{
  "id": 199311,
  "title": "Keep TPU Notebook alive for more than 15 minutes?",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/199311",
  "author_name": "",
  "post_date": "2020-11-25T09:53:51.229869100Z",
  "votes": 1,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hey everyone. So I've been running my training on TPU kernels, and I've really been struggling because of the 15 minute time limit to keep the notebook active. </p>\n<p>Is there any way to keep a TPU kernel active for more than 15 minutes without manual external actions? <br>\n(Like how in Colab there are certain code snippets that keep the notebook awake/alive by automatically clicking a part of the notebook periodically)</p>",
  "messages": [
    {
      "id": "1090381",
      "postDate": "11/25/2020 09:53:51",
      "content": "<p>Hey everyone. So I've been running my training on TPU kernels, and I've really been struggling because of the 15 minute time limit to keep the notebook active. </p>\n<p>Is there any way to keep a TPU kernel active for more than 15 minutes without manual external actions? <br>\n(Like how in Colab there are certain code snippets that keep the notebook awake/alive by automatically clicking a part of the notebook periodically)</p>",
      "rawMarkdown": "Hey everyone. So I've been running my training on TPU kernels, and I've really been struggling because of the 15 minute time limit to keep the notebook active. \n\nIs there any way to keep a TPU kernel active for more than 15 minutes without manual external actions? \n(Like how in Colab there are certain code snippets that keep the notebook awake/alive by automatically clicking a part of the notebook periodically)",
      "votes": null
    },
    {
      "id": "1090389",
      "postDate": "11/25/2020 09:59:12",
      "content": "<p>For longer runs you have to press \"Save and Commit\" button, after that, you will have access to all output files generated by the notebook. </p>",
      "rawMarkdown": "For longer runs you have to press \"Save and Commit\" button, after that, you will have access to all output files generated by the notebook.",
      "votes": null
    },
    {
      "id": "1090393",
      "postDate": "11/25/2020 10:01:22",
      "content": "<p><a href=\"https://www.kaggle.com/szacho\" target=\"_blank\">@szacho</a> I meant that while the model is training, how do I keep it active?</p>",
      "rawMarkdown": "szacho I meant that while the model is training, how do I keep it active?",
      "votes": null
    },
    {
      "id": "1090402",
      "postDate": "11/25/2020 10:05:51",
      "content": "<p>Why do you need to keep it active while training? If you save and commit, the model will be trained, saved and ready to use for inference</p>",
      "rawMarkdown": "Why do you need to keep it active while training? If you save and commit, the model will be trained, saved and ready to use for inference",
      "votes": null
    },
    {
      "id": "1090413",
      "postDate": "11/25/2020 10:08:50",
      "content": "<p>Oh alright. That's cool then. Thanks for confirming.  So is my TPU quota used when I commit?</p>",
      "rawMarkdown": "Oh alright. That's cool then. Thanks for confirming.  So is my TPU quota used when I commit?",
      "votes": null
    },
    {
      "id": "1090419",
      "postDate": "11/25/2020 10:13:28",
      "content": "<p>Yes, it is. And don't forget to save in the notebook files that you need (e.g. state_dict of a model), because only these will be available as output after committing</p>",
      "rawMarkdown": "Yes, it is. And don't forget to save in the notebook files that you need (e.g. state_dict of a model), because only these will be available as output after committing",
      "votes": null
    },
    {
      "id": "1090486",
      "postDate": "11/25/2020 11:15:58",
      "content": "<p>Yup got it!</p>",
      "rawMarkdown": "Yup got it!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1090389,
      "author_name": "szacho",
      "author_url": "",
      "post_date": "11/25/2020 09:59:12",
      "content": "<p>For longer runs you have to press \"Save and Commit\" button, after that, you will have access to all output files generated by the notebook. </p>",
      "votes": null,
      "replies": [
        {
          "id": 1090393,
          "author_name": "nikhiljohnk",
          "author_url": "",
          "post_date": "11/25/2020 10:01:22",
          "content": "<p><a href=\"https://www.kaggle.com/szacho\" target=\"_blank\">@szacho</a> I meant that while the model is training, how do I keep it active?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1090402,
          "author_name": "szacho",
          "author_url": "",
          "post_date": "11/25/2020 10:05:51",
          "content": "<p>Why do you need to keep it active while training? If you save and commit, the model will be trained, saved and ready to use for inference</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1090413,
          "author_name": "nikhiljohnk",
          "author_url": "",
          "post_date": "11/25/2020 10:08:50",
          "content": "<p>Oh alright. That's cool then. Thanks for confirming.  So is my TPU quota used when I commit?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1090419,
          "author_name": "szacho",
          "author_url": "",
          "post_date": "11/25/2020 10:13:28",
          "content": "<p>Yes, it is. And don't forget to save in the notebook files that you need (e.g. state_dict of a model), because only these will be available as output after committing</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1090486,
          "author_name": "nikhiljohnk",
          "author_url": "",
          "post_date": "11/25/2020 11:15:58",
          "content": "<p>Yup got it!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1090381": "Hey everyone. So I've been running my training on TPU kernels, and I've really been struggling because of the 15 minute time limit to keep the notebook active. \n\nIs there any way to keep a TPU kernel active for more than 15 minutes without manual external actions? \n(Like how in Colab there are certain code snippets that keep the notebook awake/alive by automatically clicking a part of the notebook periodically)",
    "1090389": "For longer runs you have to press \"Save and Commit\" button, after that, you will have access to all output files generated by the notebook.",
    "1090393": "szacho I meant that while the model is training, how do I keep it active?",
    "1090402": "Why do you need to keep it active while training? If you save and commit, the model will be trained, saved and ready to use for inference",
    "1090413": "Oh alright. That's cool then. Thanks for confirming.  So is my TPU quota used when I commit?",
    "1090419": "Yes, it is. And don't forget to save in the notebook files that you need (e.g. state_dict of a model), because only these will be available as output after committing",
    "1090486": "Yup got it!"
  },
  "source": "meta"
}