{
  "id": 143750,
  "title": "What is the optimal batch size?",
  "url": "/competitions/flower-classification-with-tpus/discussion/143750",
  "author_name": "",
  "post_date": "2020-04-16T05:30:02.000046400Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1310697%2F9312404c655ae521c2fd3fa156b83124%2FScreenshot_2020-04-16%20Introduction%20kernel%20whats%20EfficientNet%20Kaggle.png?generation=1587014842515439&amp;alt=media\" alt=\"\"></p>\n\n<p>I was getting a resource allocation error when I was training with batch size of 1024 and 256. Yet for bengali AI competition, I saw a public kernel using batch size of 1024 working fine.</p>",
  "messages": [
    {
      "id": "809363",
      "postDate": "04/16/2020 05:30:02",
      "content": "<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1310697%2F9312404c655ae521c2fd3fa156b83124%2FScreenshot_2020-04-16%20Introduction%20kernel%20whats%20EfficientNet%20Kaggle.png?generation=1587014842515439&amp;alt=media\" alt=\"\"></p>\n\n<p>I was getting a resource allocation error when I was training with batch size of 1024 and 256. Yet for bengali AI competition, I saw a public kernel using batch size of 1024 working fine.</p>",
      "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1310697%2F9312404c655ae521c2fd3fa156b83124%2FScreenshot_2020-04-16%20Introduction%20kernel%20whats%20EfficientNet%20Kaggle.png?generation=1587014842515439&amp;alt=media)\n\nI was getting a resource allocation error when I was training with batch size of 1024 and 256. Yet for bengali AI competition, I saw a public kernel using batch size of 1024 working fine.",
      "votes": null
    },
    {
      "id": "810144",
      "postDate": "04/16/2020 18:46:08",
      "content": "<p>Depends on the size of the model and the input images. Here, you are trining on high-res 512x512 images. A batch of those is a lot of RAM.</p>",
      "rawMarkdown": "Depends on the size of the model and the input images. Here, you are trining on high-res 512x512 images. A batch of those is a lot of RAM.",
      "votes": null
    },
    {
      "id": "810626",
      "postDate": "04/17/2020 06:11:39",
      "content": "<p>Thanks for your answer. Now it makes sense</p>",
      "rawMarkdown": "Thanks for your answer. Now it makes sense",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 810144,
      "author_name": "mgorner",
      "author_url": "",
      "post_date": "04/16/2020 18:46:08",
      "content": "<p>Depends on the size of the model and the input images. Here, you are trining on high-res 512x512 images. A batch of those is a lot of RAM.</p>",
      "votes": null,
      "replies": [
        {
          "id": 810626,
          "author_name": "kurianbenoy",
          "author_url": "",
          "post_date": "04/17/2020 06:11:39",
          "content": "<p>Thanks for your answer. Now it makes sense</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "809363": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1310697%2F9312404c655ae521c2fd3fa156b83124%2FScreenshot_2020-04-16%20Introduction%20kernel%20whats%20EfficientNet%20Kaggle.png?generation=1587014842515439&amp;alt=media)\n\nI was getting a resource allocation error when I was training with batch size of 1024 and 256. Yet for bengali AI competition, I saw a public kernel using batch size of 1024 working fine.",
    "810144": "Depends on the size of the model and the input images. Here, you are trining on high-res 512x512 images. A batch of those is a lot of RAM.",
    "810626": "Thanks for your answer. Now it makes sense"
  },
  "source": "meta"
}