{
  "id": 613296,
  "title": "Regarding the Use of Pre-trained Models",
  "url": "/competitions/MABe-mouse-behavior-detection/discussion/613296",
  "author_name": "",
  "post_date": "2025-10-25T15:51:59.911471800Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi everyone,</p>\n<p>I’m quite new to Kaggle competitions. I noticed that the Code Requirements mention that pretrained models are allowed, but I’d like to confirm the correct way to use them.</p>\n<p>If I train a model on the official competition data within a Kaggle Notebook, save the trained weights as a private Kaggle Dataset, and then load this pretrained model in my submission Notebook for inference (without using any external data or internet access) , is this workflow fully compliant with the competition rules?</p>\n<p>I just want to avoid any rule violations, so any clarification or advice would be really appreciated!</p>\n<p>Thanks in advance! 🙏</p>",
  "messages": [
    {
      "id": "3306926",
      "postDate": "10/25/2025 15:51:59",
      "content": "<p>Hi everyone,</p>\n<p>I’m quite new to Kaggle competitions. I noticed that the Code Requirements mention that pretrained models are allowed, but I’d like to confirm the correct way to use them.</p>\n<p>If I train a model on the official competition data within a Kaggle Notebook, save the trained weights as a private Kaggle Dataset, and then load this pretrained model in my submission Notebook for inference (without using any external data or internet access) , is this workflow fully compliant with the competition rules?</p>\n<p>I just want to avoid any rule violations, so any clarification or advice would be really appreciated!</p>\n<p>Thanks in advance! 🙏</p>",
      "rawMarkdown": "Hi everyone,\n\nI’m quite new to Kaggle competitions. I noticed that the Code Requirements mention that pretrained models are allowed, but I’d like to confirm the correct way to use them.\n\nIf I train a model on the official competition data within a Kaggle Notebook, save the trained weights as a private Kaggle Dataset, and then load this pretrained model in my submission Notebook for inference (without using any external data or internet access) , is this workflow fully compliant with the competition rules?\n\nI just want to avoid any rule violations, so any clarification or advice would be really appreciated!\n\nThanks in advance! 🙏",
      "votes": null
    },
    {
      "id": "3306991",
      "postDate": "10/25/2025 21:43:13",
      "content": "<p>Sure, this is what most people on Kaggle are doing, separate the training and inference stages.</p>",
      "rawMarkdown": "Sure, this is what most people on Kaggle are doing, separate the training and inference stages.",
      "votes": null
    },
    {
      "id": "3307037",
      "postDate": "10/26/2025 02:15:11",
      "content": "<p>Thanks for the reply! Really appreciate your help 🙏</p>",
      "rawMarkdown": "Thanks for the reply! Really appreciate your help 🙏",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3306991,
      "author_name": "lizhecheng",
      "author_url": "",
      "post_date": "10/25/2025 21:43:13",
      "content": "<p>Sure, this is what most people on Kaggle are doing, separate the training and inference stages.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3307037,
          "author_name": "bioconvolutionyt",
          "author_url": "",
          "post_date": "10/26/2025 02:15:11",
          "content": "<p>Thanks for the reply! Really appreciate your help 🙏</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3306926": "Hi everyone,\n\nI’m quite new to Kaggle competitions. I noticed that the Code Requirements mention that pretrained models are allowed, but I’d like to confirm the correct way to use them.\n\nIf I train a model on the official competition data within a Kaggle Notebook, save the trained weights as a private Kaggle Dataset, and then load this pretrained model in my submission Notebook for inference (without using any external data or internet access) , is this workflow fully compliant with the competition rules?\n\nI just want to avoid any rule violations, so any clarification or advice would be really appreciated!\n\nThanks in advance! 🙏",
    "3306991": "Sure, this is what most people on Kaggle are doing, separate the training and inference stages.",
    "3307037": "Thanks for the reply! Really appreciate your help 🙏"
  },
  "source": "meta"
}