{
  "id": 228884,
  "title": "Why the submission is so slow?",
  "url": "/competitions/plant-pathology-2021-fgvc8/discussion/228884",
  "author_name": "",
  "post_date": "2021-03-27T01:58:53.201530600Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>It takes me about 45min to train and generate the submission.csv for the given 3 test images, but after I submit my notebook, I waited for about 10 hours and the score haven't generated</p>",
  "messages": [
    {
      "id": "1253748",
      "postDate": "03/27/2021 01:58:53",
      "content": "<p>It takes me about 45min to train and generate the submission.csv for the given 3 test images, but after I submit my notebook, I waited for about 10 hours and the score haven't generated</p>",
      "rawMarkdown": "It takes me about 45min to train and generate the submission.csv for the given 3 test images, but after I submit my notebook, I waited for about 10 hours and the score haven't generated",
      "votes": null
    },
    {
      "id": "1254210",
      "postDate": "03/27/2021 12:39:31",
      "content": "<p>Hello!</p>\n<p>The thing is when you're submitting a notebook, it runs over again and makes predictions for ~5,000 images. If your notebook takes 45 minutes to make predictions for only 3 images it would definitely run out of 2 hours limit (for GPU) and probably out of 9 hours limit (for CPU). It's hard to give any advice unless you provide more information (e.g. what backbone and image size you're using, how do you load images, etc.)</p>\n<p>If you're a Keras user, consider using <strong><a href=\"https://www.kaggle.com/nickuzmenkov/pp2021-tpu-tf-inference\" target=\"_blank\">this notebook</a></strong> as a template (submission with 5 EfficientNetB4 and 512x512 images takes just 14 minutes).</p>",
      "rawMarkdown": "Hello!\n\nThe thing is when you're submitting a notebook, it runs over again and makes predictions for ~5,000 images. If your notebook takes 45 minutes to make predictions for only 3 images it would definitely run out of 2 hours limit (for GPU) and probably out of 9 hours limit (for CPU). It's hard to give any advice unless you provide more information (e.g. what backbone and image size you're using, how do you load images, etc.)\n\nIf you're a Keras user, consider using **[this notebook](https://www.kaggle.com/nickuzmenkov/pp2021-tpu-tf-inference)** as a template (submission with 5 EfficientNetB4 and 512x512 images takes just 14 minutes).",
      "votes": null
    },
    {
      "id": "1254783",
      "postDate": "03/28/2021 06:02:42",
      "content": "<p>If it took more than 10hrs, I think it is a submission error.</p>",
      "rawMarkdown": "If it took more than 10hrs, I think it is a submission error.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1254210,
      "author_name": "nickuzmenkov",
      "author_url": "",
      "post_date": "03/27/2021 12:39:31",
      "content": "<p>Hello!</p>\n<p>The thing is when you're submitting a notebook, it runs over again and makes predictions for ~5,000 images. If your notebook takes 45 minutes to make predictions for only 3 images it would definitely run out of 2 hours limit (for GPU) and probably out of 9 hours limit (for CPU). It's hard to give any advice unless you provide more information (e.g. what backbone and image size you're using, how do you load images, etc.)</p>\n<p>If you're a Keras user, consider using <strong><a href=\"https://www.kaggle.com/nickuzmenkov/pp2021-tpu-tf-inference\" target=\"_blank\">this notebook</a></strong> as a template (submission with 5 EfficientNetB4 and 512x512 images takes just 14 minutes).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1254783,
      "author_name": "shanmukh05",
      "author_url": "",
      "post_date": "03/28/2021 06:02:42",
      "content": "<p>If it took more than 10hrs, I think it is a submission error.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1253748": "It takes me about 45min to train and generate the submission.csv for the given 3 test images, but after I submit my notebook, I waited for about 10 hours and the score haven't generated",
    "1254210": "Hello!\n\nThe thing is when you're submitting a notebook, it runs over again and makes predictions for ~5,000 images. If your notebook takes 45 minutes to make predictions for only 3 images it would definitely run out of 2 hours limit (for GPU) and probably out of 9 hours limit (for CPU). It's hard to give any advice unless you provide more information (e.g. what backbone and image size you're using, how do you load images, etc.)\n\nIf you're a Keras user, consider using **[this notebook](https://www.kaggle.com/nickuzmenkov/pp2021-tpu-tf-inference)** as a template (submission with 5 EfficientNetB4 and 512x512 images takes just 14 minutes).",
    "1254783": "If it took more than 10hrs, I think it is a submission error."
  },
  "source": "meta"
}