{
  "id": 290677,
  "title": "External training allowed?",
  "url": "/competitions/tensorflow-great-barrier-reef/discussion/290677",
  "author_name": "Rishiraj Acharya",
  "post_date": "2021-11-25T18:04:56.961000",
  "votes": 8,
  "comment_count": 4,
  "views": 0,
  "content": "<p>The 9 hours run-time limit of Kaggle notebooks and also the limited RAM makes it very difficult for larger object detection models to be trained. Are we allowed to use external GPUs for the model training part and then load the trained model for inference in a Kaggle notebook from \"+ Add Data\" section? Forgive me if this sounds silly, I'm just a beginner in Kaggle.<br>\n<a href=\"https://www.kaggle.com/addisonhoward\" target=\"_blank\">@addisonhoward</a> </p>",
  "messages": [
    {
      "id": 1595480,
      "postDate": "2021-11-25T18:04:56.963Z",
      "content": "<p>The 9 hours run-time limit of Kaggle notebooks and also the limited RAM makes it very difficult for larger object detection models to be trained. Are we allowed to use external GPUs for the model training part and then load the trained model for inference in a Kaggle notebook from \"+ Add Data\" section? Forgive me if this sounds silly, I'm just a beginner in Kaggle.<br>\n<a href=\"https://www.kaggle.com/addisonhoward\" target=\"_blank\">@addisonhoward</a> </p>",
      "rawMarkdown": "The 9 hours run-time limit of Kaggle notebooks and also the limited RAM makes it very difficult for larger object detection models to be trained. Are we allowed to use external GPUs for the model training part and then load the trained model for inference in a Kaggle notebook from \"+ Add Data\" section? Forgive me if this sounds silly, I'm just a beginner in Kaggle.\n@addisonhoward ",
      "votes": 8
    },
    {
      "id": 1597218,
      "postDate": "2021-11-27T10:19:28.350Z",
      "content": "<blockquote>\n  <p>I'm just a beginner in Kaggle</p>\n</blockquote>\n<p>A beginner that is 7th. Respect! </p>\n<p>On the topic: You can train the model in one notebook, save the weights and then use another notebook for inference if you want to only use Kaggle notebooks. <br>\nAnyhow, you are always allowed to train the model wherever you like: on local GPU, On a cloud, compute the derivatives by hand.. Just upload trained weights and you are good to go! </p>",
      "rawMarkdown": ">I'm just a beginner in Kaggle\n\nA beginner that is 7th. Respect! \n\nOn the topic: You can train the model in one notebook, save the weights and then use another notebook for inference if you want to only use Kaggle notebooks. \nAnyhow, you are always allowed to train the model wherever you like: on local GPU, On a cloud, compute the derivatives by hand.. Just upload trained weights and you are good to go! ",
      "votes": 4
    },
    {
      "id": 1596580,
      "postDate": "2021-11-26T16:37:03.017Z",
      "content": "<p>You can run your code locally, using your GPU, and saving the model.<br>\nAnother possibility is using Colab, or Cloud Services, and importing the pre-trained model.<br>\nKaggle limits are already a gift hahaha.<br>\nWhat I'm currently doing is: Pre-train on Colab, import and run on Kaggle to present remotely (With the code running).</p>",
      "rawMarkdown": "You can run your code locally, using your GPU, and saving the model.\nAnother possibility is using Colab, or Cloud Services, and importing the pre-trained model.\nKaggle limits are already a gift hahaha.\nWhat I'm currently doing is: Pre-train on Colab, import and run on Kaggle to present remotely (With the code running).",
      "votes": 2
    },
    {
      "id": 1595542,
      "postDate": "2021-11-25T18:58:04.267Z",
      "content": "<p>Yes - create the model anywhere in the world - run the inference on Kaggle using the model from a data set.</p>",
      "rawMarkdown": "Yes - create the model anywhere in the world - run the inference on Kaggle using the model from a data set.",
      "votes": 2
    },
    {
      "id": 1595570,
      "postDate": "2021-11-25T19:34:26.703Z",
      "rawMarkdown": "",
      "votes": 1,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 1597218,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-11-27T10:19:28.350000",
      "content": "<blockquote>\n  <p>I'm just a beginner in Kaggle</p>\n</blockquote>\n<p>A beginner that is 7th. Respect! </p>\n<p>On the topic: You can train the model in one notebook, save the weights and then use another notebook for inference if you want to only use Kaggle notebooks. <br>\nAnyhow, you are always allowed to train the model wherever you like: on local GPU, On a cloud, compute the derivatives by hand.. Just upload trained weights and you are good to go! </p>",
      "votes": 4,
      "replies": []
    },
    {
      "id": 1596580,
      "author_name": "Lucas Silva",
      "author_url": "",
      "post_date": "2021-11-26T16:37:03.017000",
      "content": "<p>You can run your code locally, using your GPU, and saving the model.<br>\nAnother possibility is using Colab, or Cloud Services, and importing the pre-trained model.<br>\nKaggle limits are already a gift hahaha.<br>\nWhat I'm currently doing is: Pre-train on Colab, import and run on Kaggle to present remotely (With the code running).</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1595542,
      "author_name": "PC Jimmmy",
      "author_url": "",
      "post_date": "2021-11-25T18:58:04.267000",
      "content": "<p>Yes - create the model anywhere in the world - run the inference on Kaggle using the model from a data set.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1595570,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-11-25T19:34:26.703000",
      "content": "",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1595480": "The 9 hours run-time limit of Kaggle notebooks and also the limited RAM makes it very difficult for larger object detection models to be trained. Are we allowed to use external GPUs for the model training part and then load the trained model for inference in a Kaggle notebook from \"+ Add Data\" section? Forgive me if this sounds silly, I'm just a beginner in Kaggle.\n@addisonhoward ",
    "1597218": ">I'm just a beginner in Kaggle\n\nA beginner that is 7th. Respect! \n\nOn the topic: You can train the model in one notebook, save the weights and then use another notebook for inference if you want to only use Kaggle notebooks. \nAnyhow, you are always allowed to train the model wherever you like: on local GPU, On a cloud, compute the derivatives by hand.. Just upload trained weights and you are good to go! ",
    "1596580": "You can run your code locally, using your GPU, and saving the model.\nAnother possibility is using Colab, or Cloud Services, and importing the pre-trained model.\nKaggle limits are already a gift hahaha.\nWhat I'm currently doing is: Pre-train on Colab, import and run on Kaggle to present remotely (With the code running).",
    "1595542": "Yes - create the model anywhere in the world - run the inference on Kaggle using the model from a data set.",
    "1595570": ""
  }
}