{
  "id": 172593,
  "title": "Detailed guide to custom training with TPUs",
  "url": "/competitions/tpu-getting-started/discussion/172593",
  "author_name": "Yih-Dar SHIEH",
  "post_date": "2020-08-05T17:36:33.203000",
  "votes": 3,
  "comment_count": 1,
  "views": null,
  "content": "<p>Hi,</p>\n\n<p>I published a new notebook  <a href=\"https://www.kaggle.com/yihdarshieh/detailed-guide-to-custom-training-with-tpus\">Detailed guide to custom training with TPUs</a></p>\n\n<p>It is for people who want to learn how to use TPUs with a bit more details through a presentation of custom training with TPUs. There is no <code>model.fit</code> in this notebook.</p>\n\n<p>It combines several of my previous  notebooks, including </p>\n\n<ul>\n<li>custom training</li>\n<li>gradient accumulation</li>\n<li>oversampling</li>\n<li>perspective transformation (as data augmentation)</li>\n</ul>\n\n<p>with more explanation on some TPU stuffs.</p>\n\n<p>I hope it could be helpful for some people.</p>",
  "messages": [
    {
      "id": 959561,
      "postDate": "2020-08-05T17:36:33.203Z",
      "content": "<p>Hi,</p>\n\n<p>I published a new notebook  <a href=\"https://www.kaggle.com/yihdarshieh/detailed-guide-to-custom-training-with-tpus\">Detailed guide to custom training with TPUs</a></p>\n\n<p>It is for people who want to learn how to use TPUs with a bit more details through a presentation of custom training with TPUs. There is no <code>model.fit</code> in this notebook.</p>\n\n<p>It combines several of my previous  notebooks, including </p>\n\n<ul>\n<li>custom training</li>\n<li>gradient accumulation</li>\n<li>oversampling</li>\n<li>perspective transformation (as data augmentation)</li>\n</ul>\n\n<p>with more explanation on some TPU stuffs.</p>\n\n<p>I hope it could be helpful for some people.</p>",
      "rawMarkdown": "Hi,\n\nI published a new notebook  [Detailed guide to custom training with TPUs](https://www.kaggle.com/yihdarshieh/detailed-guide-to-custom-training-with-tpus)\n\nIt is for people who want to learn how to use TPUs with a bit more details through a presentation of custom training with TPUs. There is no `model.fit` in this notebook.\n\nIt combines several of my previous  notebooks, including \n\n- custom training\n- gradient accumulation\n- oversampling\n- perspective transformation (as data augmentation)\n\nwith more explanation on some TPU stuffs.\n\nI hope it could be helpful for some people.",
      "votes": 3
    },
    {
      "id": 1001339,
      "postDate": "2020-09-07T08:49:08.343Z",
      "content": "<p>Thank you, it helps a lot</p>",
      "rawMarkdown": "Thank you, it helps a lot"
    }
  ],
  "comments": [
    {
      "id": 1001339,
      "author_name": "Hanq",
      "author_url": "",
      "post_date": "2020-09-07T08:49:08.343000",
      "content": "<p>Thank you, it helps a lot</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "959561": "Hi,\n\nI published a new notebook  [Detailed guide to custom training with TPUs](https://www.kaggle.com/yihdarshieh/detailed-guide-to-custom-training-with-tpus)\n\nIt is for people who want to learn how to use TPUs with a bit more details through a presentation of custom training with TPUs. There is no `model.fit` in this notebook.\n\nIt combines several of my previous  notebooks, including \n\n- custom training\n- gradient accumulation\n- oversampling\n- perspective transformation (as data augmentation)\n\nwith more explanation on some TPU stuffs.\n\nI hope it could be helpful for some people.",
    "1001339": "Thank you, it helps a lot"
  }
}