{
  "id": 401904,
  "title": "Why does everyone use efficientnet b0 or b1 instead of a larger models ?",
  "url": "/competitions/birdclef-2023/discussion/401904",
  "author_name": "",
  "post_date": "2023-04-15T15:37:28.098759500Z",
  "votes": 7,
  "comment_count": 3,
  "views": 0,
  "content": "<p>In the most obvious example, b7 would be more accurate than b0. Is this an idea to avoid exceeding the inference time limit?</p>",
  "messages": [
    {
      "id": "2222826",
      "postDate": "04/15/2023 15:37:28",
      "content": "<p>In the most obvious example, b7 would be more accurate than b0. Is this an idea to avoid exceeding the inference time limit?</p>",
      "rawMarkdown": "In the most obvious example, b7 would be more accurate than b0. Is this an idea to avoid exceeding the inference time limit?",
      "votes": null
    },
    {
      "id": "2223581",
      "postDate": "04/16/2023 12:14:20",
      "content": "<p>Yes, you are right. It is done to avoid exceeding the time limit. But also in old Birdclef competitions larger models did not perform better. Most of the models used were smaller models.</p>",
      "rawMarkdown": "Yes, you are right. It is done to avoid exceeding the time limit. But also in old Birdclef competitions larger models did not perform better. Most of the models used were smaller models.",
      "votes": null
    },
    {
      "id": "2228578",
      "postDate": "04/20/2023 16:21:14",
      "content": "<p>Mostly compute.<br>\nAlso bigger models didnt produce better results in the past on this problem.</p>\n<p>The Devastator.</p>",
      "rawMarkdown": "Mostly compute.\nAlso bigger models didnt produce better results in the past on this problem.\n\nThe Devastator.",
      "votes": null
    },
    {
      "id": "2229174",
      "postDate": "04/21/2023 06:05:21",
      "content": "<p>smaller models perform better.</p>",
      "rawMarkdown": "smaller models perform better.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2223581,
      "author_name": "ivanaerlic",
      "author_url": "",
      "post_date": "04/16/2023 12:14:20",
      "content": "<p>Yes, you are right. It is done to avoid exceeding the time limit. But also in old Birdclef competitions larger models did not perform better. Most of the models used were smaller models.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2228578,
      "author_name": "thedevastator",
      "author_url": "",
      "post_date": "04/20/2023 16:21:14",
      "content": "<p>Mostly compute.<br>\nAlso bigger models didnt produce better results in the past on this problem.</p>\n<p>The Devastator.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2229174,
      "author_name": "gnannan",
      "author_url": "",
      "post_date": "04/21/2023 06:05:21",
      "content": "<p>smaller models perform better.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2222826": "In the most obvious example, b7 would be more accurate than b0. Is this an idea to avoid exceeding the inference time limit?",
    "2223581": "Yes, you are right. It is done to avoid exceeding the time limit. But also in old Birdclef competitions larger models did not perform better. Most of the models used were smaller models.",
    "2228578": "Mostly compute.\nAlso bigger models didnt produce better results in the past on this problem.\n\nThe Devastator.",
    "2229174": "smaller models perform better."
  },
  "source": "meta"
}