{
  "id": 71479,
  "title": "Pytorch config torch.backends.cudnn.benchmark",
  "url": "/competitions/quickdraw-doodle-recognition/discussion/71479",
  "author_name": "Finlay",
  "post_date": "2018-11-14T02:36:57.267000",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p><a href=\"https://discuss.pytorch.org/t/what-does-torch-backends-cudnn-benchmark-do/5936\">https://discuss.pytorch.org/t/what-does-torch-backends-cudnn-benchmark-do/5936</a></p>\n\n<p><strong>It depends on the task. If your input size is changing a lot, then it might hurt runtime, if not, it should be much faster.</strong></p>\n\n<p>so you should set it false if your input image size do not match the original model. </p>",
  "messages": [
    {
      "id": 420709,
      "postDate": "2018-11-14T02:36:57.267Z",
      "content": "<p><a href=\"https://discuss.pytorch.org/t/what-does-torch-backends-cudnn-benchmark-do/5936\">https://discuss.pytorch.org/t/what-does-torch-backends-cudnn-benchmark-do/5936</a></p>\n\n<p><strong>It depends on the task. If your input size is changing a lot, then it might hurt runtime, if not, it should be much faster.</strong></p>\n\n<p>so you should set it false if your input image size do not match the original model. </p>",
      "rawMarkdown": "https://discuss.pytorch.org/t/what-does-torch-backends-cudnn-benchmark-do/5936\n\n**It depends on the task. If your input size is changing a lot, then it might hurt runtime, if not, it should be much faster.**\n\nso you should set it false if your input image size do not match the original model. \n",
      "votes": 4
    },
    {
      "id": 456238,
      "postDate": "2019-01-15T11:55:57.760Z",
      "content": "<p>How the input size can change? The model has a predefined input size (e.g. nn.Linear(10, 10)).\nAm I missing something?</p>",
      "rawMarkdown": "How the input size can change? The model has a predefined input size (e.g. nn.Linear(10, 10)).\nAm I missing something?\n"
    },
    {
      "id": 420873,
      "postDate": "2018-11-14T09:12:24.430Z",
      "content": "<p>Thanks!</p>",
      "rawMarkdown": "Thanks!"
    }
  ],
  "comments": [
    {
      "id": 456238,
      "author_name": "Andreas Kouzelis",
      "author_url": "",
      "post_date": "2019-01-15T11:55:57.760000",
      "content": "<p>How the input size can change? The model has a predefined input size (e.g. nn.Linear(10, 10)).\nAm I missing something?</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 420873,
      "author_name": "RW",
      "author_url": "",
      "post_date": "2018-11-14T09:12:24.430000",
      "content": "<p>Thanks!</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "420709": "https://discuss.pytorch.org/t/what-does-torch-backends-cudnn-benchmark-do/5936\n\n**It depends on the task. If your input size is changing a lot, then it might hurt runtime, if not, it should be much faster.**\n\nso you should set it false if your input image size do not match the original model. \n",
    "456238": "How the input size can change? The model has a predefined input size (e.g. nn.Linear(10, 10)).\nAm I missing something?\n",
    "420873": "Thanks!"
  }
}