{
  "id": 287929,
  "title": "3rd Place Solution",
  "content": "<p><strong>Muril-Large + Magic</strong></p>\n<p>First of all, I would like to thank Kaggle and the competition host for hosting this challenge.<br>\nThe third place came as a surprise to me and I was not expecting it at all, since my public LB score was not high.<br>\nMuril is all I have. I have found that xlm-r is more efficient when finetuning on the squad 2.0 dataset, so I tried it similarly with muril-large.</p>\n<p><strong>Data</strong></p>\n<ul>\n<li>competition data, MLQA, XQUAD, SQUAD v2</li>\n</ul>\n<p><strong>Script to finetune muril on squad 2.0:</strong></p>\n<ul>\n<li><a href=\"https://www.kaggle.com/nguyenduongthanh/fine-tune-muril-squad-v2\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/fine-tune-muril-squad-v2</a></li>\n</ul>\n<p><strong>Muril-Large-Squad-v2 Checkpoint</strong></p>\n<ul>\n<li><a href=\"https://www.kaggle.com/nguyenduongthanh/muril-large-squad-v2\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/muril-large-squad-v2</a></li>\n</ul>\n<p><strong>Fine-tuning:</strong><br>\nI just fine-tune on 2 epochs each fold</p>\n<ul>\n<li>MuRIL Large 10 folds: max-length=384, doc_stride=128, lr=1e-5, gradient accumulation=1</li>\n</ul>\n<p><strong>Ensemble</strong></p>\n<ul>\n<li>Simple average of 10 folds</li>\n</ul>\n<p><strong>Notebook</strong></p>\n<ul>\n<li>Train: <a href=\"https://www.kaggle.com/nguyenduongthanh/simple-train\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/simple-train</a></li>\n<li>Inference : <a href=\"https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large</a></li>\n</ul>",
  "comments": [
    {
      "id": 1588065,
      "author_name": "Mintwater",
      "author_url": "",
      "post_date": "2021-11-19T07:32:20.563000",
      "content": "<p>Amazing magic!</p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 1583734,
      "author_name": "KhanhVD",
      "author_url": "",
      "post_date": "2021-11-16T03:04:59.913000",
      "content": "<p>Good job <a href=\"https://www.kaggle.com/nguyenduongthanh\" target=\"_blank\">@nguyenduongthanh</a> and team, congrats on results!</p>",
      "votes": 4,
      "replies": []
    },
    {
      "id": 1591210,
      "author_name": "Parv Yadav",
      "author_url": "",
      "post_date": "2021-11-22T05:06:24.767000",
      "content": "<p>Congratulations <a href=\"https://www.kaggle.com/nguyenduongthanh\" target=\"_blank\">@nguyenduongthanh</a> !😄</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1590689,
      "author_name": "MINGOOK KIM",
      "author_url": "",
      "post_date": "2021-11-21T15:05:48.797000",
      "content": "<p>Congratulations on your achievements.</p>\n<p>Congratulations~</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1585164,
      "author_name": "adldotori",
      "author_url": "",
      "post_date": "2021-11-17T06:11:43.183000",
      "content": "<p>Congratulations. i'm envious of gold medal.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1584814,
      "author_name": "Darek Kłeczek",
      "author_url": "",
      "post_date": "2021-11-16T19:24:54.827000",
      "content": "<p>Congratulations! MURIL proved to be very effective, it was also the best model in my ensemble. </p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1584467,
      "author_name": "Mathurin Ache",
      "author_url": "",
      "post_date": "2021-11-16T14:57:21.623000",
      "content": "<p>Congratulations. Could you give more informations about Muril-Large ? A kernel to share ?</p>",
      "votes": 2,
      "replies": [
        {
          "id": 1584963,
          "author_name": "Thanh Nguyen",
          "author_url": "",
          "post_date": "2021-11-16T23:07:34.110000",
          "content": "<p>you can read it here: <a href=\"https://huggingface.co/google/muril-large-cased\" target=\"_blank\">https://huggingface.co/google/muril-large-cased</a></p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1584093,
      "author_name": "atfujita",
      "author_url": "",
      "post_date": "2021-11-16T09:12:00.650000",
      "content": "<p>Congratulations! <br>\nThis is smart.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583987,
      "author_name": "maze508",
      "author_url": "",
      "post_date": "2021-11-16T07:28:43.980000",
      "content": "<p>Congrats!!!</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583909,
      "author_name": "ysss",
      "author_url": "",
      "post_date": "2021-11-16T06:25:03.663000",
      "content": "<p>Simple is best! Congrats! </p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583893,
      "author_name": "Luca Massaron",
      "author_url": "",
      "post_date": "2021-11-16T06:06:05.933000",
      "content": "<p>So simple, but so effective! Thank you a lot for sharing the finetuning solution, I will study how it works! </p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583859,
      "author_name": "Mr_KnowNothing",
      "author_url": "",
      "post_date": "2021-11-16T05:34:40.990000",
      "content": "<p>This is absolutely wow , well done <a href=\"https://www.kaggle.com/nguyenduongthanh\" target=\"_blank\">@nguyenduongthanh</a> </p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583753,
      "author_name": "van73",
      "author_url": "",
      "post_date": "2021-11-16T03:31:31.183000",
      "content": "<p>Congratulations! 👍</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583743,
      "author_name": "Chris Deotte",
      "author_url": "",
      "post_date": "2021-11-16T03:20:08.270000",
      "content": "<p>Congratulations! Was your solution only <code>Muril-Large</code>?</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583706,
      "author_name": "Lake Arrowhead",
      "author_url": "",
      "post_date": "2021-11-16T02:39:50.573000",
      "content": "<p>i can't believe……….</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583678,
      "author_name": "Raghawendra Singh",
      "author_url": "",
      "post_date": "2021-11-16T01:58:53.560000",
      "content": "<p>Congrats for gold….</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583676,
      "author_name": "Nicholas Broad",
      "author_url": "",
      "post_date": "2021-11-16T01:58:20.327000",
      "content": "<p>What was your CV? Your notebook isn't public</p>",
      "votes": 2,
      "replies": [
        {
          "id": 1583679,
          "author_name": "Thanh Nguyen",
          "author_url": "",
          "post_date": "2021-11-16T02:00:06.243000",
          "content": "<p>My notebook here: <a href=\"https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large</a></p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1583690,
          "author_name": "Nicholas Broad",
          "author_url": "",
          "post_date": "2021-11-16T02:12:15.233000",
          "content": "<p>Sorry I meant the fine-tuning one isn't public</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 1583700,
          "author_name": "Thanh Nguyen",
          "author_url": "",
          "post_date": "2021-11-16T02:21:46.520000",
          "content": "<p>Fine-tuning: <a href=\"https://www.kaggle.com/nguyenduongthanh/simple-train\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/simple-train</a></p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1583662,
      "author_name": "DrHB",
      "author_url": "",
      "post_date": "2021-11-16T01:41:52.630000",
      "content": "<p>Congrats! Great finish</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583660,
      "author_name": "Kiran Chaudhary",
      "author_url": "",
      "post_date": "2021-11-16T01:41:07.180000",
      "content": "<p>Congratulations!!!!!</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1583649,
      "author_name": "Leon",
      "author_url": "",
      "post_date": "2021-11-16T01:31:25.883000",
      "content": "<p>Simple but effective solution! Congratulations!</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 1587618,
      "author_name": "",
      "author_url": "",
      "post_date": "2021-11-18T20:19:14.917000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "comment_count": 25,
  "messages": [
    {
      "id": 1583647,
      "postDate": "2021-11-16T01:29:51.487Z",
      "content": "<p><strong>Muril-Large + Magic</strong></p>\n<p>First of all, I would like to thank Kaggle and the competition host for hosting this challenge.<br>\nThe third place came as a surprise to me and I was not expecting it at all, since my public LB score was not high.<br>\nMuril is all I have. I have found that xlm-r is more efficient when finetuning on the squad 2.0 dataset, so I tried it similarly with muril-large.</p>\n<p><strong>Data</strong></p>\n<ul>\n<li>competition data, MLQA, XQUAD, SQUAD v2</li>\n</ul>\n<p><strong>Script to finetune muril on squad 2.0:</strong></p>\n<ul>\n<li><a href=\"https://www.kaggle.com/nguyenduongthanh/fine-tune-muril-squad-v2\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/fine-tune-muril-squad-v2</a></li>\n</ul>\n<p><strong>Muril-Large-Squad-v2 Checkpoint</strong></p>\n<ul>\n<li><a href=\"https://www.kaggle.com/nguyenduongthanh/muril-large-squad-v2\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/muril-large-squad-v2</a></li>\n</ul>\n<p><strong>Fine-tuning:</strong><br>\nI just fine-tune on 2 epochs each fold</p>\n<ul>\n<li>MuRIL Large 10 folds: max-length=384, doc_stride=128, lr=1e-5, gradient accumulation=1</li>\n</ul>\n<p><strong>Ensemble</strong></p>\n<ul>\n<li>Simple average of 10 folds</li>\n</ul>\n<p><strong>Notebook</strong></p>\n<ul>\n<li>Train: <a href=\"https://www.kaggle.com/nguyenduongthanh/simple-train\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/simple-train</a></li>\n<li>Inference : <a href=\"https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large</a></li>\n</ul>",
      "rawMarkdown": "**Muril-Large + Magic**\n\nFirst of all, I would like to thank Kaggle and the competition host for hosting this challenge.\nThe third place came as a surprise to me and I was not expecting it at all, since my public LB score was not high.\nMuril is all I have. I have found that xlm-r is more efficient when finetuning on the squad 2.0 dataset, so I tried it similarly with muril-large.\n\n**Data**\n- competition data, MLQA, XQUAD, SQUAD v2\n\n**Script to finetune muril on squad 2.0:**\n- https://www.kaggle.com/nguyenduongthanh/fine-tune-muril-squad-v2\n\n**Muril-Large-Squad-v2 Checkpoint**\n- https://www.kaggle.com/nguyenduongthanh/muril-large-squad-v2\n\n**Fine-tuning:**\nI just fine-tune on 2 epochs each fold\n- MuRIL Large 10 folds: max-length=384, doc_stride=128, lr=1e-5, gradient accumulation=1\n\n**Ensemble**\n- Simple average of 10 folds\n\n**Notebook**\n- Train: https://www.kaggle.com/nguyenduongthanh/simple-train\n- Inference : https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large",
      "votes": 31
    },
    {
      "id": 1588065,
      "postDate": "2021-11-19T07:32:20.563Z",
      "content": "<p>Amazing magic!</p>",
      "rawMarkdown": "Amazing magic!",
      "votes": 3
    },
    {
      "id": 1583734,
      "postDate": "2021-11-16T03:04:59.913Z",
      "content": "<p>Good job <a href=\"https://www.kaggle.com/nguyenduongthanh\" target=\"_blank\">@nguyenduongthanh</a> and team, congrats on results!</p>",
      "rawMarkdown": "Good job @nguyenduongthanh and team, congrats on results!",
      "votes": 4
    },
    {
      "id": 1591210,
      "postDate": "2021-11-22T05:06:24.767Z",
      "content": "<p>Congratulations <a href=\"https://www.kaggle.com/nguyenduongthanh\" target=\"_blank\">@nguyenduongthanh</a> !😄</p>",
      "rawMarkdown": "Congratulations @nguyenduongthanh !😄",
      "votes": 1
    },
    {
      "id": 1590689,
      "postDate": "2021-11-21T15:05:48.797Z",
      "content": "<p>Congratulations on your achievements.</p>\n<p>Congratulations~</p>",
      "rawMarkdown": "Congratulations on your achievements.\n\nCongratulations~",
      "votes": 1
    },
    {
      "id": 1585164,
      "postDate": "2021-11-17T06:11:43.183Z",
      "content": "<p>Congratulations. i'm envious of gold medal.</p>",
      "rawMarkdown": "Congratulations. i'm envious of gold medal.",
      "votes": 2
    },
    {
      "id": 1584814,
      "postDate": "2021-11-16T19:24:54.827Z",
      "content": "<p>Congratulations! MURIL proved to be very effective, it was also the best model in my ensemble. </p>",
      "rawMarkdown": "Congratulations! MURIL proved to be very effective, it was also the best model in my ensemble. ",
      "votes": 2
    },
    {
      "id": 1584467,
      "postDate": "2021-11-16T14:57:21.623Z",
      "content": "<p>Congratulations. Could you give more informations about Muril-Large ? A kernel to share ?</p>",
      "rawMarkdown": "Congratulations. Could you give more informations about Muril-Large ? A kernel to share ?",
      "votes": 2,
      "replies": [
        {
          "id": 1584963,
          "postDate": "2021-11-16T23:07:34.110Z",
          "content": "<p>you can read it here: <a href=\"https://huggingface.co/google/muril-large-cased\" target=\"_blank\">https://huggingface.co/google/muril-large-cased</a></p>",
          "rawMarkdown": "you can read it here: https://huggingface.co/google/muril-large-cased"
        }
      ]
    },
    {
      "id": 1584093,
      "postDate": "2021-11-16T09:12:00.650Z",
      "content": "<p>Congratulations! <br>\nThis is smart.</p>",
      "rawMarkdown": "Congratulations! \nThis is smart.",
      "votes": 2
    },
    {
      "id": 1583987,
      "postDate": "2021-11-16T07:28:43.980Z",
      "content": "<p>Congrats!!!</p>",
      "rawMarkdown": "Congrats!!!",
      "votes": 2
    },
    {
      "id": 1583909,
      "postDate": "2021-11-16T06:25:03.663Z",
      "content": "<p>Simple is best! Congrats! </p>",
      "rawMarkdown": "Simple is best! Congrats! ",
      "votes": 2
    },
    {
      "id": 1583893,
      "postDate": "2021-11-16T06:06:05.933Z",
      "content": "<p>So simple, but so effective! Thank you a lot for sharing the finetuning solution, I will study how it works! </p>",
      "rawMarkdown": "So simple, but so effective! Thank you a lot for sharing the finetuning solution, I will study how it works! ",
      "votes": 2
    },
    {
      "id": 1583859,
      "postDate": "2021-11-16T05:34:40.990Z",
      "content": "<p>This is absolutely wow , well done <a href=\"https://www.kaggle.com/nguyenduongthanh\" target=\"_blank\">@nguyenduongthanh</a> </p>",
      "rawMarkdown": "This is absolutely wow , well done @nguyenduongthanh ",
      "votes": 2
    },
    {
      "id": 1583753,
      "postDate": "2021-11-16T03:31:31.183Z",
      "content": "<p>Congratulations! 👍</p>",
      "rawMarkdown": "Congratulations! 👍",
      "votes": 2
    },
    {
      "id": 1583743,
      "postDate": "2021-11-16T03:20:08.270Z",
      "content": "<p>Congratulations! Was your solution only <code>Muril-Large</code>?</p>",
      "rawMarkdown": "Congratulations! Was your solution only `Muril-Large`?",
      "votes": 2
    },
    {
      "id": 1583706,
      "postDate": "2021-11-16T02:39:50.573Z",
      "content": "<p>i can't believe……….</p>",
      "rawMarkdown": "i can't believe..........",
      "votes": 2
    },
    {
      "id": 1583678,
      "postDate": "2021-11-16T01:58:53.560Z",
      "content": "<p>Congrats for gold….</p>",
      "rawMarkdown": "Congrats for gold....",
      "votes": 2
    },
    {
      "id": 1583676,
      "postDate": "2021-11-16T01:58:20.327Z",
      "content": "<p>What was your CV? Your notebook isn't public</p>",
      "rawMarkdown": "What was your CV? Your notebook isn't public",
      "votes": 2,
      "replies": [
        {
          "id": 1583679,
          "postDate": "2021-11-16T02:00:06.243Z",
          "content": "<p>My notebook here: <a href=\"https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large</a></p>",
          "rawMarkdown": "My notebook here: https://www.kaggle.com/nguyenduongthanh/simple-10-folds-of-muril-large"
        },
        {
          "id": 1583690,
          "postDate": "2021-11-16T02:12:15.233Z",
          "content": "<p>Sorry I meant the fine-tuning one isn't public</p>",
          "rawMarkdown": "Sorry I meant the fine-tuning one isn't public",
          "votes": 1
        },
        {
          "id": 1583700,
          "postDate": "2021-11-16T02:21:46.520Z",
          "content": "<p>Fine-tuning: <a href=\"https://www.kaggle.com/nguyenduongthanh/simple-train\" target=\"_blank\">https://www.kaggle.com/nguyenduongthanh/simple-train</a></p>",
          "rawMarkdown": "Fine-tuning: https://www.kaggle.com/nguyenduongthanh/simple-train"
        }
      ]
    },
    {
      "id": 1583662,
      "postDate": "2021-11-16T01:41:52.630Z",
      "content": "<p>Congrats! Great finish</p>",
      "rawMarkdown": "Congrats! Great finish",
      "votes": 2
    },
    {
      "id": 1583660,
      "postDate": "2021-11-16T01:41:07.180Z",
      "content": "<p>Congratulations!!!!!</p>",
      "rawMarkdown": "Congratulations!!!!!",
      "votes": 2
    },
    {
      "id": 1583649,
      "postDate": "2021-11-16T01:31:25.883Z",
      "content": "<p>Simple but effective solution! Congratulations!</p>",
      "rawMarkdown": "Simple but effective solution! Congratulations!",
      "votes": 2
    },
    {
      "id": 1587618,
      "postDate": "2021-11-18T20:19:14.917Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "complete": true,
  "source": "live",
  "fetched_at": "2026-10-19T16:21:08Z",
  "competition_id": 30060,
  "competition_slug": "chaii-hindi-and-tamil-question-answering",
  "team_id": 7251967,
  "solution_url": "/competitions/chaii-hindi-and-tamil-question-answering/writeups/magic-3rd-place-solution"
}