{
  "id": 185365,
  "title": "Efficient ways to train models with the Kaggle's GPU and TPU.",
  "url": "/competitions/landmark-recognition-2020/discussion/185365",
  "author_name": "",
  "post_date": "2020-09-20T15:24:34.748536900Z",
  "votes": 2,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Please share your views on how to train models with the limited GPU and TPU time of Kaggle.</p>",
  "messages": [
    {
      "id": "1019663",
      "postDate": "09/20/2020 15:24:34",
      "content": "<p>Please share your views on how to train models with the limited GPU and TPU time of Kaggle.</p>",
      "rawMarkdown": "Please share your views on how to train models with the limited GPU and TPU time of Kaggle.",
      "votes": null
    },
    {
      "id": "1019674",
      "postDate": "09/20/2020 15:32:51",
      "content": "<p><a href=\"https://www.kaggle.com/vikrant06\" target=\"_blank\">@vikrant06</a> You can try to train models by 9 hours (on GPU), and save their weights as kernel output.<br>\nYou can also try google colab.</p>",
      "rawMarkdown": "vikrant06 You can try to train models by 9 hours (on GPU), and save their weights as kernel output.\nYou can also try google colab.",
      "votes": null
    },
    {
      "id": "1019704",
      "postDate": "09/20/2020 16:03:45",
      "content": "<p>So how many times did u run to get the final model? Can you through some light on your model hyperparameters.</p>",
      "rawMarkdown": "So how many times did u run to get the final model? Can you through some light on your model hyperparameters.",
      "votes": null
    },
    {
      "id": "1021193",
      "postDate": "09/21/2020 17:29:29",
      "content": "<p>I think google colab is a good way to train and save your model and then import it in kaggle.</p>",
      "rawMarkdown": "I think google colab is a good way to train and save your model and then import it in kaggle.",
      "votes": null
    },
    {
      "id": "1021217",
      "postDate": "09/21/2020 17:40:59",
      "content": "<p>Thanks. Saw that you are a notebook expert.  Have u published any notebook to solve the above problem?</p>",
      "rawMarkdown": "Thanks. Saw that you are a notebook expert.  Have u published any notebook to solve the above problem?",
      "votes": null
    },
    {
      "id": "1021865",
      "postDate": "09/22/2020 07:10:07",
      "content": "<p>Hey <a href=\"https://www.kaggle.com/vikrant06\" target=\"_blank\">@vikrant06</a>. Please checkout my notebook where you can access either GPU or TPU. </p>\n<p><a href=\"https://www.kaggle.com/sanjaydsb/efficientnets-in-pytorch-using-tpus-and-gpus\" target=\"_blank\">https://www.kaggle.com/sanjaydsb/efficientnets-in-pytorch-using-tpus-and-gpus</a> </p>",
      "rawMarkdown": "Hey @vikrant06. Please checkout my notebook where you can access either GPU or TPU. \n\nhttps://www.kaggle.com/sanjaydsb/efficientnets-in-pytorch-using-tpus-and-gpus",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1019674,
      "author_name": "ihelon",
      "author_url": "",
      "post_date": "09/20/2020 15:32:51",
      "content": "<p><a href=\"https://www.kaggle.com/vikrant06\" target=\"_blank\">@vikrant06</a> You can try to train models by 9 hours (on GPU), and save their weights as kernel output.<br>\nYou can also try google colab.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1019704,
          "author_name": "vikrant06",
          "author_url": "",
          "post_date": "09/20/2020 16:03:45",
          "content": "<p>So how many times did u run to get the final model? Can you through some light on your model hyperparameters.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1021193,
      "author_name": "",
      "author_url": "",
      "post_date": "09/21/2020 17:29:29",
      "content": "<p>I think google colab is a good way to train and save your model and then import it in kaggle.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1021217,
          "author_name": "vikrant06",
          "author_url": "",
          "post_date": "09/21/2020 17:40:59",
          "content": "<p>Thanks. Saw that you are a notebook expert.  Have u published any notebook to solve the above problem?</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1021865,
      "author_name": "sanjaydsb",
      "author_url": "",
      "post_date": "09/22/2020 07:10:07",
      "content": "<p>Hey <a href=\"https://www.kaggle.com/vikrant06\" target=\"_blank\">@vikrant06</a>. Please checkout my notebook where you can access either GPU or TPU. </p>\n<p><a href=\"https://www.kaggle.com/sanjaydsb/efficientnets-in-pytorch-using-tpus-and-gpus\" target=\"_blank\">https://www.kaggle.com/sanjaydsb/efficientnets-in-pytorch-using-tpus-and-gpus</a> </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1019663": "Please share your views on how to train models with the limited GPU and TPU time of Kaggle.",
    "1019674": "vikrant06 You can try to train models by 9 hours (on GPU), and save their weights as kernel output.\nYou can also try google colab.",
    "1019704": "So how many times did u run to get the final model? Can you through some light on your model hyperparameters.",
    "1021193": "I think google colab is a good way to train and save your model and then import it in kaggle.",
    "1021217": "Thanks. Saw that you are a notebook expert.  Have u published any notebook to solve the above problem?",
    "1021865": "Hey @vikrant06. Please checkout my notebook where you can access either GPU or TPU. \n\nhttps://www.kaggle.com/sanjaydsb/efficientnets-in-pytorch-using-tpus-and-gpus"
  },
  "source": "meta"
}