{
  "id": 412865,
  "title": "Training loss NaN",
  "url": "/competitions/google-research-identify-contrails-reduce-global-warming/discussion/412865",
  "author_name": "Arunodhayan",
  "post_date": "2023-05-25T14:46:10.312000",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Did anyone face this , my training loss after some iterations is NaN i used dice and bce loss , it looks weird</p>",
  "messages": [
    {
      "id": 2280627,
      "postDate": "2023-05-30T08:14:00.540Z",
      "content": "<p>amp=True also gives you nans for large Lr</p>",
      "rawMarkdown": "amp=True also gives you nans for large Lr",
      "votes": 1
    },
    {
      "id": 2278212,
      "postDate": "2023-05-28T12:38:09.803Z",
      "content": "<p>You have to add a smoothing to avoid Division by Zero in Dice Loss</p>",
      "rawMarkdown": "You have to add a smoothing to avoid Division by Zero in Dice Loss",
      "votes": 1
    },
    {
      "id": 2274003,
      "postDate": "2023-05-25T14:46:10.313Z",
      "content": "<p>Did anyone face this , my training loss after some iterations is NaN i used dice and bce loss , it looks weird</p>",
      "rawMarkdown": "Did anyone face this , my training loss after some iterations is NaN i used dice and bce loss , it looks weird"
    }
  ],
  "comments": [
    {
      "id": 2280627,
      "author_name": "dmitrykonovalov",
      "author_url": "",
      "post_date": "2023-05-30T08:14:00.540000",
      "content": "<p>amp=True also gives you nans for large Lr</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2278212,
      "author_name": "Patchef",
      "author_url": "",
      "post_date": "2023-05-28T12:38:09.803000",
      "content": "<p>You have to add a smoothing to avoid Division by Zero in Dice Loss</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2280627": "amp=True also gives you nans for large Lr",
    "2278212": "You have to add a smoothing to avoid Division by Zero in Dice Loss",
    "2274003": "Did anyone face this , my training loss after some iterations is NaN i used dice and bce loss , it looks weird"
  }
}