{
  "id": 162211,
  "title": "[PyTorch] Training on GPU takes too long",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/162211",
  "author_name": "",
  "post_date": "2020-06-27T20:16:38.440036900Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hello,</p>\n\n<p>Im training on EF_B1 with an image-size of 240x240 and a batch-size of 32 and one epoch of training takes 1:30 hours its incredible long. I have seen notebooks where people need only 5 minutes. I dont know why. I would like to share my notebook with someone maybe you can spot the mistake.</p>\n\n<p>Thank you in advance.</p>",
  "messages": [
    {
      "id": "904691",
      "postDate": "06/27/2020 20:16:38",
      "content": "<p>Hello,</p>\n\n<p>Im training on EF_B1 with an image-size of 240x240 and a batch-size of 32 and one epoch of training takes 1:30 hours its incredible long. I have seen notebooks where people need only 5 minutes. I dont know why. I would like to share my notebook with someone maybe you can spot the mistake.</p>\n\n<p>Thank you in advance.</p>",
      "rawMarkdown": "Hello,\n\nIm training on EF_B1 with an image-size of 240x240 and a batch-size of 32 and one epoch of training takes 1:30 hours its incredible long. I have seen notebooks where people need only 5 minutes. I dont know why. I would like to share my notebook with someone maybe you can spot the mistake.\n\nThank you in advance.",
      "votes": null
    },
    {
      "id": "905743",
      "postDate": "06/28/2020 18:51:56",
      "content": "<p>I fixed my problem, the image-sizes are way to big and my train-dataloader took way too long to get a batch of 32 images (20 seconds) Im using size reduced training data now from here: <a href=\"https://www.kaggle.com/nroman/melanoma-external-malignant-256\">https://www.kaggle.com/nroman/melanoma-external-malignant-256</a> this causes my dataloader to load 32 images within a second and fixed my whole train problem. </p>\n\n<p>One epoch takes around 3 minutes now on GPU</p>",
      "rawMarkdown": "I fixed my problem, the image-sizes are way to big and my train-dataloader took way too long to get a batch of 32 images (20 seconds) Im using size reduced training data now from here: https://www.kaggle.com/nroman/melanoma-external-malignant-256 this causes my dataloader to load 32 images within a second and fixed my whole train problem. \n\nOne epoch takes around 3 minutes now on GPU",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 905743,
      "author_name": "aliabdin1",
      "author_url": "",
      "post_date": "06/28/2020 18:51:56",
      "content": "<p>I fixed my problem, the image-sizes are way to big and my train-dataloader took way too long to get a batch of 32 images (20 seconds) Im using size reduced training data now from here: <a href=\"https://www.kaggle.com/nroman/melanoma-external-malignant-256\">https://www.kaggle.com/nroman/melanoma-external-malignant-256</a> this causes my dataloader to load 32 images within a second and fixed my whole train problem. </p>\n\n<p>One epoch takes around 3 minutes now on GPU</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "904691": "Hello,\n\nIm training on EF_B1 with an image-size of 240x240 and a batch-size of 32 and one epoch of training takes 1:30 hours its incredible long. I have seen notebooks where people need only 5 minutes. I dont know why. I would like to share my notebook with someone maybe you can spot the mistake.\n\nThank you in advance.",
    "905743": "I fixed my problem, the image-sizes are way to big and my train-dataloader took way too long to get a batch of 32 images (20 seconds) Im using size reduced training data now from here: https://www.kaggle.com/nroman/melanoma-external-malignant-256 this causes my dataloader to load 32 images within a second and fixed my whole train problem. \n\nOne epoch takes around 3 minutes now on GPU"
  },
  "source": "meta"
}