{
  "id": 141927,
  "title": "kernel stops after several epochs",
  "url": "/competitions/flower-classification-with-tpus/discussion/141927",
  "author_name": "",
  "post_date": "2020-04-08T03:23:08.018649500Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I'm trying to train a densenet201 with torch_xla, while the cpu sharply drops to 0.00% after the model runs for several epochs, and the training process stucks. The batch size is 16, so does anyone have ideas about that? thanks ~~</p>",
  "messages": [
    {
      "id": "801083",
      "postDate": "04/08/2020 03:23:08",
      "content": "<p>I'm trying to train a densenet201 with torch_xla, while the cpu sharply drops to 0.00% after the model runs for several epochs, and the training process stucks. The batch size is 16, so does anyone have ideas about that? thanks ~~</p>",
      "rawMarkdown": "I'm trying to train a densenet201 with torch_xla, while the cpu sharply drops to 0.00% after the model runs for several epochs, and the training process stucks. The batch size is 16, so does anyone have ideas about that? thanks ~~",
      "votes": null
    },
    {
      "id": "809360",
      "postDate": "04/16/2020 05:23:55",
      "content": "<p><a href=\"/chc278\">@chc278</a> since you are using Pytorch, I am not sure about the issue, Pytorch xla has some issues with TPU. DO check out Multilingual text competition, where multiple folks have used TPUs successfully</p>",
      "rawMarkdown": "chc278 since you are using Pytorch, I am not sure about the issue, Pytorch xla has some issues with TPU. DO check out Multilingual text competition, where multiple folks have used TPUs successfully",
      "votes": null
    },
    {
      "id": "810141",
      "postDate": "04/16/2020 18:44:51",
      "content": "<p>Link to <a href=\"/abhishek\">@abhishek</a> 's <a href=\"https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification/discussion/143613\">Super-clean PyTorch TPU training notebook</a>.</p>",
      "rawMarkdown": "Link to @abhishek 's [Super-clean PyTorch TPU training notebook](https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification/discussion/143613).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 809360,
      "author_name": "kurianbenoy",
      "author_url": "",
      "post_date": "04/16/2020 05:23:55",
      "content": "<p><a href=\"/chc278\">@chc278</a> since you are using Pytorch, I am not sure about the issue, Pytorch xla has some issues with TPU. DO check out Multilingual text competition, where multiple folks have used TPUs successfully</p>",
      "votes": null,
      "replies": [
        {
          "id": 810141,
          "author_name": "mgorner",
          "author_url": "",
          "post_date": "04/16/2020 18:44:51",
          "content": "<p>Link to <a href=\"/abhishek\">@abhishek</a> 's <a href=\"https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification/discussion/143613\">Super-clean PyTorch TPU training notebook</a>.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "801083": "I'm trying to train a densenet201 with torch_xla, while the cpu sharply drops to 0.00% after the model runs for several epochs, and the training process stucks. The batch size is 16, so does anyone have ideas about that? thanks ~~",
    "809360": "chc278 since you are using Pytorch, I am not sure about the issue, Pytorch xla has some issues with TPU. DO check out Multilingual text competition, where multiple folks have used TPUs successfully",
    "810141": "Link to @abhishek 's [Super-clean PyTorch TPU training notebook](https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification/discussion/143613)."
  },
  "source": "meta"
}