019-12-31 08:47:13,749 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:47:13,749 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:47:13,749 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:47:13,749 Example #2\n",
            "2019-12-31 08:47:13,749 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:47:13,749 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:47:13,749 \tHypothesis: Kẹ egbomọphẹ vọ yen o vwo ruo ?\n",
            "2019-12-31 08:47:13,749 Example #3\n",
            "2019-12-31 08:47:13,749 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:47:13,749 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:47:13,749 \tHypothesis: Ọ je hepha kẹ ukoko wẹn vẹ ukoko wẹn jovwo re .\n",
            "2019-12-31 08:47:13,750 Validation result (greedy) at epoch 141, step    38000: bleu:  14.58, loss: 65116.2500, ppl:  22.5036, duration: 33.5468s\n",
            "2019-12-31 08:47:28,335 Epoch 141 Step:    38100 Batch Loss:     0.362765 Tokens per Sec:    14178, Lr: 0.000050\n",
            "2019-12-31 08:47:28,920 Epoch 141: total training loss 160.48\n",
            "2019-12-31 08:47:28,920 EPOCH 142\n",
            "2019-12-31 08:47:42,803 Epoch 142 Step:    38200 Batch Loss:     0.508243 Tokens per Sec:    13863, Lr: 0.000050\n",
            "2019-12-31 08:47:57,412 Epoch 142 Step:    38300 Batch Loss:     0.531780 Tokens per Sec:    13963, Lr: 0.000050\n",
            "2019-12-31 08:48:08,312 Epoch 142: total training loss 161.02\n",
            "2019-12-31 08:48:08,312 EPOCH 143\n",
            "2019-12-31 08:48:11,844 Epoch 143 Step:    38400 Batch Loss:     0.644076 Tokens per Sec:    13629, Lr: 0.000050\n",
            "2019-12-31 08:48:26,408 Epoch 143 Step:    38500 Batch Loss:     0.673954 Tokens per Sec:    14067, Lr: 0.000050\n",
            "2019-12-31 08:48:40,746 Epoch 143 Step:    38600 Batch Loss:     0.714925 Tokens per Sec:    13698, Lr: 0.000050\n",
            "2019-12-31 08:48:47,597 Epoch 143: total training loss 161.21\n",
            "2019-12-31 08:48:47,597 EPOCH 144\n",
            "2019-12-31 08:48:55,251 Epoch 144 Step:    38700 Batch Loss:     0.735921 Tokens per Sec:    13769, Lr: 0.000050\n",
            "2019-12-31 08:49:09,757 Epoch 144 Step:    38800 Batch Loss:     0.715990 Tokens per Sec:    13943, Lr: 0.000050\n",
            "2019-12-31 08:49:24,254 Epoch 144 Step:    38900 Batch Loss:     0.320705 Tokens per Sec:    14152, Lr: 0.000050\n",
            "2019-12-31 08:49:26,711 Epoch 144: total training loss 160.16\n",
            "2019-12-31 08:49:26,711 EPOCH 145\n",
            "2019-12-31 08:49:38,919 Epoch 145 Step:    39000 Batch Loss:     0.636216 Tokens per Sec:    14106, Lr: 0.000050\n",
            "2019-12-31 08:50:12,367 Example #0\n",
            "2019-12-31 08:50:12,367 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:50:12,367 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:50:12,367 \tHypothesis: Ihoho nana pha ghanghanre vwẹ idjerhe rẹ ẹwẹn avwanre se vwo muegbe rẹ ayen vwo nene odjekẹ na .\n",
            "2019-12-31 08:50:12,367 Example #1\n",
            "2019-12-31 08:50:12,367 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:50:12,367 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:50:12,367 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:50:12,367 Example #2\n",
            "2019-12-31 08:50:12,367 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:50:12,368 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:50:12,368 \tHypothesis: Die kọyen egbomọphẹ ?\n",
            "2019-12-31 08:50:12,368 Example #3\n",
            "2019-12-31 08:50:12,368 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:50:12,368 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:50:12,368 \tHypothesis: Ọ je dianẹ ovwan je hepha vwẹ ukoko wẹn ro vwo oba na ọfa - a .\n",
            "2019-12-31 08:50:12,368 Validation result (greedy) at epoch 145, step    39000: bleu:  14.54, loss: 65140.5664, ppl:  22.5297, duration: 33.4482s\n",
            "2019-12-31 08:50:26,878 Epoch 145 Step:    39100 Batch Loss:     0.511276 Tokens per Sec:    13772, Lr: 0.000050\n",
            "2019-12-31 08:50:39,297 Epoch 145: total training loss 158.48\n",
            "2019-12-31 08:50:39,298 EPOCH 146\n",
            "2019-12-31 08:50:41,497 Epoch 146 Step:    39200 Batch Loss:     0.620622 Tokens per Sec:    13332, Lr: 0.000050\n",
            "2019-12-31 08:50:56,134 Epoch 146 Step:    39300 Batch Loss:     0.690100 Tokens per Sec:    13946, Lr: 0.000050\n",
            "2019-12-31 08:51:10,724 Epoch 146 Step:    39400 Batch Loss:     0.558163 Tokens per Sec:    13737, Lr: 0.000050\n",
            "2019-12-31 08:51:18,756 Epoch 146: total training loss 159.18\n",
            "2019-12-31 08:51:18,756 EPOCH 147\n",
            "2019-12-31 08:51:25,382 Epoch 147 Step:    39500 Batch Loss:     0.600398 Tokens per Sec:    14074, Lr: 0.000050\n",
            "2019-12-31 08:51:39,805 Epoch 147 Step:    39600 Batch Loss:     0.676031 Tokens per Sec:    13651, Lr: 0.000050\n",
            "2019-12-31 08:51:54,327 Epoch 147 Step:    39700 Batch Loss:     0.660990 Tokens per Sec:    14191, Lr: 0.000050\n",
            "2019-12-31 08:51:58,040 Epoch 147: total training loss 159.34\n",
            "2019-12-31 08:51:58,040 EPOCH 148\n",
            "2019-12-31 08:52:08,862 Epoch 148 Step:    39800 Batch Loss:     0.424601 Tokens per Sec:    13635, Lr: 0.000050\n",
            "2019-12-31 08:52:23,395 Epoch 148 Step:    39900 Batch Loss:     0.631629 Tokens per Sec:    13917, Lr: 0.000050\n",
            "2019-12-31 08:52:37,440 Epoch 148: total training loss 158.73\n",
            "2019-12-31 08:52:37,440 EPOCH 149\n",
            "2019-12-31 08:52:38,070 Epoch 149 Step:    40000 Batch Loss:     0.667331 Tokens per Sec:    13908, Lr: 0.000050\n",
            "2019-12-31 08:53:11,522 Example #0\n",
            "2019-12-31 08:53:11,523 \tSource:     These orchestral arrangements are composed in such a way that they will prepare our heart and mind for the program to follow .\n",
            "2019-12-31 08:53:11,523 \tReference:  E ru uhworo nana vwẹ idjerhe ro de se muegbe rẹ ubiudu rẹ avwanre hẹrhẹ ọrhuẹrẹphiyọ rẹ ẹdẹ yena .\n",
            "2019-12-31 08:53:11,523 \tHypothesis: Ihoho nana nẹrhẹ ihwo roro nẹ ayen che muegbe rẹ ubiudu avwanre , je reyọ oborẹ avwanre yonori .\n",
            "2019-12-31 08:53:11,523 Example #1\n",
            "2019-12-31 08:53:11,523 \tSource:     Today he is serving at Bethel .\n",
            "2019-12-31 08:53:11,523 \tReference:  Nonẹna , ọ ga vwẹ Bẹtẹl .\n",
            "2019-12-31 08:53:11,523 \tHypothesis: Asaọkiephana , ọ ga vwẹ Bẹtẹl asaọkiephana .\n",
            "2019-12-31 08:53:11,523 Example #2\n",
            "2019-12-31 08:53:11,523 \tSource:     But freedom from what ?\n",
            "2019-12-31 08:53:11,523 \tReference:  Ẹkẹvuọvo , ẹdia vọ yen egbomọphẹ na che si ayen nu ?\n",
            "2019-12-31 08:53:11,524 \tHypothesis: Die kọyen egbomọphẹ ?\n",
            "2019-12-31 08:53:11,524 Example #3\n",
            "2019-12-31 08:53:11,524 \tSource:     Avoid comparing your new congregation with your previous one .\n",
            "2019-12-31 08:53:11,524 \tReference:  Wọ vwẹ ukoko kpokpọ na vwọ vwanvwen ọ rẹ wo nurhe na - a .\n",
            "2019-12-31 08:53:11,524 \tHypothesis: Ọ je dianẹ ovwan vẹ ukoko kpokpọ na ọfa vwo oka rẹ ovwan ro chekọ - a .\n",
            "2019-12-31 08:53:11,524 Validation result (greedy) at epoch 149, step    40000: bleu:  15.22, loss: 65266.7891, ppl:  22.6661, duration: 33.4538s\n",
            "2019-12-31 08:53:25,998 Epoch 149 Step:    40100 Batch Loss:     0.552748 Tokens per Sec:    14109, Lr: 0.000050\n",
            "2019-12-31 08:53:40,499 Epoch 149 Step:    40200 Batch Loss:     0.584819 Tokens per Sec:    13803, Lr: 0.000050\n",
            "2019-12-31 08:53:49,957 Epoch 149: total training loss 157.60\n",
            "2019-12-31 08:53:49,958 EPOCH 150\n",
            "2019-12-31 08:53:54,998 Epoch 150 Step:    40300 Batch Loss:     0.602546 Tokens per Sec:    15056, Lr: 0.000050\n",
            "2019-12-31 08:54:09,371 Epoch 150 Step:    40400 Batch Loss:     0.319972 Tokens per Sec:    13679, Lr: 0.000050\n",
            "2019-12-31 08:54:23,947 Epoch 150 Step:    40500 Batch Loss:     0.531706 Tokens per Sec:    13793, Lr: 0.000050\n",
            "2019-12-31 08:54:29,146 Epoch 150: total training loss 157.35\n",
            "2019-12-31 08:54:29,146 Training ended after 150 epochs.\n",
            "2019-12-31 08:54:29,146 Best validation result (greedy) at step     7000:  13.40 ppl.\n",
            "2019-12-31 08:54:48,104  dev bleu:  11.80 [Beam search decoding with beam size = 5 and alpha = 1.0]\n",
            "2019-12-31 08:54:48,104 Translations saved to: models/enurh_transformer/00007000.hyps.dev\n",
            "2019-12-31 08:55:16,292 test bleu:  22.39 [Beam search decoding with beam size = 5 and alpha = 1.0]\n",
            "2019-12-31 08:55:16,293 Translations saved to: models/enurh_transformer/00007000.hyps.test\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "MBoDS09JM807",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "ace2c1d7-6b25-4f09-9fef-c5b150787a61"
      },
      "source": [
        "# Copy the created models from the notebook storage to google drive for persistant storage \n",
        "#!cp -r joeynmt/models/${src}${tgt}_transformer/* \"$gdrive_path/models/${src}${tgt}_transformer/\"\n",
        "!cp -r joeynmt/models/${src}${tgt}_transformer/* drive/'My Drive'/masakhane/en-urh-baseline/models/enurh_transformer"
      ],
      "execution_count": 45,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "cp: cannot create symbolic link 'drive/My Drive/masakhane/en-urh-baseline/models/enurh_transformer/best.ckpt': Operation not supported\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "n94wlrCjVc17",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 697
        },
        "outputId": "0208b579-1ac9-4bdd-ad80-8ded7258c35d"
      },
      "source": [
        "# Output our validation accuracy\n",
        "! cat \"$gdrive_path/models/${src}${tgt}_transformer/validations.txt\""
      ],
      "execution_count": 46,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Steps: 1000\tLoss: 78227.96875\tPPL: 42.12487\tbleu: 0.43998\tLR: 0.00030000\t*\n",
            "Steps: 2000\tLoss: 67111.74219\tPPL: 24.75660\tbleu: 2.83394\tLR: 0.00030000\t*\n",
            "Steps: 3000\tLoss: 61367.63281\tPPL: 18.81069\tbleu: 4.77393\tLR: 0.00030000\t*\n",
            "Steps: 4000\tLoss: 57640.03906\tPPL: 15.73964\tbleu: 6.48525\tLR: 0.00030000\t*\n",
            "Steps: 5000\tLoss: 55712.81250\tPPL: 14.35399\tbleu: 8.95064\tLR: 0.00030000\t*\n",
            "Steps: 6000\tLoss: 54999.32422\tPPL: 13.87253\tbleu: 10.15556\tLR: 0.00030000\t*\n",
            "Steps: 7000\tLoss: 54282.33984\tPPL: 13.40498\tbleu: 11.26286\tLR: 0.00030000\t*\n",
            "Steps: 8000\tLoss: 54547.91406\tPPL: 13.57630\tbleu: 12.18470\tLR: 0.00030000\t\n",
            "Steps: 9000\tLoss: 54701.26953\tPPL: 13.67622\tbleu: 12.61264\tLR: 0.00030000\t\n",
            "Steps: 10000\tLoss: 55179.74609\tPPL: 13.99273\tbleu: 12.95071\tLR: 0.00030000\t\n",
            "Steps: 11000\tLoss: 55769.63672\tPPL: 14.39304\tbleu: 13.16508\tLR: 0.00030000\t\n",
            "Steps: 12000\tLoss: 56564.67188\tPPL: 14.95075\tbleu: 13.49432\tLR: 0.00030000\t\n",
            "Steps: 13000\tLoss: 56796.05859\tPPL: 15.11708\tbleu: 13.26063\tLR: 0.00021000\t\n",
            "Steps: 14000\tLoss: 57613.84375\tPPL: 15.71993\tbleu: 13.94025\tLR: 0.00021000\t\n",
            "Steps: 15000\tLoss: 58165.76562\tPPL: 16.14033\tbleu: 13.85035\tLR: 0.00021000\t\n",
            "Steps: 16000\tLoss: 58579.19922\tPPL: 16.46258\tbleu: 14.42836\tLR: 0.00021000\t\n",
            "Steps: 17000\tLoss: 59104.80469\tPPL: 16.88158\tbleu: 14.35280\tLR: 0.00021000\t\n",
            "Steps: 18000\tLoss: 59971.57812\tPPL: 17.59597\tbleu: 13.90183\tLR: 0.00021000\t\n",
            "Steps: 19000\tLoss: 60328.69141\tPPL: 17.89902\tbleu: 14.09911\tLR: 0.00014700\t\n",
            "Steps: 20000\tLoss: 60525.47266\tPPL: 18.06823\tbleu: 14.55474\tLR: 0.00014700\t\n",
            "Steps: 21000\tLoss: 61023.71094\tPPL: 18.50387\tbleu: 14.48248\tLR: 0.00014700\t\n",
            "Steps: 22000\tLoss: 61665.95703\tPPL: 19.08094\tbleu: 14.40739\tLR: 0.00014700\t\n",
            "Steps: 23000\tLoss: 61697.98047\tPPL: 19.11018\tbleu: 14.12732\tLR: 0.00014700\t\n",
            "Steps: 24000\tLoss: 62261.51953\tPPL: 19.63214\tbleu: 14.51305\tLR: 0.00014700\t\n",
            "Steps: 25000\tLoss: 62652.58594\tPPL: 20.00271\tbleu: 14.48856\tLR: 0.00010290\t\n",
            "Steps: 26000\tLoss: 62752.07422\tPPL: 20.09810\tbleu: 14.72953\tLR: 0.00010290\t\n",
            "Steps: 27000\tLoss: 62960.05859\tPPL: 20.29897\tbleu: 15.06399\tLR: 0.00010290\t\n",
            "Steps: 28000\tLoss: 63579.70703\tPPL: 20.90943\tbleu: 15.01341\tLR: 0.00010290\t\n",
            "Steps: 29000\tLoss: 63843.86719\tPPL: 21.17521\tbleu: 14.88903\tLR: 0.00010290\t\n",
            "Steps: 30000\tLoss: 63885.11719\tPPL: 21.21703\tbleu: 14.17094\tLR: 0.00010290\t\n",
            "Steps: 31000\tLoss: 64090.25391\tPPL: 21.42617\tbleu: 14.80125\tLR: 0.00007203\t\n",
            "Steps: 32000\tLoss: 64068.02734\tPPL: 21.40341\tbleu: 15.04242\tLR: 0.00007203\t\n",
            "Steps: 33000\tLoss: 64300.80859\tPPL: 21.64298\tbleu: 15.01238\tLR: 0.00007203\t\n",
            "Steps: 34000\tLoss: 64409.68359\tPPL: 21.75595\tbleu: 14.89129\tLR: 0.00007203\t\n",
            "Steps: 35000\tLoss: 64785.31250\tPPL: 22.15025\tbleu: 14.91799\tLR: 0.00007203\t\n",
            "Steps: 36000\tLoss: 64826.36719\tPPL: 22.19377\tbleu: 14.49856\tLR: 0.00007203\t\n",
            "Steps: 37000\tLoss: 65046.75000\tPPL: 22.42889\tbleu: 14.71071\tLR: 0.00005042\t\n",
            "Steps: 38000\tLoss: 65116.25000\tPPL: 22.50355\tbleu: 14.57850\tLR: 0.00005042\t\n",
            "Steps: 39000\tLoss: 65140.56641\tPPL: 22.52973\tbleu: 14.54128\tLR: 0.00005042\t\n",
            "Steps: 40000\tLoss: 65266.78906\tPPL: 22.66612\tbleu: 15.21854\tLR: 0.00005042\t\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab_type": "code",
        "id": "66WhRE9lIhoD",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 68
        },
        "outputId": "fe87bc05-5f40-41f8-95f9-8bd7d982a49d"
      },
      "source": [
        "# Test our model\n",
        "! cd joeynmt; python3 -m joeynmt test \"$gdrive_path/models/${src}${tgt}_transformer/config.yaml\""
      ],
      "execution_count": 47,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "2019-12-31 08:55:58,017 Hello! This is Joey-NMT.\n",
            "2019-12-31 08:56:19,658  dev bleu:  11.80 [Beam search decoding with beam size = 5 and alpha = 1.0]\n",
            "2019-12-31 08:56:47,548 test bleu:  22.39 [Beam search decoding with beam size = 5 and alpha = 1.0]\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}