{
  "id": 674104,
  "title": "Stuck because of Inference Time for past few days",
  "url": "/competitions/vesuvius-challenge-surface-detection/discussion/674104",
  "author_name": "",
  "post_date": "2026-02-18T18:17:07.254127600Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>sorry for asking a stupid question but I have been stuck on this for few days.</p>\n<p>When using cross-validation with cv = 5, inference runs on five different models, which increases the prediction time to roughly five times that of a single model.</p>\n<p>Is there a way to combine the cross-validation models into a single model to reduce inference time?</p>",
  "messages": [
    {
      "id": "3407608",
      "postDate": "02/18/2026 18:17:07",
      "content": "<p>Hi,</p>\n<p>sorry for asking a stupid question but I have been stuck on this for few days.</p>\n<p>When using cross-validation with cv = 5, inference runs on five different models, which increases the prediction time to roughly five times that of a single model.</p>\n<p>Is there a way to combine the cross-validation models into a single model to reduce inference time?</p>",
      "rawMarkdown": "Hi,\n\nsorry for asking a stupid question but I have been stuck on this for few days.\n\nWhen using cross-validation with cv = 5, inference runs on five different models, which increases the prediction time to roughly five times that of a single model.\n\nIs there a way to combine the cross-validation models into a single model to reduce inference time?",
      "votes": null
    },
    {
      "id": "3407653",
      "postDate": "02/18/2026 21:09:00",
      "content": "<p>No, but you can use T4x2 gpu runtime while inferencing parallelly. Load 2 model on single gpu and other 3 on another gpu, then use workers and make volumes inference sequentially but parallelly on both gpus. This will almost half your runtime.  </p>",
      "rawMarkdown": "No, but you can use T4x2 gpu runtime while inferencing parallelly. Load 2 model on single gpu and other 3 on another gpu, then use workers and make volumes inference sequentially but parallelly on both gpus. This will almost half your runtime.",
      "votes": null
    },
    {
      "id": "3408497",
      "postDate": "02/20/2026 18:49:39",
      "content": "<p>I still get timeout error :( If a single test sample is taking 6 minutes, It shouldnt timeout</p>",
      "rawMarkdown": "I still get timeout error :( If a single test sample is taking 6 minutes, It shouldnt timeout",
      "votes": null
    },
    {
      "id": "3408827",
      "postDate": "02/21/2026 11:56:28",
      "content": "<p>sometimes the same notebook gets timeout, and if you submit it again, it get successfully scored. Try re-submitting it.</p>",
      "rawMarkdown": "sometimes the same notebook gets timeout, and if you submit it again, it get successfully scored. Try re-submitting it.",
      "votes": null
    },
    {
      "id": "3409101",
      "postDate": "02/22/2026 05:21:03",
      "content": "<p>yeah, its not consistent across submissions.</p>",
      "rawMarkdown": "yeah, its not consistent across submissions.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3407653,
      "author_name": "muhammadibrahim3093",
      "author_url": "",
      "post_date": "02/18/2026 21:09:00",
      "content": "<p>No, but you can use T4x2 gpu runtime while inferencing parallelly. Load 2 model on single gpu and other 3 on another gpu, then use workers and make volumes inference sequentially but parallelly on both gpus. This will almost half your runtime.  </p>",
      "votes": null,
      "replies": [
        {
          "id": 3408497,
          "author_name": "ppilania1985",
          "author_url": "",
          "post_date": "02/20/2026 18:49:39",
          "content": "<p>I still get timeout error :( If a single test sample is taking 6 minutes, It shouldnt timeout</p>",
          "votes": null,
          "replies": [
            {
              "id": 3408827,
              "author_name": "muhammadibrahim3093",
              "author_url": "",
              "post_date": "02/21/2026 11:56:28",
              "content": "<p>sometimes the same notebook gets timeout, and if you submit it again, it get successfully scored. Try re-submitting it.</p>",
              "votes": null,
              "replies": [
                {
                  "id": 3409101,
                  "author_name": "ppilania1985",
                  "author_url": "",
                  "post_date": "02/22/2026 05:21:03",
                  "content": "<p>yeah, its not consistent across submissions.</p>",
                  "votes": null,
                  "replies": []
                }
              ]
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3407608": "Hi,\n\nsorry for asking a stupid question but I have been stuck on this for few days.\n\nWhen using cross-validation with cv = 5, inference runs on five different models, which increases the prediction time to roughly five times that of a single model.\n\nIs there a way to combine the cross-validation models into a single model to reduce inference time?",
    "3407653": "No, but you can use T4x2 gpu runtime while inferencing parallelly. Load 2 model on single gpu and other 3 on another gpu, then use workers and make volumes inference sequentially but parallelly on both gpus. This will almost half your runtime.",
    "3408497": "I still get timeout error :( If a single test sample is taking 6 minutes, It shouldnt timeout",
    "3408827": "sometimes the same notebook gets timeout, and if you submit it again, it get successfully scored. Try re-submitting it.",
    "3409101": "yeah, its not consistent across submissions."
  },
  "source": "meta"
}