{
  "id": 569730,
  "title": "GPU disabled? Hold my CPU",
  "url": "/competitions/birdclef-2025/discussion/569730",
  "author_name": "",
  "post_date": "2025-03-23T19:27:41.623065500Z",
  "votes": 10,
  "comment_count": 4,
  "views": 0,
  "content": "<p>From the competition rules: </p>\n<blockquote>\n  <p>GPU Notebook submissions are disabled. You can technically submit but will only have 1 minute of runtime.</p>\n</blockquote>\n<p>Is that a restriction, or a challenge? 😏</p>\n<p>After playing around with joblib and batching, I was about to submit a baseline efficientnet-b0 model using P100 GPU! It runs in 57 seconds when testing on 700 train_soundscapes (see version 1)</p>\n<p>Notebooks:</p>\n<ul>\n<li><a href=\"https://www.kaggle.com/code/robbynevels/bc25-train-specs-signed-db-80-n-16\" target=\"_blank\">Train preprocessing</a></li>\n<li><a href=\"https://www.kaggle.com/code/robbynevels/bc25-train-baseline\" target=\"_blank\">Training</a></li>\n<li><a href=\"https://www.kaggle.com/code/robbynevels/bc25-infer-baseline-gpu/\" target=\"_blank\">GPU Inference</a></li>\n</ul>",
  "messages": [
    {
      "id": "3157769",
      "postDate": "03/23/2025 19:27:41",
      "content": "<p>From the competition rules: </p>\n<blockquote>\n  <p>GPU Notebook submissions are disabled. You can technically submit but will only have 1 minute of runtime.</p>\n</blockquote>\n<p>Is that a restriction, or a challenge? 😏</p>\n<p>After playing around with joblib and batching, I was about to submit a baseline efficientnet-b0 model using P100 GPU! It runs in 57 seconds when testing on 700 train_soundscapes (see version 1)</p>\n<p>Notebooks:</p>\n<ul>\n<li><a href=\"https://www.kaggle.com/code/robbynevels/bc25-train-specs-signed-db-80-n-16\" target=\"_blank\">Train preprocessing</a></li>\n<li><a href=\"https://www.kaggle.com/code/robbynevels/bc25-train-baseline\" target=\"_blank\">Training</a></li>\n<li><a href=\"https://www.kaggle.com/code/robbynevels/bc25-infer-baseline-gpu/\" target=\"_blank\">GPU Inference</a></li>\n</ul>",
      "rawMarkdown": "From the competition rules: \n\n> GPU Notebook submissions are disabled. You can technically submit but will only have 1 minute of runtime.\n\nIs that a restriction, or a challenge? 😏\n\nAfter playing around with joblib and batching, I was about to submit a baseline efficientnet-b0 model using P100 GPU! It runs in 57 seconds when testing on 700 train_soundscapes (see version 1)\n\nNotebooks:\n- [Train preprocessing](https://www.kaggle.com/code/robbynevels/bc25-train-specs-signed-db-80-n-16)\n- [Training](https://www.kaggle.com/code/robbynevels/bc25-train-baseline)\n- [GPU Inference](https://www.kaggle.com/code/robbynevels/bc25-infer-baseline-gpu/)",
      "votes": null
    },
    {
      "id": "3157807",
      "postDate": "03/23/2025 21:09:16",
      "content": "<p>Amazing job! <br>\nNot only you have speed up your model, but you have also overcame the bottleneck of loading the soundscape files!</p>\n<blockquote>\n  <p>It should take your submission notebook approximately five minutes to load all the test soundscapes</p>\n</blockquote>",
      "rawMarkdown": "Amazing job! \nNot only you have speed up your model, but you have also overcame the bottleneck of loading the soundscape files!\n\n> It should take your submission notebook approximately five minutes to load all the test soundscapes",
      "votes": null
    },
    {
      "id": "3157898",
      "postDate": "03/24/2025 02:18:16",
      "content": "<p>Thanks! Indeed, although it still loads in 40-50 seconds even without parallelized reads. I think that text might be left over from 2024 -- the same guidance was given back then, but this year has much less data.</p>",
      "rawMarkdown": "Thanks! Indeed, although it still loads in 40-50 seconds even without parallelized reads. I think that text might be left over from 2024 -- the same guidance was given back then, but this year has much less data.",
      "votes": null
    },
    {
      "id": "3158563",
      "postDate": "03/24/2025 16:56:41",
      "content": "<p>well done!</p>",
      "rawMarkdown": "well done!",
      "votes": null
    },
    {
      "id": "3160285",
      "postDate": "03/26/2025 15:33:26",
      "content": "<p>Ha! Perhaps we should add BirdCLEF to speedrun.com…</p>",
      "rawMarkdown": "Ha! Perhaps we should add BirdCLEF to speedrun.com...",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3157807,
      "author_name": "kdmitrie",
      "author_url": "",
      "post_date": "03/23/2025 21:09:16",
      "content": "<p>Amazing job! <br>\nNot only you have speed up your model, but you have also overcame the bottleneck of loading the soundscape files!</p>\n<blockquote>\n  <p>It should take your submission notebook approximately five minutes to load all the test soundscapes</p>\n</blockquote>",
      "votes": null,
      "replies": [
        {
          "id": 3157898,
          "author_name": "robbynevels",
          "author_url": "",
          "post_date": "03/24/2025 02:18:16",
          "content": "<p>Thanks! Indeed, although it still loads in 40-50 seconds even without parallelized reads. I think that text might be left over from 2024 -- the same guidance was given back then, but this year has much less data.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 3158563,
      "author_name": "shohinurpervezshohan",
      "author_url": "",
      "post_date": "03/24/2025 16:56:41",
      "content": "<p>well done!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3160285,
      "author_name": "tomdenton",
      "author_url": "",
      "post_date": "03/26/2025 15:33:26",
      "content": "<p>Ha! Perhaps we should add BirdCLEF to speedrun.com…</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3157769": "From the competition rules: \n\n> GPU Notebook submissions are disabled. You can technically submit but will only have 1 minute of runtime.\n\nIs that a restriction, or a challenge? 😏\n\nAfter playing around with joblib and batching, I was about to submit a baseline efficientnet-b0 model using P100 GPU! It runs in 57 seconds when testing on 700 train_soundscapes (see version 1)\n\nNotebooks:\n- [Train preprocessing](https://www.kaggle.com/code/robbynevels/bc25-train-specs-signed-db-80-n-16)\n- [Training](https://www.kaggle.com/code/robbynevels/bc25-train-baseline)\n- [GPU Inference](https://www.kaggle.com/code/robbynevels/bc25-infer-baseline-gpu/)",
    "3157807": "Amazing job! \nNot only you have speed up your model, but you have also overcame the bottleneck of loading the soundscape files!\n\n> It should take your submission notebook approximately five minutes to load all the test soundscapes",
    "3157898": "Thanks! Indeed, although it still loads in 40-50 seconds even without parallelized reads. I think that text might be left over from 2024 -- the same guidance was given back then, but this year has much less data.",
    "3158563": "well done!",
    "3160285": "Ha! Perhaps we should add BirdCLEF to speedrun.com..."
  },
  "source": "meta"
}