{
  "id": 161214,
  "title": "How to make pytorch TPU training faster??",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/161214",
  "author_name": "",
  "post_date": "2020-06-24T07:50:40.699094600Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>On GPU, it takes 2m45s to run a epoch. On TPU it takes 3m20s. Pytorch XLA\nPlease share some tips to speed up the training on TPU(pytorch)</p>",
  "messages": [
    {
      "id": "899386",
      "postDate": "06/24/2020 07:50:40",
      "content": "<p>On GPU, it takes 2m45s to run a epoch. On TPU it takes 3m20s. Pytorch XLA\nPlease share some tips to speed up the training on TPU(pytorch)</p>",
      "rawMarkdown": "On GPU, it takes 2m45s to run a epoch. On TPU it takes 3m20s. Pytorch XLA\nPlease share some tips to speed up the training on TPU(pytorch)",
      "votes": null
    },
    {
      "id": "899412",
      "postDate": "06/24/2020 08:14:38",
      "content": "<p>Did you check many epochs or just one?</p>",
      "rawMarkdown": "Did you check many epochs or just one?",
      "votes": null
    },
    {
      "id": "899424",
      "postDate": "06/24/2020 08:26:44",
      "content": "<p>Just first epoch....Thank you for your reminder, i'll check the whole training process</p>",
      "rawMarkdown": "Just first epoch....Thank you for your reminder, i'll check the whole training process",
      "votes": null
    },
    {
      "id": "919122",
      "postDate": "07/07/2020 18:09:50",
      "content": "<p>did you succeed to have competitive results about the speed ? when I tried to experiment a small resnet18 the training takes ages.</p>",
      "rawMarkdown": "did you succeed to have competitive results about the speed ? when I tried to experiment a small resnet18 the training takes ages.",
      "votes": null
    },
    {
      "id": "919274",
      "postDate": "07/07/2020 19:21:06",
      "content": "<p>See <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/164020#916082\">PyTorch TPU tips</a></p>",
      "rawMarkdown": "See [PyTorch TPU tips](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/164020#916082)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 899412,
      "author_name": "abhishek",
      "author_url": "",
      "post_date": "06/24/2020 08:14:38",
      "content": "<p>Did you check many epochs or just one?</p>",
      "votes": null,
      "replies": [
        {
          "id": 899424,
          "author_name": "changewow",
          "author_url": "",
          "post_date": "06/24/2020 08:26:44",
          "content": "<p>Just first epoch....Thank you for your reminder, i'll check the whole training process</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 919122,
          "author_name": "ludovick",
          "author_url": "",
          "post_date": "07/07/2020 18:09:50",
          "content": "<p>did you succeed to have competitive results about the speed ? when I tried to experiment a small resnet18 the training takes ages.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 919274,
          "author_name": "sirishks",
          "author_url": "",
          "post_date": "07/07/2020 19:21:06",
          "content": "<p>See <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/164020#916082\">PyTorch TPU tips</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "899386": "On GPU, it takes 2m45s to run a epoch. On TPU it takes 3m20s. Pytorch XLA\nPlease share some tips to speed up the training on TPU(pytorch)",
    "899412": "Did you check many epochs or just one?",
    "899424": "Just first epoch....Thank you for your reminder, i'll check the whole training process",
    "919122": "did you succeed to have competitive results about the speed ? when I tried to experiment a small resnet18 the training takes ages.",
    "919274": "See [PyTorch TPU tips](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/164020#916082)"
  },
  "source": "meta"
}