{
  "id": 187474,
  "title": "Distributed Training for Large Datasets with TPUs",
  "url": "/competitions/rsna-str-pulmonary-embolism-detection/discussion/187474",
  "author_name": "Marcos Novaes",
  "post_date": "2020-09-29T05:16:21.456000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hello, Everyone.</p>\n<p>I managed to implement distributed training on TPUs on a large dataset. I also built a sample dataset in TFRecord format that chunks the original dataset into chunks suitable for dynamic loading using tf.dataset(). I shared my experiences in this notebook:<br>\n<a href=\"https://www.kaggle.com/marcosnovaes/distributed-training-for-large-datasets-with-tpus/\" target=\"_blank\">https://www.kaggle.com/marcosnovaes/distributed-training-for-large-datasets-with-tpus/</a><br>\nI am still working on it, future versions will have more comments and references. <br>\nCheers!</p>",
  "messages": [
    {
      "id": 1030921,
      "postDate": "2020-09-29T05:16:21.457Z",
      "content": "<p>Hello, Everyone.</p>\n<p>I managed to implement distributed training on TPUs on a large dataset. I also built a sample dataset in TFRecord format that chunks the original dataset into chunks suitable for dynamic loading using tf.dataset(). I shared my experiences in this notebook:<br>\n<a href=\"https://www.kaggle.com/marcosnovaes/distributed-training-for-large-datasets-with-tpus/\" target=\"_blank\">https://www.kaggle.com/marcosnovaes/distributed-training-for-large-datasets-with-tpus/</a><br>\nI am still working on it, future versions will have more comments and references. <br>\nCheers!</p>",
      "rawMarkdown": "Hello, Everyone.\n\nI managed to implement distributed training on TPUs on a large dataset. I also built a sample dataset in TFRecord format that chunks the original dataset into chunks suitable for dynamic loading using tf.dataset(). I shared my experiences in this notebook:\n[https://www.kaggle.com/marcosnovaes/distributed-training-for-large-datasets-with-tpus/](https://www.kaggle.com/marcosnovaes/distributed-training-for-large-datasets-with-tpus/)\nI am still working on it, future versions will have more comments and references. \nCheers!"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1030921": "Hello, Everyone.\n\nI managed to implement distributed training on TPUs on a large dataset. I also built a sample dataset in TFRecord format that chunks the original dataset into chunks suitable for dynamic loading using tf.dataset(). I shared my experiences in this notebook:\n[https://www.kaggle.com/marcosnovaes/distributed-training-for-large-datasets-with-tpus/](https://www.kaggle.com/marcosnovaes/distributed-training-for-large-datasets-with-tpus/)\nI am still working on it, future versions will have more comments and references. \nCheers!"
  }
}