{
  "id": 114974,
  "title": "Why is TensorFlow 2 much slower than TensorFlow 1?",
  "url": "/competitions/tensorflow2-question-answering/discussion/114974",
  "author_name": "",
  "post_date": "2019-10-30T12:58:20.092795800Z",
  "votes": 10,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Since this contest is sponsored (kind of) by TF 2.0, I thought of bringing into attention the subject issue, especially since it seems that TF devs have not been responsive (disclaimer: not mine, I'm just the messenger):</p>\n\n<p><a href=\"https://github.com/tensorflow/tensorflow/issues/33487\">Github issue</a></p>\n\n<p><a href=\"https://stackoverflow.com/questions/58441514/why-is-tensorflow-2-much-slower-than-tensorflow-1\">Stack Overflow thread</a></p>",
  "messages": [
    {
      "id": "661545",
      "postDate": "10/30/2019 12:58:20",
      "content": "<p>Since this contest is sponsored (kind of) by TF 2.0, I thought of bringing into attention the subject issue, especially since it seems that TF devs have not been responsive (disclaimer: not mine, I'm just the messenger):</p>\n\n<p><a href=\"https://github.com/tensorflow/tensorflow/issues/33487\">Github issue</a></p>\n\n<p><a href=\"https://stackoverflow.com/questions/58441514/why-is-tensorflow-2-much-slower-than-tensorflow-1\">Stack Overflow thread</a></p>",
      "rawMarkdown": "Since this contest is sponsored (kind of) by TF 2.0, I thought of bringing into attention the subject issue, especially since it seems that TF devs have not been responsive (disclaimer: not mine, I'm just the messenger):\n\n[Github issue](https://github.com/tensorflow/tensorflow/issues/33487)\n\n[Stack Overflow thread](https://stackoverflow.com/questions/58441514/why-is-tensorflow-2-much-slower-than-tensorflow-1)",
      "votes": null
    },
    {
      "id": "661562",
      "postDate": "10/30/2019 13:09:33",
      "content": "<p>And whoever answers to this question in 2 days, receives a +50 reputation bounty at stack overflow  </p>",
      "rawMarkdown": "And whoever answers to this question in 2 days, receives a +50 reputation bounty at stack overflow",
      "votes": null
    },
    {
      "id": "662664",
      "postDate": "10/31/2019 20:32:39",
      "content": "<p>The github link has some answers now. \n1. Apparently there is a bug in train_on_batch(), test_on_batch() and predict_on_batch(), so they recommend using model.fit etc. until that is fixed.\n2. They recommend writing in default eager mode and using @tf.function rather than trying to use graph mode explicitly</p>",
      "rawMarkdown": "The github link has some answers now. \n1. Apparently there is a bug in train_on_batch(), test_on_batch() and predict_on_batch(), so they recommend using model.fit etc. until that is fixed.\n2. They recommend writing in default eager mode and using @tf.function rather than trying to use graph mode explicitly",
      "votes": null
    },
    {
      "id": "663053",
      "postDate": "11/01/2019 12:00:27",
      "content": "<p>Indeed; 2 high quality complementary answers have also been added in the Stack Overflow thread by the OP</p>",
      "rawMarkdown": "Indeed; 2 high quality complementary answers have also been added in the Stack Overflow thread by the OP",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 661562,
      "author_name": "khairulislam",
      "author_url": "",
      "post_date": "10/30/2019 13:09:33",
      "content": "<p>And whoever answers to this question in 2 days, receives a +50 reputation bounty at stack overflow  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 662664,
      "author_name": "aharless",
      "author_url": "",
      "post_date": "10/31/2019 20:32:39",
      "content": "<p>The github link has some answers now. \n1. Apparently there is a bug in train_on_batch(), test_on_batch() and predict_on_batch(), so they recommend using model.fit etc. until that is fixed.\n2. They recommend writing in default eager mode and using @tf.function rather than trying to use graph mode explicitly</p>",
      "votes": null,
      "replies": [
        {
          "id": 663053,
          "author_name": "ctsats",
          "author_url": "",
          "post_date": "11/01/2019 12:00:27",
          "content": "<p>Indeed; 2 high quality complementary answers have also been added in the Stack Overflow thread by the OP</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "661545": "Since this contest is sponsored (kind of) by TF 2.0, I thought of bringing into attention the subject issue, especially since it seems that TF devs have not been responsive (disclaimer: not mine, I'm just the messenger):\n\n[Github issue](https://github.com/tensorflow/tensorflow/issues/33487)\n\n[Stack Overflow thread](https://stackoverflow.com/questions/58441514/why-is-tensorflow-2-much-slower-than-tensorflow-1)",
    "661562": "And whoever answers to this question in 2 days, receives a +50 reputation bounty at stack overflow",
    "662664": "The github link has some answers now. \n1. Apparently there is a bug in train_on_batch(), test_on_batch() and predict_on_batch(), so they recommend using model.fit etc. until that is fixed.\n2. They recommend writing in default eager mode and using @tf.function rather than trying to use graph mode explicitly",
    "663053": "Indeed; 2 high quality complementary answers have also been added in the Stack Overflow thread by the OP"
  },
  "source": "meta"
}