{
  "id": 190078,
  "title": "Which device does you guys use?",
  "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/190078",
  "author_name": "Dewei Chen",
  "post_date": "2020-10-10T04:10:03.343000",
  "votes": -1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>The original dataset takes nearly 1T space. It's hard for me to download it to my laptop. I buy a new disk for my desktop, and I am still trying to download this 1T dataset(It cost really long time) to my disk.</p>\n<p>I saw there is tfrecord file, and someone transfer dataset to JPEC from.<br>\nFor me, I just have 40h GPU per week.</p>\n<p>So, what I want to ask is: </p>\n<ol>\n<li>Did you training your model on your local PC? If did, which type of dataset did you used?</li>\n<li>How much GPU/TPU running time did you cost per week?</li>\n</ol>\n<p>Thanks, for all answers.</p>",
  "messages": [
    {
      "id": 1044765,
      "postDate": "2020-10-10T05:13:33.847Z",
      "content": "<p>I use only Colab TPUs and Kaggle kernel TPUs for training, my local machine is horrible. It took ~33, 34 hrs of GPU quota to run my models on Kaggle kernels.</p>",
      "rawMarkdown": "I use only Colab TPUs and Kaggle kernel TPUs for training, my local machine is horrible. It took ~33, 34 hrs of GPU quota to run my models on Kaggle kernels.",
      "votes": 2,
      "replies": [
        {
          "id": 1044778,
          "postDate": "2020-10-10T05:28:35.483Z",
          "content": "<p>That's a pretty good news for my. Thanks for sharing.</p>",
          "rawMarkdown": "That's a pretty good news for my. Thanks for sharing.",
          "votes": 1
        }
      ]
    },
    {
      "id": 1044704,
      "postDate": "2020-10-10T04:10:03.343Z",
      "content": "<p>The original dataset takes nearly 1T space. It's hard for me to download it to my laptop. I buy a new disk for my desktop, and I am still trying to download this 1T dataset(It cost really long time) to my disk.</p>\n<p>I saw there is tfrecord file, and someone transfer dataset to JPEC from.<br>\nFor me, I just have 40h GPU per week.</p>\n<p>So, what I want to ask is: </p>\n<ol>\n<li>Did you training your model on your local PC? If did, which type of dataset did you used?</li>\n<li>How much GPU/TPU running time did you cost per week?</li>\n</ol>\n<p>Thanks, for all answers.</p>",
      "rawMarkdown": "The original dataset takes nearly 1T space. It's hard for me to download it to my laptop. I buy a new disk for my desktop, and I am still trying to download this 1T dataset(It cost really long time) to my disk.\n\nI saw there is tfrecord file, and someone transfer dataset to JPEC from.\nFor me, I just have 40h GPU per week.\n\nSo, what I want to ask is: \n1. Did you training your model on your local PC? If did, which type of dataset did you used?\n2. How much GPU/TPU running time did you cost per week?\n\nThanks, for all answers.",
      "votes": -1
    }
  ],
  "comments": [
    {
      "id": 1044765,
      "author_name": "Trigram",
      "author_url": "",
      "post_date": "2020-10-10T05:13:33.847000",
      "content": "<p>I use only Colab TPUs and Kaggle kernel TPUs for training, my local machine is horrible. It took ~33, 34 hrs of GPU quota to run my models on Kaggle kernels.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 1044778,
          "author_name": "Dewei Chen",
          "author_url": "",
          "post_date": "2020-10-10T05:28:35.483000",
          "content": "<p>That's a pretty good news for my. Thanks for sharing.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1044765": "I use only Colab TPUs and Kaggle kernel TPUs for training, my local machine is horrible. It took ~33, 34 hrs of GPU quota to run my models on Kaggle kernels.",
    "1044704": "The original dataset takes nearly 1T space. It's hard for me to download it to my laptop. I buy a new disk for my desktop, and I am still trying to download this 1T dataset(It cost really long time) to my disk.\n\nI saw there is tfrecord file, and someone transfer dataset to JPEC from.\nFor me, I just have 40h GPU per week.\n\nSo, what I want to ask is: \n1. Did you training your model on your local PC? If did, which type of dataset did you used?\n2. How much GPU/TPU running time did you cost per week?\n\nThanks, for all answers."
  }
}