{
  "id": 133015,
  "title": "[Solved] Input dataset size limit + Pytorch 1.4 - Update Notebook/Smaller whl?",
  "url": "/competitions/deepfake-detection-challenge/discussion/133015",
  "author_name": "",
  "post_date": "2020-02-29T09:01:02.077050200Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>EDIT: Downgraded fastai2 and fastcore to pre-Feb 17th version and it all works fine]</p>\n\n<hr>\n\n<p>Just wondering if anyone is using pytorch 1.4 for their models, I have a dependency (fastai2) that requires 1.4 however the notebooks come with 1.3 by default.</p>\n\n<p>Right now to use 1.4 I'm loading a .whl in a dataset I built from a local Pytorch 1.4 repo which works fine, however the file takes up 700+mb and pushes me over the 1024mb inputs dataset limit we have when it comes to making submissions :(</p>\n\n<p>Does anyone have experience getting around this? I'm not super familiar with building .whl's, if I randomly delete things from the Pytorch repo that don't seem important and rebuild the wheel will that work?? </p>\n\n<p>Alternatively is there any other way to use 1.4 in this no internet competition? Any help would be super appreciated!!</p>",
  "messages": [
    {
      "id": "759616",
      "postDate": "02/29/2020 09:01:02",
      "content": "<p>EDIT: Downgraded fastai2 and fastcore to pre-Feb 17th version and it all works fine]</p>\n\n<hr>\n\n<p>Just wondering if anyone is using pytorch 1.4 for their models, I have a dependency (fastai2) that requires 1.4 however the notebooks come with 1.3 by default.</p>\n\n<p>Right now to use 1.4 I'm loading a .whl in a dataset I built from a local Pytorch 1.4 repo which works fine, however the file takes up 700+mb and pushes me over the 1024mb inputs dataset limit we have when it comes to making submissions :(</p>\n\n<p>Does anyone have experience getting around this? I'm not super familiar with building .whl's, if I randomly delete things from the Pytorch repo that don't seem important and rebuild the wheel will that work?? </p>\n\n<p>Alternatively is there any other way to use 1.4 in this no internet competition? Any help would be super appreciated!!</p>",
      "rawMarkdown": "EDIT: Downgraded fastai2 and fastcore to pre-Feb 17th version and it all works fine]\n\n-----\nJust wondering if anyone is using pytorch 1.4 for their models, I have a dependency (fastai2) that requires 1.4 however the notebooks come with 1.3 by default.\n\nRight now to use 1.4 I'm loading a .whl in a dataset I built from a local Pytorch 1.4 repo which works fine, however the file takes up 700+mb and pushes me over the 1024mb inputs dataset limit we have when it comes to making submissions :(\n\nDoes anyone have experience getting around this? I'm not super familiar with building .whl's, if I randomly delete things from the Pytorch repo that don't seem important and rebuild the wheel will that work?? \n\nAlternatively is there any other way to use 1.4 in this no internet competition? Any help would be super appreciated!!",
      "votes": null
    },
    {
      "id": "759638",
      "postDate": "02/29/2020 09:08:09",
      "content": "<p>I'm using fastai2 with pytorch 1.3 without any issue (so far) so maybe that's an option for you?</p>",
      "rawMarkdown": "I'm using fastai2 with pytorch 1.3 without any issue (so far) so maybe that's an option for you?",
      "votes": null
    },
    {
      "id": "759799",
      "postDate": "02/29/2020 13:30:48",
      "content": "<p>What version are you using? I’m on 0.11, maybe I should downgrade</p>",
      "rawMarkdown": "What version are you using? I’m on 0.11, maybe I should downgrade",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 759638,
      "author_name": "davecoates",
      "author_url": "",
      "post_date": "02/29/2020 09:08:09",
      "content": "<p>I'm using fastai2 with pytorch 1.3 without any issue (so far) so maybe that's an option for you?</p>",
      "votes": null,
      "replies": [
        {
          "id": 759799,
          "author_name": "morganmcg",
          "author_url": "",
          "post_date": "02/29/2020 13:30:48",
          "content": "<p>What version are you using? I’m on 0.11, maybe I should downgrade</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "759616": "EDIT: Downgraded fastai2 and fastcore to pre-Feb 17th version and it all works fine]\n\n-----\nJust wondering if anyone is using pytorch 1.4 for their models, I have a dependency (fastai2) that requires 1.4 however the notebooks come with 1.3 by default.\n\nRight now to use 1.4 I'm loading a .whl in a dataset I built from a local Pytorch 1.4 repo which works fine, however the file takes up 700+mb and pushes me over the 1024mb inputs dataset limit we have when it comes to making submissions :(\n\nDoes anyone have experience getting around this? I'm not super familiar with building .whl's, if I randomly delete things from the Pytorch repo that don't seem important and rebuild the wheel will that work?? \n\nAlternatively is there any other way to use 1.4 in this no internet competition? Any help would be super appreciated!!",
    "759638": "I'm using fastai2 with pytorch 1.3 without any issue (so far) so maybe that's an option for you?",
    "759799": "What version are you using? I’m on 0.11, maybe I should downgrade"
  },
  "source": "meta"
}