{
  "id": 148788,
  "title": "tpu time commit",
  "url": "/competitions/jigsaw-multilingual-toxic-comment-classification/discussion/148788",
  "author_name": "",
  "post_date": "2020-05-05T16:33:04.839540Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi guys,</p>\n\n<p>I am new to TPU use on Kaggle. The transformer NLP models take a long time to train on TPU. I am making modifications  either architecture the model or processing of dataset and to see the effect I am fine-tuning a pre-trained model, which is demanding for TPU time as you know. Then when I create a submission, I need to commit the notebook I modified, which runs everything again also takes time. Moreover, as you know, TPU time is limited, so what can I do to efficiently spent my TPU time?</p>",
  "messages": [
    {
      "id": "834596",
      "postDate": "05/05/2020 16:33:04",
      "content": "<p>Hi guys,</p>\n\n<p>I am new to TPU use on Kaggle. The transformer NLP models take a long time to train on TPU. I am making modifications  either architecture the model or processing of dataset and to see the effect I am fine-tuning a pre-trained model, which is demanding for TPU time as you know. Then when I create a submission, I need to commit the notebook I modified, which runs everything again also takes time. Moreover, as you know, TPU time is limited, so what can I do to efficiently spent my TPU time?</p>",
      "rawMarkdown": "Hi guys,\n\nI am new to TPU use on Kaggle. The transformer NLP models take a long time to train on TPU. I am making modifications  either architecture the model or processing of dataset and to see the effect I am fine-tuning a pre-trained model, which is demanding for TPU time as you know. Then when I create a submission, I need to commit the notebook I modified, which runs everything again also takes time. Moreover, as you know, TPU time is limited, so what can I do to efficiently spent my TPU time?",
      "votes": null
    },
    {
      "id": "835377",
      "postDate": "05/06/2020 08:27:36",
      "content": "<p>You don't have to rerun the same code again for the commit. What you can do is save the model, or the predictions and submit them in a separate notebook.</p>",
      "rawMarkdown": "You don't have to rerun the same code again for the commit. What you can do is save the model, or the predictions and submit them in a separate notebook.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 835377,
      "author_name": "rafiko1",
      "author_url": "",
      "post_date": "05/06/2020 08:27:36",
      "content": "<p>You don't have to rerun the same code again for the commit. What you can do is save the model, or the predictions and submit them in a separate notebook.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "834596": "Hi guys,\n\nI am new to TPU use on Kaggle. The transformer NLP models take a long time to train on TPU. I am making modifications  either architecture the model or processing of dataset and to see the effect I am fine-tuning a pre-trained model, which is demanding for TPU time as you know. Then when I create a submission, I need to commit the notebook I modified, which runs everything again also takes time. Moreover, as you know, TPU time is limited, so what can I do to efficiently spent my TPU time?",
    "835377": "You don't have to rerun the same code again for the commit. What you can do is save the model, or the predictions and submit them in a separate notebook."
  },
  "source": "meta"
}