{
  "id": 131321,
  "title": "Is there anything wrong with colab TF 2.1?",
  "url": "/competitions/bengaliai-cv19/discussion/131321",
  "author_name": "Urvish",
  "post_date": "2020-02-19T09:33:31.805000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi guys, </p>\n\n<p>I am stuck at something and need help. </p>\n\n<p>I am using @ipythonx 's this kernel <a href=\"https://www.kaggle.com/ipythonx/keras-grapheme-gridmask-augmix-in-efficientnet\">https://www.kaggle.com/ipythonx/keras-grapheme-gridmask-augmix-in-efficientnet</a> \nI only changed it from <code>keras</code> to <code>tf 2.1</code>. I am training on <code>colab</code> and in my local and I get two different training outputs. On <code>colab</code>, nor loss or accuracy is improving at all. While in a local machine it is improving really well. Why is this happening? \nBelow is a picture of training on <code>colab</code>. </p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1127972%2Fda80756e58f5ad226951a51ba1dd911d%2Fcolab_pic.PNG?generation=1582103653073125&amp;alt=media\" alt=\"\"></p>\n\n<p>Below is a picture of my local machine training. \n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1127972%2F4199dbfbd05cba432ad700e67c6037ef%2Fjupyter_run.PNG?generation=1582103706563622&amp;alt=media\" alt=\"\"></p>",
  "messages": [
    {
      "id": 750318,
      "postDate": "2020-02-19T09:33:31.807Z",
      "content": "<p>Hi guys, </p>\n\n<p>I am stuck at something and need help. </p>\n\n<p>I am using @ipythonx 's this kernel <a href=\"https://www.kaggle.com/ipythonx/keras-grapheme-gridmask-augmix-in-efficientnet\">https://www.kaggle.com/ipythonx/keras-grapheme-gridmask-augmix-in-efficientnet</a> \nI only changed it from <code>keras</code> to <code>tf 2.1</code>. I am training on <code>colab</code> and in my local and I get two different training outputs. On <code>colab</code>, nor loss or accuracy is improving at all. While in a local machine it is improving really well. Why is this happening? \nBelow is a picture of training on <code>colab</code>. </p>\n\n<p><img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1127972%2Fda80756e58f5ad226951a51ba1dd911d%2Fcolab_pic.PNG?generation=1582103653073125&amp;alt=media\" alt=\"\"></p>\n\n<p>Below is a picture of my local machine training. \n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1127972%2F4199dbfbd05cba432ad700e67c6037ef%2Fjupyter_run.PNG?generation=1582103706563622&amp;alt=media\" alt=\"\"></p>",
      "rawMarkdown": "Hi guys, \n\nI am stuck at something and need help. \n\nI am using @ipythonx 's this kernel https://www.kaggle.com/ipythonx/keras-grapheme-gridmask-augmix-in-efficientnet \nI only changed it from `keras` to `tf 2.1`. I am training on `colab` and in my local and I get two different training outputs. On `colab`, nor loss or accuracy is improving at all. While in a local machine it is improving really well. Why is this happening? \nBelow is a picture of training on `colab`. \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1127972%2Fda80756e58f5ad226951a51ba1dd911d%2Fcolab_pic.PNG?generation=1582103653073125&amp;alt=media)\n\nBelow is a picture of my local machine training. \n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1127972%2F4199dbfbd05cba432ad700e67c6037ef%2Fjupyter_run.PNG?generation=1582103706563622&amp;alt=media)\n\n"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "750318": "Hi guys, \n\nI am stuck at something and need help. \n\nI am using @ipythonx 's this kernel https://www.kaggle.com/ipythonx/keras-grapheme-gridmask-augmix-in-efficientnet \nI only changed it from `keras` to `tf 2.1`. I am training on `colab` and in my local and I get two different training outputs. On `colab`, nor loss or accuracy is improving at all. While in a local machine it is improving really well. Why is this happening? \nBelow is a picture of training on `colab`. \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1127972%2Fda80756e58f5ad226951a51ba1dd911d%2Fcolab_pic.PNG?generation=1582103653073125&amp;alt=media)\n\nBelow is a picture of my local machine training. \n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1127972%2F4199dbfbd05cba432ad700e67c6037ef%2Fjupyter_run.PNG?generation=1582103706563622&amp;alt=media)\n\n"
  }
}