{
  "id": 146530,
  "title": "Hyperparameter Tuning",
  "url": "/competitions/jigsaw-multilingual-toxic-comment-classification/discussion/146530",
  "author_name": "Amogh ",
  "post_date": "2020-04-27T14:41:46.421000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>I am relatively new to NLP and its my first time training models on TPUs, so I am admittedly struggling to train a large dataset and experiment with it enough so as to perform hyperparameter tuning? Do you use some special techniques for this? \nWhat learning rate schedule are you following, and what is your rationale behind it? How do you decide on a learning rate in the first place?\nPlease feel free to leave comments with any advice that might be helpful, I shall consolidate them for better sharing.</p>\n\n<p>PS. I am using PyTorch XLA</p>",
  "messages": [
    {
      "id": 823293,
      "postDate": "2020-04-27T14:41:46.420Z",
      "content": "<p>I am relatively new to NLP and its my first time training models on TPUs, so I am admittedly struggling to train a large dataset and experiment with it enough so as to perform hyperparameter tuning? Do you use some special techniques for this? \nWhat learning rate schedule are you following, and what is your rationale behind it? How do you decide on a learning rate in the first place?\nPlease feel free to leave comments with any advice that might be helpful, I shall consolidate them for better sharing.</p>\n\n<p>PS. I am using PyTorch XLA</p>",
      "rawMarkdown": "I am relatively new to NLP and its my first time training models on TPUs, so I am admittedly struggling to train a large dataset and experiment with it enough so as to perform hyperparameter tuning? Do you use some special techniques for this? \nWhat learning rate schedule are you following, and what is your rationale behind it? How do you decide on a learning rate in the first place?\nPlease feel free to leave comments with any advice that might be helpful, I shall consolidate them for better sharing.\n\nPS. I am using PyTorch XLA",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "823293": "I am relatively new to NLP and its my first time training models on TPUs, so I am admittedly struggling to train a large dataset and experiment with it enough so as to perform hyperparameter tuning? Do you use some special techniques for this? \nWhat learning rate schedule are you following, and what is your rationale behind it? How do you decide on a learning rate in the first place?\nPlease feel free to leave comments with any advice that might be helpful, I shall consolidate them for better sharing.\n\nPS. I am using PyTorch XLA"
  }
}