{
  "id": 164060,
  "title": "How to use the GPU training model provided by kaggle?",
  "url": "/competitions/global-wheat-detection/discussion/164060",
  "author_name": "",
  "post_date": "2020-07-04T14:41:25.367011800Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hi, I'm a begginer. I ran<a href=\"https://www.kaggle.com/orkatz2/yolov5-train\">yolov5-train </a> public notebook, hoping to get model checkpoint to inference. But my GPU quota is not enough, what should I do? </p>",
  "messages": [
    {
      "id": "915197",
      "postDate": "07/04/2020 14:41:25",
      "content": "<p>Hi, I'm a begginer. I ran<a href=\"https://www.kaggle.com/orkatz2/yolov5-train\">yolov5-train </a> public notebook, hoping to get model checkpoint to inference. But my GPU quota is not enough, what should I do? </p>",
      "rawMarkdown": "Hi, I'm a begginer. I ran[yolov5-train ](https://www.kaggle.com/orkatz2/yolov5-train) public notebook, hoping to get model checkpoint to inference. But my GPU quota is not enough, what should I do?",
      "votes": null
    },
    {
      "id": "915374",
      "postDate": "07/04/2020 17:19:31",
      "content": "<p>You will need some GPU quota in order to run the inference commit and there is no way to obtain extra quota during the week. Hence you'll have to wait until the quota resets. Next time just keep an eye on your quota for the week to make sure you don't run out</p>",
      "rawMarkdown": "You will need some GPU quota in order to run the inference commit and there is no way to obtain extra quota during the week. Hence you'll have to wait until the quota resets. Next time just keep an eye on your quota for the week to make sure you don't run out",
      "votes": null
    },
    {
      "id": "915403",
      "postDate": "07/04/2020 17:46:20",
      "content": "<p>what matthew said below, plus you can train in segments and continue training from where you left off once you get your quota back. For the notebook you mentioned you can do a few epochs, save the model weights somewhere, and do the following the next time.</p>\n\n<p><code>!python train.py --img 1024 --batch 2 --epochs 1 --data ../input/configyolo5/wheat0.yaml --cfg ../input/configyolo5/yolov5x.yaml --name yolov5x_fold0 --weights path/best.pt</code></p>\n\n<p>--weights is where you would pass the previous checkpoint and continue from that.</p>\n\n<p>If you want extra GPU hours, you can also continue training using Google Colab for free 12 hours at a time (if you didn't know)</p>",
      "rawMarkdown": "what matthew said below, plus you can train in segments and continue training from where you left off once you get your quota back. For the notebook you mentioned you can do a few epochs, save the model weights somewhere, and do the following the next time.\n\n`!python train.py --img 1024 --batch 2 --epochs 1 --data ../input/configyolo5/wheat0.yaml --cfg ../input/configyolo5/yolov5x.yaml --name yolov5x_fold0 --weights path/best.pt`\n\n--weights is where you would pass the previous checkpoint and continue from that.\n\nIf you want extra GPU hours, you can also continue training using Google Colab for free 12 hours at a time (if you didn't know)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 915374,
      "author_name": "matthewmasters",
      "author_url": "",
      "post_date": "07/04/2020 17:19:31",
      "content": "<p>You will need some GPU quota in order to run the inference commit and there is no way to obtain extra quota during the week. Hence you'll have to wait until the quota resets. Next time just keep an eye on your quota for the week to make sure you don't run out</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 915403,
      "author_name": "jonykarki",
      "author_url": "",
      "post_date": "07/04/2020 17:46:20",
      "content": "<p>what matthew said below, plus you can train in segments and continue training from where you left off once you get your quota back. For the notebook you mentioned you can do a few epochs, save the model weights somewhere, and do the following the next time.</p>\n\n<p><code>!python train.py --img 1024 --batch 2 --epochs 1 --data ../input/configyolo5/wheat0.yaml --cfg ../input/configyolo5/yolov5x.yaml --name yolov5x_fold0 --weights path/best.pt</code></p>\n\n<p>--weights is where you would pass the previous checkpoint and continue from that.</p>\n\n<p>If you want extra GPU hours, you can also continue training using Google Colab for free 12 hours at a time (if you didn't know)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "915197": "Hi, I'm a begginer. I ran[yolov5-train ](https://www.kaggle.com/orkatz2/yolov5-train) public notebook, hoping to get model checkpoint to inference. But my GPU quota is not enough, what should I do?",
    "915374": "You will need some GPU quota in order to run the inference commit and there is no way to obtain extra quota during the week. Hence you'll have to wait until the quota resets. Next time just keep an eye on your quota for the week to make sure you don't run out",
    "915403": "what matthew said below, plus you can train in segments and continue training from where you left off once you get your quota back. For the notebook you mentioned you can do a few epochs, save the model weights somewhere, and do the following the next time.\n\n`!python train.py --img 1024 --batch 2 --epochs 1 --data ../input/configyolo5/wheat0.yaml --cfg ../input/configyolo5/yolov5x.yaml --name yolov5x_fold0 --weights path/best.pt`\n\n--weights is where you would pass the previous checkpoint and continue from that.\n\nIf you want extra GPU hours, you can also continue training using Google Colab for free 12 hours at a time (if you didn't know)"
  },
  "source": "meta"
}