{
  "id": 116681,
  "title": "Colab now has P100 GPUs",
  "url": "/competitions/understanding_cloud_organization/discussion/116681",
  "author_name": "",
  "post_date": "2019-11-10T19:18:54.572693500Z",
  "votes": 15,
  "comment_count": 8,
  "views": 0,
  "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1173254%2F0544ec5e10d82fb3e6aeee2aed6173af%2FColab.png?generation=1573413526850411&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": "669960",
      "postDate": "11/10/2019 19:18:54",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1173254%2F0544ec5e10d82fb3e6aeee2aed6173af%2FColab.png?generation=1573413526850411&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1173254%2F0544ec5e10d82fb3e6aeee2aed6173af%2FColab.png?generation=1573413526850411&amp;alt=media)",
      "votes": null
    },
    {
      "id": "669961",
      "postDate": "11/10/2019 19:20:51",
      "content": "<p>great news</p>",
      "rawMarkdown": "great news",
      "votes": null
    },
    {
      "id": "670038",
      "postDate": "11/10/2019 23:26:35",
      "content": "<p>Awesome!!!</p>",
      "rawMarkdown": "Awesome!!!",
      "votes": null
    },
    {
      "id": "670116",
      "postDate": "11/11/2019 03:07:46",
      "content": "<p>Awesome , and  what's  the  time  limit?</p>",
      "rawMarkdown": "Awesome , and  what's  the  time  limit?",
      "votes": null
    },
    {
      "id": "670187",
      "postDate": "11/11/2019 06:38:09",
      "content": "<p>I guess it would be same as before, i.e, 12 hours, but when one uses Colab much, they will be provided slower GPUs(like Tesla T4 or K80). </p>",
      "rawMarkdown": "I guess it would be same as before, i.e, 12 hours, but when one uses Colab much, they will be provided slower GPUs(like Tesla T4 or K80).",
      "votes": null
    },
    {
      "id": "670897",
      "postDate": "11/12/2019 02:33:38",
      "content": "<p>I hope google allocate them back to kaggle!!!</p>",
      "rawMarkdown": "I hope google allocate them back to kaggle!!!",
      "votes": null
    },
    {
      "id": "671622",
      "postDate": "11/13/2019 00:42:31",
      "content": "<p>Thanks for bringing this to our attention. I created a \"How To Use Colab\" discussion <a href=\"https://www.kaggle.com/c/understanding_cloud_organization/discussion/117021\">here</a>. The main thing is just transferring the Cloud Comp data to Google Colab. Instead of wasting Colab time resizing original images, I created a dataset <a href=\"https://www.kaggle.com/cdeotte/cloud-images-resized\">here</a> with all the images resized to six different sizes: <code>384x576</code>, <code>320x480</code>, <code>256x384</code>, <code>192x288</code>, <code>128x192</code>, and <code>64x96</code>. </p>\n\n<p>If you want to sit and watch your experiments, use small image sizes and your epochs can be under 1 minute each! I also posted a Keras metric which displays your Kaggle Dice score including post process <a href=\"https://www.kaggle.com/c/understanding_cloud_organization/discussion/117021\">here</a>. So you can watch your actual LB score as your models train.</p>",
      "rawMarkdown": "Thanks for bringing this to our attention. I created a \"How To Use Colab\" discussion [here][1]. The main thing is just transferring the Cloud Comp data to Google Colab. Instead of wasting Colab time resizing original images, I created a dataset [here][2] with all the images resized to six different sizes: `384x576`, `320x480`, `256x384`, `192x288`, `128x192`, and `64x96`. \n  \nIf you want to sit and watch your experiments, use small image sizes and your epochs can be under 1 minute each! I also posted a Keras metric which displays your Kaggle Dice score including post process [here][1]. So you can watch your actual LB score as your models train.\n\n[1]: https://www.kaggle.com/c/understanding_cloud_organization/discussion/117021\n[2]: https://www.kaggle.com/cdeotte/cloud-images-resized",
      "votes": null
    },
    {
      "id": "672115",
      "postDate": "11/13/2019 14:21:58",
      "content": "<p>Nice!</p>",
      "rawMarkdown": "Nice!",
      "votes": null
    },
    {
      "id": "683163",
      "postDate": "11/28/2019 06:42:34",
      "content": "<p>thats great</p>",
      "rawMarkdown": "thats great",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 669961,
      "author_name": "mobassir",
      "author_url": "",
      "post_date": "11/10/2019 19:20:51",
      "content": "<p>great news</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 670038,
      "author_name": "phunghieu",
      "author_url": "",
      "post_date": "11/10/2019 23:26:35",
      "content": "<p>Awesome!!!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 670116,
      "author_name": "xujingzhao",
      "author_url": "",
      "post_date": "11/11/2019 03:07:46",
      "content": "<p>Awesome , and  what's  the  time  limit?</p>",
      "votes": null,
      "replies": [
        {
          "id": 670187,
          "author_name": "salilm23",
          "author_url": "",
          "post_date": "11/11/2019 06:38:09",
          "content": "<p>I guess it would be same as before, i.e, 12 hours, but when one uses Colab much, they will be provided slower GPUs(like Tesla T4 or K80). </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 670897,
      "author_name": "bamps53",
      "author_url": "",
      "post_date": "11/12/2019 02:33:38",
      "content": "<p>I hope google allocate them back to kaggle!!!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 671622,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "11/13/2019 00:42:31",
      "content": "<p>Thanks for bringing this to our attention. I created a \"How To Use Colab\" discussion <a href=\"https://www.kaggle.com/c/understanding_cloud_organization/discussion/117021\">here</a>. The main thing is just transferring the Cloud Comp data to Google Colab. Instead of wasting Colab time resizing original images, I created a dataset <a href=\"https://www.kaggle.com/cdeotte/cloud-images-resized\">here</a> with all the images resized to six different sizes: <code>384x576</code>, <code>320x480</code>, <code>256x384</code>, <code>192x288</code>, <code>128x192</code>, and <code>64x96</code>. </p>\n\n<p>If you want to sit and watch your experiments, use small image sizes and your epochs can be under 1 minute each! I also posted a Keras metric which displays your Kaggle Dice score including post process <a href=\"https://www.kaggle.com/c/understanding_cloud_organization/discussion/117021\">here</a>. So you can watch your actual LB score as your models train.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 672115,
      "author_name": "honihitak",
      "author_url": "",
      "post_date": "11/13/2019 14:21:58",
      "content": "<p>Nice!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 683163,
      "author_name": "sujitnoronha",
      "author_url": "",
      "post_date": "11/28/2019 06:42:34",
      "content": "<p>thats great</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "669960": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1173254%2F0544ec5e10d82fb3e6aeee2aed6173af%2FColab.png?generation=1573413526850411&amp;alt=media)",
    "669961": "great news",
    "670038": "Awesome!!!",
    "670116": "Awesome , and  what's  the  time  limit?",
    "670187": "I guess it would be same as before, i.e, 12 hours, but when one uses Colab much, they will be provided slower GPUs(like Tesla T4 or K80).",
    "670897": "I hope google allocate them back to kaggle!!!",
    "671622": "Thanks for bringing this to our attention. I created a \"How To Use Colab\" discussion [here][1]. The main thing is just transferring the Cloud Comp data to Google Colab. Instead of wasting Colab time resizing original images, I created a dataset [here][2] with all the images resized to six different sizes: `384x576`, `320x480`, `256x384`, `192x288`, `128x192`, and `64x96`. \n  \nIf you want to sit and watch your experiments, use small image sizes and your epochs can be under 1 minute each! I also posted a Keras metric which displays your Kaggle Dice score including post process [here][1]. So you can watch your actual LB score as your models train.\n\n[1]: https://www.kaggle.com/c/understanding_cloud_organization/discussion/117021\n[2]: https://www.kaggle.com/cdeotte/cloud-images-resized",
    "672115": "Nice!",
    "683163": "thats great"
  },
  "source": "meta"
}