{
  "id": 264041,
  "title": "still need more GPU/RAM to do the job",
  "url": "/competitions/seti-breakthrough-listen/discussion/264041",
  "author_name": "dragon zhang",
  "post_date": "2021-08-10T21:22:08.782000",
  "votes": 3,
  "comment_count": 4,
  "views": 0,
  "content": "<p>just back to this competition again.</p>\n<p>I changed the efficientnet b0 to b1 by forking learned image resizing, LB got  0.1 up.</p>\n<p>once tried b5 or higher , OOM.  </p>\n<p>now out of time to modify it to run on colab.</p>",
  "messages": [
    {
      "id": 1465093,
      "postDate": "2021-08-10T21:22:08.783Z",
      "content": "<p>just back to this competition again.</p>\n<p>I changed the efficientnet b0 to b1 by forking learned image resizing, LB got  0.1 up.</p>\n<p>once tried b5 or higher , OOM.  </p>\n<p>now out of time to modify it to run on colab.</p>",
      "rawMarkdown": "just back to this competition again.\n\nI changed the efficientnet b0 to b1 by forking learned image resizing, LB got  0.1 up.\n\nonce tried b5 or higher , OOM.  \n\n now out of time to modify it to run on colab.\n\n",
      "votes": 3
    },
    {
      "id": 1468134,
      "postDate": "2021-08-12T09:16:44.430Z",
      "content": "<p>Facing the same problem, and Colab won't have enough disks.</p>\n<p>I tried to train only one fold at a time but it still takes longer than 9hrs if epochs &gt; 30, besides 9hrs * 5folds =45hrs &gt; GPU Quota😭</p>",
      "rawMarkdown": "Facing the same problem, and Colab won't have enough disks.\n\nI tried to train only one fold at a time but it still takes longer than 9hrs if epochs > 30, besides 9hrs * 5folds =45hrs > GPU Quota😭",
      "votes": 1
    },
    {
      "id": 1473027,
      "postDate": "2021-08-15T09:50:24.283Z",
      "content": "<p>I think so. I still need a lot of GPU.   Collab pro hasn't enough disk.. 😂</p>",
      "rawMarkdown": "I think so. I still need a lot of GPU.   Collab pro hasn't enough disk.. 😂"
    },
    {
      "id": 1465760,
      "postDate": "2021-08-11T07:11:43.337Z",
      "content": "<p>no time left .</p>",
      "rawMarkdown": "no time left ."
    },
    {
      "id": 1465145,
      "postDate": "2021-08-10T22:06:19.023Z",
      "content": "<p>How about using TPU?</p>",
      "rawMarkdown": "How about using TPU?"
    }
  ],
  "comments": [
    {
      "id": 1468134,
      "author_name": "Bingliang Li",
      "author_url": "",
      "post_date": "2021-08-12T09:16:44.430000",
      "content": "<p>Facing the same problem, and Colab won't have enough disks.</p>\n<p>I tried to train only one fold at a time but it still takes longer than 9hrs if epochs &gt; 30, besides 9hrs * 5folds =45hrs &gt; GPU Quota😭</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1473027,
      "author_name": "tensor choko",
      "author_url": "",
      "post_date": "2021-08-15T09:50:24.283000",
      "content": "<p>I think so. I still need a lot of GPU.   Collab pro hasn't enough disk.. 😂</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1465760,
      "author_name": "dragon zhang",
      "author_url": "",
      "post_date": "2021-08-11T07:11:43.337000",
      "content": "<p>no time left .</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1465145,
      "author_name": "Araik Tamazian",
      "author_url": "",
      "post_date": "2021-08-10T22:06:19.023000",
      "content": "<p>How about using TPU?</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1465093": "just back to this competition again.\n\nI changed the efficientnet b0 to b1 by forking learned image resizing, LB got  0.1 up.\n\nonce tried b5 or higher , OOM.  \n\n now out of time to modify it to run on colab.\n\n",
    "1468134": "Facing the same problem, and Colab won't have enough disks.\n\nI tried to train only one fold at a time but it still takes longer than 9hrs if epochs > 30, besides 9hrs * 5folds =45hrs > GPU Quota😭",
    "1473027": "I think so. I still need a lot of GPU.   Collab pro hasn't enough disk.. 😂",
    "1465760": "no time left .",
    "1465145": "How about using TPU?"
  }
}