{
  "id": 436835,
  "title": "Direct inference with Meta Seamless M4T large",
  "url": "/competitions/bengaliai-speech/discussion/436835",
  "author_name": "",
  "post_date": "2023-09-04T10:14:27.566431800Z",
  "votes": 6,
  "comment_count": 7,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>I released a <a href=\"https://www.kaggle.com/code/baohaoliao/bengali-m4t-submission\" target=\"_blank\">notebook</a> about direct inference with the newly released Meta model, Seamless M4T.</p>\n<p>I also did inference with the medium model, the results are:</p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>LB score</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Medium</td>\n<td>0.629</td>\n</tr>\n<tr>\n<td>Large</td>\n<td>0.632</td>\n</tr>\n</tbody>\n</table>\n<p>The results from the pre-trained models are not good. One reason might be that I didn't do any post-processing. </p>\n<p><strong>Be careful to use this model, since its license is cc-by-nc 4.0</strong></p>",
  "messages": [
    {
      "id": "2422887",
      "postDate": "09/04/2023 10:14:27",
      "content": "<p>Hi,</p>\n<p>I released a <a href=\"https://www.kaggle.com/code/baohaoliao/bengali-m4t-submission\" target=\"_blank\">notebook</a> about direct inference with the newly released Meta model, Seamless M4T.</p>\n<p>I also did inference with the medium model, the results are:</p>\n<table>\n<thead>\n<tr>\n<th>Model</th>\n<th>LB score</th>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>Medium</td>\n<td>0.629</td>\n</tr>\n<tr>\n<td>Large</td>\n<td>0.632</td>\n</tr>\n</tbody>\n</table>\n<p>The results from the pre-trained models are not good. One reason might be that I didn't do any post-processing. </p>\n<p><strong>Be careful to use this model, since its license is cc-by-nc 4.0</strong></p>",
      "rawMarkdown": "Hi,\n\nI released a [notebook](https://www.kaggle.com/code/baohaoliao/bengali-m4t-submission) about direct inference with the newly released Meta model, Seamless M4T.\n\nI also did inference with the medium model, the results are:\n| Model | LB score  |\n| --- | --- |\n| Medium | 0.629 |\n|  Large | 0.632 |\n\nThe results from the pre-trained models are not good. One reason might be that I didn't do any post-processing. \n\n**Be careful to use this model, since its license is cc-by-nc 4.0**",
      "votes": null
    },
    {
      "id": "2423018",
      "postDate": "09/04/2023 12:02:45",
      "content": "<p>Nice, but atleast it correlates as expected with size</p>",
      "rawMarkdown": "Nice, but atleast it correlates as expected with size",
      "votes": null
    },
    {
      "id": "2423026",
      "postDate": "09/04/2023 12:05:46",
      "content": "<p>not really, the smaller model, medium, achieves better</p>",
      "rawMarkdown": "not really, the smaller model, medium, achieves better",
      "votes": null
    },
    {
      "id": "2423208",
      "postDate": "09/04/2023 14:03:51",
      "content": "<p><a href=\"https://www.kaggle.com/imtiazprio\" target=\"_blank\">@imtiazprio</a> Is it allowed to use this model? Since its license is cc-by-nc 4.0</p>",
      "rawMarkdown": "imtiazprio Is it allowed to use this model? Since its license is cc-by-nc 4.0",
      "votes": null
    },
    {
      "id": "2423219",
      "postDate": "09/04/2023 14:12:18",
      "content": "<p>Codes with non-commercial licenses will not be eligible for competition prizes sadly :(</p>",
      "rawMarkdown": "Codes with non-commercial licenses will not be eligible for competition prizes sadly :(",
      "votes": null
    },
    {
      "id": "2424268",
      "postDate": "09/05/2023 06:54:32",
      "content": "<p>Thanks for your contribution <a href=\"https://www.kaggle.com/baohaoliao\" target=\"_blank\">@baohaoliao</a> </p>",
      "rawMarkdown": "Thanks for your contribution @baohaoliao",
      "votes": null
    },
    {
      "id": "2424476",
      "postDate": "09/05/2023 09:52:34",
      "content": "<p>Thank you for sharing your experiments! </p>",
      "rawMarkdown": "Thank you for sharing your experiments!",
      "votes": null
    },
    {
      "id": "2434343",
      "postDate": "09/12/2023 08:37:45",
      "content": "<p>good notebook ty for sharing</p>",
      "rawMarkdown": "good notebook ty for sharing",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2423018,
      "author_name": "botdeveloper11",
      "author_url": "",
      "post_date": "09/04/2023 12:02:45",
      "content": "<p>Nice, but atleast it correlates as expected with size</p>",
      "votes": null,
      "replies": [
        {
          "id": 2423026,
          "author_name": "baohaoliao",
          "author_url": "",
          "post_date": "09/04/2023 12:05:46",
          "content": "<p>not really, the smaller model, medium, achieves better</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2423208,
      "author_name": "baohaoliao",
      "author_url": "",
      "post_date": "09/04/2023 14:03:51",
      "content": "<p><a href=\"https://www.kaggle.com/imtiazprio\" target=\"_blank\">@imtiazprio</a> Is it allowed to use this model? Since its license is cc-by-nc 4.0</p>",
      "votes": null,
      "replies": [
        {
          "id": 2423219,
          "author_name": "imtiazprio",
          "author_url": "",
          "post_date": "09/04/2023 14:12:18",
          "content": "<p>Codes with non-commercial licenses will not be eligible for competition prizes sadly :(</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2424268,
      "author_name": "aifahim",
      "author_url": "",
      "post_date": "09/05/2023 06:54:32",
      "content": "<p>Thanks for your contribution <a href=\"https://www.kaggle.com/baohaoliao\" target=\"_blank\">@baohaoliao</a> </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2424476,
      "author_name": "snnclsr",
      "author_url": "",
      "post_date": "09/05/2023 09:52:34",
      "content": "<p>Thank you for sharing your experiments! </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2434343,
      "author_name": "thomcat",
      "author_url": "",
      "post_date": "09/12/2023 08:37:45",
      "content": "<p>good notebook ty for sharing</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2422887": "Hi,\n\nI released a [notebook](https://www.kaggle.com/code/baohaoliao/bengali-m4t-submission) about direct inference with the newly released Meta model, Seamless M4T.\n\nI also did inference with the medium model, the results are:\n| Model | LB score  |\n| --- | --- |\n| Medium | 0.629 |\n|  Large | 0.632 |\n\nThe results from the pre-trained models are not good. One reason might be that I didn't do any post-processing. \n\n**Be careful to use this model, since its license is cc-by-nc 4.0**",
    "2423018": "Nice, but atleast it correlates as expected with size",
    "2423026": "not really, the smaller model, medium, achieves better",
    "2423208": "imtiazprio Is it allowed to use this model? Since its license is cc-by-nc 4.0",
    "2423219": "Codes with non-commercial licenses will not be eligible for competition prizes sadly :(",
    "2424268": "Thanks for your contribution @baohaoliao",
    "2424476": "Thank you for sharing your experiments!",
    "2434343": "good notebook ty for sharing"
  },
  "source": "meta"
}