{
  "id": 309935,
  "title": "colab version of forked  pytorch arcface gempooling TPU training",
  "url": "/competitions/happy-whale-and-dolphin/discussion/309935",
  "author_name": "",
  "post_date": "2022-02-26T14:05:07.536736Z",
  "votes": 16,
  "comment_count": 8,
  "views": 0,
  "content": "<p><a href=\"https://github.com/flydragon2018/colab_notebooks/blob/main/colab_pytorch_arcface_gempooling_tpu_train_b6.ipynb\" target=\"_blank\">https://github.com/flydragon2018/colab_notebooks/blob/main/colab_pytorch_arcface_gempooling_tpu_train_b6.ipynb</a></p>\n<p>I have copied one version of dataset to my gdrive for training.</p>",
  "messages": [
    {
      "id": "1705458",
      "postDate": "02/26/2022 14:05:07",
      "content": "<p><a href=\"https://github.com/flydragon2018/colab_notebooks/blob/main/colab_pytorch_arcface_gempooling_tpu_train_b6.ipynb\" target=\"_blank\">https://github.com/flydragon2018/colab_notebooks/blob/main/colab_pytorch_arcface_gempooling_tpu_train_b6.ipynb</a></p>\n<p>I have copied one version of dataset to my gdrive for training.</p>",
      "rawMarkdown": "https://github.com/flydragon2018/colab_notebooks/blob/main/colab_pytorch_arcface_gempooling_tpu_train_b6.ipynb\n\nI have copied one version of dataset to my gdrive for training.",
      "votes": null
    },
    {
      "id": "1705483",
      "postDate": "02/26/2022 14:25:55",
      "content": "<p>Nice. Thank you! Are you going to add inference/submission part as well?</p>",
      "rawMarkdown": "Nice. Thank you! Are you going to add inference/submission part as well?",
      "votes": null
    },
    {
      "id": "1705486",
      "postDate": "02/26/2022 14:26:28",
      "content": "<p>B4 no problem. when I switch to B6, something wrong.  I am dealing with it.</p>",
      "rawMarkdown": "B4 no problem. when I switch to B6, something wrong.  I am dealing with it.",
      "votes": null
    },
    {
      "id": "1705489",
      "postDate": "02/26/2022 14:27:37",
      "content": "<p>there is  inference notebook shared already.  </p>",
      "rawMarkdown": "there is  inference notebook shared already.",
      "votes": null
    },
    {
      "id": "1705533",
      "postDate": "02/26/2022 15:14:14",
      "content": "<p>Where did you get the dataset?<br>\n\"train_images-384-384\"<br>\n\"test_images-384-384\"</p>",
      "rawMarkdown": "Where did you get the dataset?\n\"train_images-384-384\"\n\"test_images-384-384\"",
      "votes": null
    },
    {
      "id": "1705563",
      "postDate": "02/26/2022 15:39:46",
      "content": "<p>from Kaggle shared public dataset</p>\n<p><a href=\"https://www.kaggle.com/rdizzl3/jpeg-happywhale-384x384\" target=\"_blank\">https://www.kaggle.com/rdizzl3/jpeg-happywhale-384x384</a></p>",
      "rawMarkdown": "from Kaggle shared public dataset\n\nhttps://www.kaggle.com/rdizzl3/jpeg-happywhale-384x384",
      "votes": null
    },
    {
      "id": "1705567",
      "postDate": "02/26/2022 15:40:56",
      "content": "<p>have to reduce batch size.<br>\nIt is much slower compared to TF TPU training.</p>",
      "rawMarkdown": "have to reduce batch size.\nIt is much slower compared to TF TPU training.",
      "votes": null
    },
    {
      "id": "1705953",
      "postDate": "02/27/2022 02:33:26",
      "content": "<p>nice work！</p>",
      "rawMarkdown": "nice work！",
      "votes": null
    },
    {
      "id": "1706377",
      "postDate": "02/27/2022 12:49:14",
      "content": "<p>Perhaps Colab pro+ could run much faster?<br>\nColab pro  settings not good enough. </p>",
      "rawMarkdown": "Perhaps Colab pro+ could run much faster?\nColab pro  settings not good enough.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1705483,
      "author_name": "remekkinas",
      "author_url": "",
      "post_date": "02/26/2022 14:25:55",
      "content": "<p>Nice. Thank you! Are you going to add inference/submission part as well?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1705489,
          "author_name": "dragonzhang",
          "author_url": "",
          "post_date": "02/26/2022 14:27:37",
          "content": "<p>there is  inference notebook shared already.  </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1705486,
      "author_name": "dragonzhang",
      "author_url": "",
      "post_date": "02/26/2022 14:26:28",
      "content": "<p>B4 no problem. when I switch to B6, something wrong.  I am dealing with it.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1705567,
          "author_name": "dragonzhang",
          "author_url": "",
          "post_date": "02/26/2022 15:40:56",
          "content": "<p>have to reduce batch size.<br>\nIt is much slower compared to TF TPU training.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1705533,
      "author_name": "hzp22333",
      "author_url": "",
      "post_date": "02/26/2022 15:14:14",
      "content": "<p>Where did you get the dataset?<br>\n\"train_images-384-384\"<br>\n\"test_images-384-384\"</p>",
      "votes": null,
      "replies": [
        {
          "id": 1705563,
          "author_name": "dragonzhang",
          "author_url": "",
          "post_date": "02/26/2022 15:39:46",
          "content": "<p>from Kaggle shared public dataset</p>\n<p><a href=\"https://www.kaggle.com/rdizzl3/jpeg-happywhale-384x384\" target=\"_blank\">https://www.kaggle.com/rdizzl3/jpeg-happywhale-384x384</a></p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1705953,
      "author_name": "rainfalllove",
      "author_url": "",
      "post_date": "02/27/2022 02:33:26",
      "content": "<p>nice work！</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1706377,
      "author_name": "dragonzhang",
      "author_url": "",
      "post_date": "02/27/2022 12:49:14",
      "content": "<p>Perhaps Colab pro+ could run much faster?<br>\nColab pro  settings not good enough. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1705458": "https://github.com/flydragon2018/colab_notebooks/blob/main/colab_pytorch_arcface_gempooling_tpu_train_b6.ipynb\n\nI have copied one version of dataset to my gdrive for training.",
    "1705483": "Nice. Thank you! Are you going to add inference/submission part as well?",
    "1705486": "B4 no problem. when I switch to B6, something wrong.  I am dealing with it.",
    "1705489": "there is  inference notebook shared already.",
    "1705533": "Where did you get the dataset?\n\"train_images-384-384\"\n\"test_images-384-384\"",
    "1705563": "from Kaggle shared public dataset\n\nhttps://www.kaggle.com/rdizzl3/jpeg-happywhale-384x384",
    "1705567": "have to reduce batch size.\nIt is much slower compared to TF TPU training.",
    "1705953": "nice work！",
    "1706377": "Perhaps Colab pro+ could run much faster?\nColab pro  settings not good enough."
  },
  "source": "meta"
}