{
  "id": 610543,
  "title": "GPU-time for submissions",
  "url": "/competitions/rsna-intracranial-aneurysm-detection/discussion/610543",
  "author_name": "thomas rost",
  "post_date": "2025-10-04T10:38:47.988000",
  "votes": 1,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Is the time it takes to compute the predictions on the hidden test set actually deducted from the users' GPU-time quota?</p>",
  "messages": [
    {
      "id": 3298027,
      "postDate": "2025-10-04T10:47:03.730Z",
      "content": "<p>No - you submit once to the dummy LB and this time is deducted from your quota. The actual rerun is separate and out of your quota <a href=\"https://www.kaggle.com/thalro\" target=\"_blank\">@thalro</a> </p>",
      "rawMarkdown": "No - you submit once to the dummy LB and this time is deducted from your quota. The actual rerun is separate and out of your quota @thalro ",
      "votes": 3,
      "replies": [
        {
          "id": 3298144,
          "postDate": "2025-10-04T16:12:20.797Z",
          "content": "<p>Ok, thanks!</p>",
          "rawMarkdown": "Ok, thanks!",
          "votes": 1,
          "replies": [
            {
              "id": 3298146,
              "postDate": "2025-10-04T16:25:47.557Z",
              "content": "<p><a href=\"https://www.kaggle.com/thalro\" target=\"_blank\">@thalro</a> I can give you 1 tip here - <br>\nuse <code>os.environ[\"IS_COMPETITION_RERUN\"]</code> variable - if this is false, then directly submit the sample submission and close the process in 1-2 seconds. Else run the process (out of your GPU quota) </p>",
              "rawMarkdown": "@thalro I can give you 1 tip here - \nuse `os.environ[\"IS_COMPETITION_RERUN\"]` variable - if this is false, then directly submit the sample submission and close the process in 1-2 seconds. Else run the process (out of your GPU quota) "
            }
          ]
        }
      ]
    },
    {
      "id": 3298024,
      "postDate": "2025-10-04T10:44:52.040Z",
      "content": "<p>No. Is not.</p>\n<p>EDIT: But the save before score needs to be triggered and needs GPU quota remaining.</p>",
      "rawMarkdown": "No. Is not.\n\nEDIT: But the save before score needs to be triggered and needs GPU quota remaining.",
      "votes": 1,
      "replies": [
        {
          "id": 3298145,
          "postDate": "2025-10-04T16:12:43.983Z",
          "content": "<p>Thanks, that‘s good to know!</p>",
          "rawMarkdown": "Thanks, that‘s good to know!"
        }
      ]
    },
    {
      "id": 3298022,
      "postDate": "2025-10-04T10:38:47.987Z",
      "content": "<p>Is the time it takes to compute the predictions on the hidden test set actually deducted from the users' GPU-time quota?</p>",
      "rawMarkdown": "Is the time it takes to compute the predictions on the hidden test set actually deducted from the users' GPU-time quota?",
      "votes": 1
    },
    {
      "id": 3298602,
      "postDate": "2025-10-06T00:36:43.797Z",
      "content": "<p>Is the GPU actually faster for inference? I thought it wouldn't give speed up because predictions are one at a time.</p>",
      "rawMarkdown": "Is the GPU actually faster for inference? I thought it wouldn't give speed up because predictions are one at a time."
    }
  ],
  "comments": [
    {
      "id": 3298027,
      "author_name": "Ravi Ramakrishnan",
      "author_url": "",
      "post_date": "2025-10-04T10:47:03.730000",
      "content": "<p>No - you submit once to the dummy LB and this time is deducted from your quota. The actual rerun is separate and out of your quota <a href=\"https://www.kaggle.com/thalro\" target=\"_blank\">@thalro</a> </p>",
      "votes": 3,
      "replies": [
        {
          "id": 3298144,
          "author_name": "thomas rost",
          "author_url": "",
          "post_date": "2025-10-04T16:12:20.797000",
          "content": "<p>Ok, thanks!</p>",
          "votes": 1,
          "replies": [
            {
              "id": 3298146,
              "author_name": "Ravi Ramakrishnan",
              "author_url": "",
              "post_date": "2025-10-04T16:25:47.557000",
              "content": "<p><a href=\"https://www.kaggle.com/thalro\" target=\"_blank\">@thalro</a> I can give you 1 tip here - <br>\nuse <code>os.environ[\"IS_COMPETITION_RERUN\"]</code> variable - if this is false, then directly submit the sample submission and close the process in 1-2 seconds. Else run the process (out of your GPU quota) </p>",
              "votes": 0,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 3298024,
      "author_name": "Ángel Jacinto Sánchez Ruiz",
      "author_url": "",
      "post_date": "2025-10-04T10:44:52.040000",
      "content": "<p>No. Is not.</p>\n<p>EDIT: But the save before score needs to be triggered and needs GPU quota remaining.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 3298145,
          "author_name": "thomas rost",
          "author_url": "",
          "post_date": "2025-10-04T16:12:43.983000",
          "content": "<p>Thanks, that‘s good to know!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 3298602,
      "author_name": "Ravon",
      "author_url": "",
      "post_date": "2025-10-06T00:36:43.797000",
      "content": "<p>Is the GPU actually faster for inference? I thought it wouldn't give speed up because predictions are one at a time.</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3298027": "No - you submit once to the dummy LB and this time is deducted from your quota. The actual rerun is separate and out of your quota @thalro ",
    "3298024": "No. Is not.\n\nEDIT: But the save before score needs to be triggered and needs GPU quota remaining.",
    "3298022": "Is the time it takes to compute the predictions on the hidden test set actually deducted from the users' GPU-time quota?",
    "3298602": "Is the GPU actually faster for inference? I thought it wouldn't give speed up because predictions are one at a time."
  }
}