When I read public kernels for this competition I often see TPU used. \nDoes it mean you are using 30h of free time per week to train the models or is there something else I don't know?\nAs far as I know you can't buy TPU for home, so do you use some paid service to work on this competition or is there some free way to use TPU here?
", "messages": [ { "id": "898691", "postDate": "06/23/2020 17:14:19", "content": "When I read public kernels for this competition I often see TPU used. \nDoes it mean you are using 30h of free time per week to train the models or is there something else I don't know?\nAs far as I know you can't buy TPU for home, so do you use some paid service to work on this competition or is there some free way to use TPU here?
", "rawMarkdown": "When I read public kernels for this competition I often see TPU used. \nDoes it mean you are using 30h of free time per week to train the models or is there something else I don't know?\nAs far as I know you can't buy TPU for home, so do you use some paid service to work on this competition or is there some free way to use TPU here?", "votes": null }, { "id": "898849", "postDate": "06/23/2020 19:41:44", "content": "You can use 30 hours of Kaggle TPU time AND you have 30 hours of GPU time also available - per week. Most of the top voted kernels have the code needed to setup and run a Kaggle TPU.
\n\nTPU's must run 3 hours or less - so you can get around 10 commits per week.
\n\nThis new competition has lots of info on using the TPU's.\nhttps://www.kaggle.com/c/tpu-getting-started
", "rawMarkdown": "You can use 30 hours of Kaggle TPU time AND you have 30 hours of GPU time also available - per week. Most of the top voted kernels have the code needed to setup and run a Kaggle TPU.\n\nTPU's must run 3 hours or less - so you can get around 10 commits per week. \n\nThis new competition has lots of info on using the TPU's.\nhttps://www.kaggle.com/c/tpu-getting-started", "votes": null }, { "id": "898876", "postDate": "06/23/2020 20:03:06", "content": "Is there a kernel that had the code to easily switch from GPU to TPU execution that you recommend?
", "rawMarkdown": "Is there a kernel that had the code to easily switch from GPU to TPU execution that you recommend?", "votes": null }, { "id": "899192", "postDate": "06/24/2020 04:37:37", "content": "You can get a TPU on Colab which you can run for 12 hours in free version and for 24 hours if you sign up for Colab Pro (~$10 per month). Here is my Kaggle notebook which can be adjusted for Colab: https://www.kaggle.com/graf10a/effnb0-tabular-features-tf-cv5-512x512
\n\n(More discussion here: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/158395)
", "rawMarkdown": "You can get a TPU on Colab which you can run for 12 hours in free version and for 24 hours if you sign up for Colab Pro (~$10 per month). Here is my Kaggle notebook which can be adjusted for Colab: https://www.kaggle.com/graf10a/effnb0-tabular-features-tf-cv5-512x512\n\n(More discussion here: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/158395)", "votes": null }, { "id": "915018", "postDate": "07/04/2020 12:25:38", "content": "Thanks for the info!
", "rawMarkdown": "Thanks for the info!", "votes": null }, { "id": "915353", "postDate": "07/04/2020 17:01:09", "content": "If you use Colab i suggest that you don't download the full 110GB of data. Instead just download 1GB, 2GB, 3GB, 5GB depending on what size you want, 256x256, 384x384, 512x512, and 768x768. I posted both JPEGS for PyTorch and TFRecords for TensorFlow here
", "rawMarkdown": "If you use Colab i suggest that you don't download the full 110GB of data. Instead just download 1GB, 2GB, 3GB, 5GB depending on what size you want, 256x256, 384x384, 512x512, and 768x768. I posted both JPEGS for PyTorch and TFRecords for TensorFlow [here][1]\n\n[1]: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/164092", "votes": null } ], "comments": [ { "id": 898849, "author_name": "pcjimmmy", "author_url": "", "post_date": "06/23/2020 19:41:44", "content": "You can use 30 hours of Kaggle TPU time AND you have 30 hours of GPU time also available - per week. Most of the top voted kernels have the code needed to setup and run a Kaggle TPU.
\n\nTPU's must run 3 hours or less - so you can get around 10 commits per week.
\n\nThis new competition has lots of info on using the TPU's.\nhttps://www.kaggle.com/c/tpu-getting-started
", "votes": null, "replies": [ { "id": 898876, "author_name": "hiramcho", "author_url": "", "post_date": "06/23/2020 20:03:06", "content": "Is there a kernel that had the code to easily switch from GPU to TPU execution that you recommend?
", "votes": null, "replies": [] }, { "id": 915018, "author_name": "erebusue", "author_url": "", "post_date": "07/04/2020 12:25:38", "content": "Thanks for the info!
", "votes": null, "replies": [] } ] }, { "id": 899192, "author_name": "graf10a", "author_url": "", "post_date": "06/24/2020 04:37:37", "content": "You can get a TPU on Colab which you can run for 12 hours in free version and for 24 hours if you sign up for Colab Pro (~$10 per month). Here is my Kaggle notebook which can be adjusted for Colab: https://www.kaggle.com/graf10a/effnb0-tabular-features-tf-cv5-512x512
\n\n(More discussion here: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/158395)
", "votes": null, "replies": [] }, { "id": 915353, "author_name": "cdeotte", "author_url": "", "post_date": "07/04/2020 17:01:09", "content": "If you use Colab i suggest that you don't download the full 110GB of data. Instead just download 1GB, 2GB, 3GB, 5GB depending on what size you want, 256x256, 384x384, 512x512, and 768x768. I posted both JPEGS for PyTorch and TFRecords for TensorFlow here
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "898691": "When I read public kernels for this competition I often see TPU used. \nDoes it mean you are using 30h of free time per week to train the models or is there something else I don't know?\nAs far as I know you can't buy TPU for home, so do you use some paid service to work on this competition or is there some free way to use TPU here?", "898849": "You can use 30 hours of Kaggle TPU time AND you have 30 hours of GPU time also available - per week. Most of the top voted kernels have the code needed to setup and run a Kaggle TPU.\n\nTPU's must run 3 hours or less - so you can get around 10 commits per week. \n\nThis new competition has lots of info on using the TPU's.\nhttps://www.kaggle.com/c/tpu-getting-started", "898876": "Is there a kernel that had the code to easily switch from GPU to TPU execution that you recommend?", "899192": "You can get a TPU on Colab which you can run for 12 hours in free version and for 24 hours if you sign up for Colab Pro (~$10 per month). Here is my Kaggle notebook which can be adjusted for Colab: https://www.kaggle.com/graf10a/effnb0-tabular-features-tf-cv5-512x512\n\n(More discussion here: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/158395)", "915018": "Thanks for the info!", "915353": "If you use Colab i suggest that you don't download the full 110GB of data. Instead just download 1GB, 2GB, 3GB, 5GB depending on what size you want, 256x256, 384x384, 512x512, and 768x768. I posted both JPEGS for PyTorch and TFRecords for TensorFlow [here][1]\n\n[1]: https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/164092" }, "source": "meta" }