{
  "id": 671282,
  "title": "Submission file too large (10GB+) – timeout / out of disk error",
  "url": "/competitions/cafa-6-protein-function-prediction/discussion/671282",
  "author_name": "",
  "post_date": "2026-02-01T04:28:32.904460900Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hello CAFA6 organizers,</p>\n<p>I am trying to submit a submission.tsv file, but the file size becomes larger than 10GB.\nWhen uploading the file, I encounter errors such as \"Timeout exceeded\" or \"Out of disk\".</p>\n<p>Is there an official or recommended way to handle very large submission files?\nFor example:</p>\n<ul>\n<li>Is it allowed to split the submission into multiple parts?</li>\n<li>Can we submit a compressed file (e.g., gzip)?</li>\n<li>Or is there an expected strategy such as applying a cutoff / threshold on predicted GO terms?</li>\n</ul>\n<p>Any guidance on how to properly submit large-scale predictions would be greatly appreciated.</p>\n<p>Thank you.</p>",
  "messages": [
    {
      "id": "3400174",
      "postDate": "02/01/2026 04:28:32",
      "content": "<p>Hello CAFA6 organizers,</p>\n<p>I am trying to submit a submission.tsv file, but the file size becomes larger than 10GB.\nWhen uploading the file, I encounter errors such as \"Timeout exceeded\" or \"Out of disk\".</p>\n<p>Is there an official or recommended way to handle very large submission files?\nFor example:</p>\n<ul>\n<li>Is it allowed to split the submission into multiple parts?</li>\n<li>Can we submit a compressed file (e.g., gzip)?</li>\n<li>Or is there an expected strategy such as applying a cutoff / threshold on predicted GO terms?</li>\n</ul>\n<p>Any guidance on how to properly submit large-scale predictions would be greatly appreciated.</p>\n<p>Thank you.</p>",
      "rawMarkdown": "Hello CAFA6 organizers,\n\nI am trying to submit a submission.tsv file, but the file size becomes larger than 10GB.\nWhen uploading the file, I encounter errors such as \"Timeout exceeded\" or \"Out of disk\".\n\nIs there an official or recommended way to handle very large submission files?\nFor example:\n- Is it allowed to split the submission into multiple parts?\n- Can we submit a compressed file (e.g., gzip)?\n- Or is there an expected strategy such as applying a cutoff / threshold on predicted GO terms?\n\nAny guidance on how to properly submit large-scale predictions would be greatly appreciated.\n\nThank you.",
      "votes": null
    },
    {
      "id": "3400416",
      "postDate": "02/01/2026 14:13:16",
      "content": "<p>My files also exceed 10GB. Would it be okay to upload them separately as three files (CCO, MFO, and BPO) without affecting the evaluation?</p>",
      "rawMarkdown": "My files also exceed 10GB. Would it be okay to upload them separately as three files (CCO, MFO, and BPO) without affecting the evaluation?",
      "votes": null
    },
    {
      "id": "3400455",
      "postDate": "02/01/2026 15:29:53",
      "content": "<p>Did you limit to 1500 GO terms for each protein in the testsuperset.fasta? You can also decrease the number of significant digits which helps decrease the file size. Another strategy is to only submit the child term predictions and omit the parent terms, since all the predictions will be propagated based on the given GO graph. After propagation, the parent terms will get the score of the child terms if no scores for the parent terms are found.</p>",
      "rawMarkdown": "Did you limit to 1500 GO terms for each protein in the testsuperset.fasta? You can also decrease the number of significant digits which helps decrease the file size. Another strategy is to only submit the child term predictions and omit the parent terms, since all the predictions will be propagated based on the given GO graph. After propagation, the parent terms will get the score of the child terms if no scores for the parent terms are found.",
      "votes": null
    },
    {
      "id": "3400610",
      "postDate": "02/01/2026 23:19:55",
      "content": "<p>Reply to An Phan:\nThank you for the helpful suggestions.\nRegarding the third strategy (submitting only child terms), our model directly predicts both parent and child GO terms simultaneously, rather than predicting only child terms and relying on propagation. Therefore, this approach may not be applicable to our case.\nWe will try reducing the number of significant digits and limiting the GO terms per protein. However, could you please clarify what the maximum allowed file size is for submission? Knowing the exact limit would help us determine how aggressively we need to optimize our submission file.\nAdditionally, I previously asked whether it would be acceptable to split the submission into three separate files (CCO, MFO, and BPO). Could you confirm if this is a valid submission format?\nThank you for your guidance.</p>",
      "rawMarkdown": "Reply to An Phan:\nThank you for the helpful suggestions.\nRegarding the third strategy (submitting only child terms), our model directly predicts both parent and child GO terms simultaneously, rather than predicting only child terms and relying on propagation. Therefore, this approach may not be applicable to our case.\nWe will try reducing the number of significant digits and limiting the GO terms per protein. However, could you please clarify what the maximum allowed file size is for submission? Knowing the exact limit would help us determine how aggressively we need to optimize our submission file.\nAdditionally, I previously asked whether it would be acceptable to split the submission into three separate files (CCO, MFO, and BPO). Could you confirm if this is a valid submission format?\nThank you for your guidance.",
      "votes": null
    },
    {
      "id": "3400611",
      "postDate": "02/01/2026 23:29:16",
      "content": "<p>We have tested with a file size of 10GB and it took around an hour to finish scoring on the leaderboard, so I would recommend file sizes below 10GB. 10GB is also the estimated upper bound for submission file if every protein is predicted with 1500 terms and 3 significant figures for the score. </p>\n<p>All three ontologies (CCO, MFO, and BPO) should be combined in one prediction file since the current parser only takes in one file.</p>",
      "rawMarkdown": "We have tested with a file size of 10GB and it took around an hour to finish scoring on the leaderboard, so I would recommend file sizes below 10GB. 10GB is also the estimated upper bound for submission file if every protein is predicted with 1500 terms and 3 significant figures for the score. \n\nAll three ontologies (CCO, MFO, and BPO) should be combined in one prediction file since the current parser only takes in one file.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3400416,
      "author_name": "sdkparkforbi",
      "author_url": "",
      "post_date": "02/01/2026 14:13:16",
      "content": "<p>My files also exceed 10GB. Would it be okay to upload them separately as three files (CCO, MFO, and BPO) without affecting the evaluation?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3400455,
      "author_name": "ahphan",
      "author_url": "",
      "post_date": "02/01/2026 15:29:53",
      "content": "<p>Did you limit to 1500 GO terms for each protein in the testsuperset.fasta? You can also decrease the number of significant digits which helps decrease the file size. Another strategy is to only submit the child term predictions and omit the parent terms, since all the predictions will be propagated based on the given GO graph. After propagation, the parent terms will get the score of the child terms if no scores for the parent terms are found.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3400610,
          "author_name": "sdkparkforbi",
          "author_url": "",
          "post_date": "02/01/2026 23:19:55",
          "content": "<p>Reply to An Phan:\nThank you for the helpful suggestions.\nRegarding the third strategy (submitting only child terms), our model directly predicts both parent and child GO terms simultaneously, rather than predicting only child terms and relying on propagation. Therefore, this approach may not be applicable to our case.\nWe will try reducing the number of significant digits and limiting the GO terms per protein. However, could you please clarify what the maximum allowed file size is for submission? Knowing the exact limit would help us determine how aggressively we need to optimize our submission file.\nAdditionally, I previously asked whether it would be acceptable to split the submission into three separate files (CCO, MFO, and BPO). Could you confirm if this is a valid submission format?\nThank you for your guidance.</p>",
          "votes": null,
          "replies": [
            {
              "id": 3400611,
              "author_name": "ahphan",
              "author_url": "",
              "post_date": "02/01/2026 23:29:16",
              "content": "<p>We have tested with a file size of 10GB and it took around an hour to finish scoring on the leaderboard, so I would recommend file sizes below 10GB. 10GB is also the estimated upper bound for submission file if every protein is predicted with 1500 terms and 3 significant figures for the score. </p>\n<p>All three ontologies (CCO, MFO, and BPO) should be combined in one prediction file since the current parser only takes in one file.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3400174": "Hello CAFA6 organizers,\n\nI am trying to submit a submission.tsv file, but the file size becomes larger than 10GB.\nWhen uploading the file, I encounter errors such as \"Timeout exceeded\" or \"Out of disk\".\n\nIs there an official or recommended way to handle very large submission files?\nFor example:\n- Is it allowed to split the submission into multiple parts?\n- Can we submit a compressed file (e.g., gzip)?\n- Or is there an expected strategy such as applying a cutoff / threshold on predicted GO terms?\n\nAny guidance on how to properly submit large-scale predictions would be greatly appreciated.\n\nThank you.",
    "3400416": "My files also exceed 10GB. Would it be okay to upload them separately as three files (CCO, MFO, and BPO) without affecting the evaluation?",
    "3400455": "Did you limit to 1500 GO terms for each protein in the testsuperset.fasta? You can also decrease the number of significant digits which helps decrease the file size. Another strategy is to only submit the child term predictions and omit the parent terms, since all the predictions will be propagated based on the given GO graph. After propagation, the parent terms will get the score of the child terms if no scores for the parent terms are found.",
    "3400610": "Reply to An Phan:\nThank you for the helpful suggestions.\nRegarding the third strategy (submitting only child terms), our model directly predicts both parent and child GO terms simultaneously, rather than predicting only child terms and relying on propagation. Therefore, this approach may not be applicable to our case.\nWe will try reducing the number of significant digits and limiting the GO terms per protein. However, could you please clarify what the maximum allowed file size is for submission? Knowing the exact limit would help us determine how aggressively we need to optimize our submission file.\nAdditionally, I previously asked whether it would be acceptable to split the submission into three separate files (CCO, MFO, and BPO). Could you confirm if this is a valid submission format?\nThank you for your guidance.",
    "3400611": "We have tested with a file size of 10GB and it took around an hour to finish scoring on the leaderboard, so I would recommend file sizes below 10GB. 10GB is also the estimated upper bound for submission file if every protein is predicted with 1500 terms and 3 significant figures for the score. \n\nAll three ontologies (CCO, MFO, and BPO) should be combined in one prediction file since the current parser only takes in one file."
  },
  "source": "meta"
}