{
  "id": 399020,
  "title": "Code Requirements: Does *everything* needs to run <= 9 hours ?",
  "url": "/competitions/vesuvius-challenge-ink-detection/discussion/399020",
  "author_name": "Nils",
  "post_date": "2023-04-02T03:50:18.855000",
  "votes": 4,
  "comment_count": 9,
  "views": 0,
  "content": "<p>Does all the tasks need to happen under 9 hours, or just the evaluation part is fine?</p>\n<p>As in: </p>\n<ul>\n<li>We build a model and make everything public (code, docs, …), then load into an evaluation notebook, or</li>\n<li>Everything we do to the data, training steps, and inference needs to happen under 9 hours?</li>\n</ul>\n<p>Thanks</p>",
  "messages": [
    {
      "id": 2205880,
      "postDate": "2023-04-02T03:50:18.857Z",
      "content": "<p>Does all the tasks need to happen under 9 hours, or just the evaluation part is fine?</p>\n<p>As in: </p>\n<ul>\n<li>We build a model and make everything public (code, docs, …), then load into an evaluation notebook, or</li>\n<li>Everything we do to the data, training steps, and inference needs to happen under 9 hours?</li>\n</ul>\n<p>Thanks</p>",
      "rawMarkdown": "Does all the tasks need to happen under 9 hours, or just the evaluation part is fine?\n\nAs in: \n\n- We build a model and make everything public (code, docs, ...), then load into an evaluation notebook, or\n- Everything we do to the data, training steps, and inference needs to happen under 9 hours?\n\nThanks",
      "votes": 4
    },
    {
      "id": 2205883,
      "postDate": "2023-04-02T04:00:42.617Z",
      "content": "<p>From the answer to a question I previously asked, it's only the former. Training can be offline and the model loaded into the notebook through a dataset.</p>",
      "rawMarkdown": "From the answer to a question I previously asked, it's only the former. Training can be offline and the model loaded into the notebook through a dataset.",
      "votes": 2,
      "replies": [
        {
          "id": 2205884,
          "postDate": "2023-04-02T04:05:42.290Z",
          "content": "<p>Yep exactly this (ty!)</p>",
          "rawMarkdown": "Yep exactly this (ty!)",
          "votes": 1,
          "replies": [
            {
              "id": 2211371,
              "postDate": "2023-04-06T01:53:46.023Z",
              "content": "<p><a href=\"https://www.kaggle.com/jpposma\" target=\"_blank\">@jpposma</a> Does it mean that I can use the offline training results and only submit a notebook with the pre-trained model and evaluation code on the test set? And the GPU run time of this notebook &lt;= 9 hours?</p>",
              "rawMarkdown": "@jpposma Does it mean that I can use the offline training results and only submit a notebook with the pre-trained model and evaluation code on the test set? And the GPU run time of this notebook <= 9 hours?",
              "votes": 1
            }
          ]
        },
        {
          "id": 2206838,
          "postDate": "2023-04-02T23:45:58.050Z",
          "content": "<p>Thank you so much!</p>",
          "rawMarkdown": "Thank you so much!",
          "replies": [
            {
              "id": 2211370,
              "postDate": "2023-04-06T01:53:14.460Z",
              "rawMarkdown": "",
              "isDeleted": true
            }
          ]
        },
        {
          "id": 2211846,
          "postDate": "2023-04-06T10:12:45.110Z",
          "content": "<p><a href=\"https://www.kaggle.com/costella\" target=\"_blank\">@costella</a> <a href=\"https://www.kaggle.com/jpposma\" target=\"_blank\">@jpposma</a> do you know whether the datasets need to be public or can they be private?</p>",
          "rawMarkdown": "@costella @jpposma do you know whether the datasets need to be public or can they be private?",
          "replies": [
            {
              "id": 2211912,
              "postDate": "2023-04-06T11:00:02.067Z",
              "content": "<p>I assume they can be private.</p>",
              "rawMarkdown": "I assume they can be private.",
              "votes": 2
            },
            {
              "id": 2212211,
              "postDate": "2023-04-06T15:40:47.140Z",
              "content": "<p>Yes, <a href=\"https://www.kaggle.com/costella\" target=\"_blank\">@costella</a> is right, they can be private.</p>",
              "rawMarkdown": "Yes, @costella is right, they can be private."
            }
          ]
        }
      ]
    },
    {
      "id": 2205997,
      "postDate": "2023-04-02T06:39:14.083Z",
      "content": "<p>There is a run length limit on kaggle - cannot expect to run a training notebook for a week.  I don't run much on kaggle but I think it times out after 12 hours for cpu and GPU notebooks and 9 hours for TPU.</p>\n<p>There is also the quota's on GPU's.  </p>\n<p>So a practical limit does exist - but only the final submission piece needs to clock in under 9 hours.</p>",
      "rawMarkdown": "There is a run length limit on kaggle - cannot expect to run a training notebook for a week.  I don't run much on kaggle but I think it times out after 12 hours for cpu and GPU notebooks and 9 hours for TPU.\n\nThere is also the quota's on GPU's.  \n\nSo a practical limit does exist - but only the final submission piece needs to clock in under 9 hours.\n\n\n\n"
    }
  ],
  "comments": [
    {
      "id": 2205883,
      "author_name": "John Costella",
      "author_url": "",
      "post_date": "2023-04-02T04:00:42.617000",
      "content": "<p>From the answer to a question I previously asked, it's only the former. Training can be offline and the model loaded into the notebook through a dataset.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 2205884,
          "author_name": "JP Posma",
          "author_url": "",
          "post_date": "2023-04-02T04:05:42.290000",
          "content": "<p>Yep exactly this (ty!)</p>",
          "votes": 1,
          "replies": [
            {
              "id": 2211371,
              "author_name": "TwinB",
              "author_url": "",
              "post_date": "2023-04-06T01:53:46.023000",
              "content": "<p><a href=\"https://www.kaggle.com/jpposma\" target=\"_blank\">@jpposma</a> Does it mean that I can use the offline training results and only submit a notebook with the pre-trained model and evaluation code on the test set? And the GPU run time of this notebook &lt;= 9 hours?</p>",
              "votes": 1,
              "replies": []
            }
          ]
        },
        {
          "id": 2206838,
          "author_name": "Nils",
          "author_url": "",
          "post_date": "2023-04-02T23:45:58.050000",
          "content": "<p>Thank you so much!</p>",
          "votes": 0,
          "replies": [
            {
              "id": 2211370,
              "author_name": "",
              "author_url": "",
              "post_date": "2023-04-06T01:53:14.460000",
              "content": "",
              "votes": 0,
              "replies": []
            }
          ]
        },
        {
          "id": 2211846,
          "author_name": "Lucas",
          "author_url": "",
          "post_date": "2023-04-06T10:12:45.110000",
          "content": "<p><a href=\"https://www.kaggle.com/costella\" target=\"_blank\">@costella</a> <a href=\"https://www.kaggle.com/jpposma\" target=\"_blank\">@jpposma</a> do you know whether the datasets need to be public or can they be private?</p>",
          "votes": 0,
          "replies": [
            {
              "id": 2211912,
              "author_name": "John Costella",
              "author_url": "",
              "post_date": "2023-04-06T11:00:02.067000",
              "content": "<p>I assume they can be private.</p>",
              "votes": 2,
              "replies": []
            },
            {
              "id": 2212211,
              "author_name": "JP Posma",
              "author_url": "",
              "post_date": "2023-04-06T15:40:47.140000",
              "content": "<p>Yes, <a href=\"https://www.kaggle.com/costella\" target=\"_blank\">@costella</a> is right, they can be private.</p>",
              "votes": 0,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2205997,
      "author_name": "PC Jimmmy",
      "author_url": "",
      "post_date": "2023-04-02T06:39:14.083000",
      "content": "<p>There is a run length limit on kaggle - cannot expect to run a training notebook for a week.  I don't run much on kaggle but I think it times out after 12 hours for cpu and GPU notebooks and 9 hours for TPU.</p>\n<p>There is also the quota's on GPU's.  </p>\n<p>So a practical limit does exist - but only the final submission piece needs to clock in under 9 hours.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2205880": "Does all the tasks need to happen under 9 hours, or just the evaluation part is fine?\n\nAs in: \n\n- We build a model and make everything public (code, docs, ...), then load into an evaluation notebook, or\n- Everything we do to the data, training steps, and inference needs to happen under 9 hours?\n\nThanks",
    "2205883": "From the answer to a question I previously asked, it's only the former. Training can be offline and the model loaded into the notebook through a dataset.",
    "2205997": "There is a run length limit on kaggle - cannot expect to run a training notebook for a week.  I don't run much on kaggle but I think it times out after 12 hours for cpu and GPU notebooks and 9 hours for TPU.\n\nThere is also the quota's on GPU's.  \n\nSo a practical limit does exist - but only the final submission piece needs to clock in under 9 hours.\n\n\n\n"
  }
}