{ "id": 160924, "title": "The Kernel has been Stopped after 3 hours of TPU continuos use", "url": "/competitions/siim-isic-melanoma-classification/discussion/160924", "author_name": "", "post_date": "2020-06-23T06:47:37.049457900Z", "votes": 1, "comment_count": 4, "views": 0, "content": "

\"\"\n\"\"

\n\n

I have received that message although i still have more than 20 hours in my Quota

", "messages": [ { "id": "897872", "postDate": "06/23/2020 06:47:37", "content": "

\"\"\n\"\"

\n\n

I have received that message although i still have more than 20 hours in my Quota

", "rawMarkdown": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2747432%2F2c252f3c450eadc1ceb134023b55da30%2FCapture1.PNG?generation=1592894815821824&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2747432%2Fb157f3f672294215dea7974093fd5f3c%2FCapture2.PNG?generation=1592894816028179&alt=media)\n\nI have received that message although i still have more than 20 hours in my Quota", "votes": null }, { "id": "898032", "postDate": "06/23/2020 08:36:16", "content": "

TPUs have a limited use time of 3h per session:

\n\n
\n

You can use up to 30 hours per week of TPUs and up to 3h at a time in a single session.

\n
\n\n

Oficial documentation: https://www.kaggle.com/docs/tpu

", "rawMarkdown": "TPUs have a limited use time of 3h per session:\n> You can use up to 30 hours per week of TPUs and up to 3h at a time in a single session.\n\nOficial documentation: https://www.kaggle.com/docs/tpu", "votes": null }, { "id": "898524", "postDate": "06/23/2020 15:12:20", "content": "

You have to use check points , if using pytorch check out Nvidia apex . It almost half the time in training.

", "rawMarkdown": "You have to use check points , if using pytorch check out Nvidia apex . It almost half the time in training.", "votes": null }, { "id": "898857", "postDate": "06/23/2020 19:47:42", "content": "

You need to start out a new kernel with an epoch run length that will lead to around a 1.5 hour run - than you can start to creep up the epochs to get close to the 3 hour limit. The execution info tells you the seconds used.

", "rawMarkdown": "You need to start out a new kernel with an epoch run length that will lead to around a 1.5 hour run - than you can start to creep up the epochs to get close to the 3 hour limit. The execution info tells you the seconds used.", "votes": null }, { "id": "899290", "postDate": "06/24/2020 06:16:53", "content": "

yes, i have already done that, thanks

", "rawMarkdown": "yes, i have already done that, thanks", "votes": null } ], "comments": [ { "id": 898032, "author_name": "cayala", "author_url": "", "post_date": "06/23/2020 08:36:16", "content": "

TPUs have a limited use time of 3h per session:

\n\n
\n

You can use up to 30 hours per week of TPUs and up to 3h at a time in a single session.

\n
\n\n

Oficial documentation: https://www.kaggle.com/docs/tpu

", "votes": null, "replies": [] }, { "id": 898524, "author_name": "yash612", "author_url": "", "post_date": "06/23/2020 15:12:20", "content": "

You have to use check points , if using pytorch check out Nvidia apex . It almost half the time in training.

", "votes": null, "replies": [] }, { "id": 898857, "author_name": "pcjimmmy", "author_url": "", "post_date": "06/23/2020 19:47:42", "content": "

You need to start out a new kernel with an epoch run length that will lead to around a 1.5 hour run - than you can start to creep up the epochs to get close to the 3 hour limit. The execution info tells you the seconds used.

", "votes": null, "replies": [ { "id": 899290, "author_name": "ahmedewida", "author_url": "", "post_date": "06/24/2020 06:16:53", "content": "

yes, i have already done that, thanks

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "897872": "![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2747432%2F2c252f3c450eadc1ceb134023b55da30%2FCapture1.PNG?generation=1592894815821824&alt=media)\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F2747432%2Fb157f3f672294215dea7974093fd5f3c%2FCapture2.PNG?generation=1592894816028179&alt=media)\n\nI have received that message although i still have more than 20 hours in my Quota", "898032": "TPUs have a limited use time of 3h per session:\n> You can use up to 30 hours per week of TPUs and up to 3h at a time in a single session.\n\nOficial documentation: https://www.kaggle.com/docs/tpu", "898524": "You have to use check points , if using pytorch check out Nvidia apex . It almost half the time in training.", "898857": "You need to start out a new kernel with an epoch run length that will lead to around a 1.5 hour run - than you can start to creep up the epochs to get close to the 3 hour limit. The execution info tells you the seconds used.", "899290": "yes, i have already done that, thanks" }, "source": "meta" }