As the training proceeds, epoch 1 2 3 is fine,but with more epochs, notebook will failed.\nYour notebook tried to allocate more memory than is available. It has restarted.\nI found that RAM memory increased with the epochs until it exploded.\nDoes anyone has good suggestions to solve this....HELP!
", "messages": [ { "id": "917029", "postDate": "07/06/2020 07:29:14", "content": "As the training proceeds, epoch 1 2 3 is fine,but with more epochs, notebook will failed.\nYour notebook tried to allocate more memory than is available. It has restarted.\nI found that RAM memory increased with the epochs until it exploded.\nDoes anyone has good suggestions to solve this....HELP!
", "rawMarkdown": "As the training proceeds, epoch 1 2 3 is fine,but with more epochs, notebook will failed.\nYour notebook tried to allocate more memory than is available. It has restarted.\nI found that RAM memory increased with the epochs until it exploded.\nDoes anyone has good suggestions to solve this....HELP!", "votes": null }, { "id": "917111", "postDate": "07/06/2020 08:47:30", "content": "To debug the issue:\n1. Use smaller size images say 224x224\n2. Lesser number of Augmentations\n3. Smaller batch size (not greater than 64)\n4. Look at earlier posts about data loading pipeline
", "rawMarkdown": "To debug the issue:\n1. Use smaller size images say 224x224\n2. Lesser number of Augmentations\n3. Smaller batch size (not greater than 64)\n4. Look at earlier posts about [data loading pipeline](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/158520#885869)", "votes": null }, { "id": "917379", "postDate": "07/06/2020 13:03:28", "content": "Use colab if you want TPU with pytorch. Colab gives 35GB of RAM and 40 core processor while Kaggle offers only 13GB RAM and 4 core processor. The benefits of more RAM and processors outweighs the difference between v2 and v3 TPU. Also check this link regarding pytorch TPU usage.
", "rawMarkdown": "Use colab if you want TPU with pytorch. Colab gives 35GB of RAM and 40 core processor while Kaggle offers only 13GB RAM and 4 core processor. The benefits of more RAM and processors outweighs the difference between v2 and v3 TPU. Also check [this link](https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification/discussion/159723) regarding pytorch TPU usage.", "votes": null }, { "id": "918283", "postDate": "07/07/2020 06:29:54", "content": "Thank you for reply! i will check some great notebook to solve
", "rawMarkdown": "Thank you for reply! i will check some great notebook to solve", "votes": null }, { "id": "918287", "postDate": "07/07/2020 06:33:37", "content": "Thank you for telling me such information.😃 I've tried colab, but it didnt work in my area.i think i should try more a few times.Thank you
", "rawMarkdown": "Thank you for telling me such information.😃 I've tried colab, but it didnt work in my area.i think i should try more a few times.Thank you", "votes": null }, { "id": "920814", "postDate": "07/08/2020 20:13:15", "content": "You can try Colab Pro. With 35GB RAM. Tell me if you need help! 👐
", "rawMarkdown": "You can try Colab Pro. With 35GB RAM. Tell me if you need help! 👐", "votes": null } ], "comments": [ { "id": 917111, "author_name": "sirishks", "author_url": "", "post_date": "07/06/2020 08:47:30", "content": "To debug the issue:\n1. Use smaller size images say 224x224\n2. Lesser number of Augmentations\n3. Smaller batch size (not greater than 64)\n4. Look at earlier posts about data loading pipeline
", "votes": null, "replies": [ { "id": 918283, "author_name": "changewow", "author_url": "", "post_date": "07/07/2020 06:29:54", "content": "Thank you for reply! i will check some great notebook to solve
", "votes": null, "replies": [] } ] }, { "id": 917379, "author_name": "utsavnandi", "author_url": "", "post_date": "07/06/2020 13:03:28", "content": "Use colab if you want TPU with pytorch. Colab gives 35GB of RAM and 40 core processor while Kaggle offers only 13GB RAM and 4 core processor. The benefits of more RAM and processors outweighs the difference between v2 and v3 TPU. Also check this link regarding pytorch TPU usage.
", "votes": null, "replies": [ { "id": 918287, "author_name": "changewow", "author_url": "", "post_date": "07/07/2020 06:33:37", "content": "Thank you for telling me such information.😃 I've tried colab, but it didnt work in my area.i think i should try more a few times.Thank you
", "votes": null, "replies": [] } ] }, { "id": 920814, "author_name": "doanquanvietnamca", "author_url": "", "post_date": "07/08/2020 20:13:15", "content": "You can try Colab Pro. With 35GB RAM. Tell me if you need help! 👐
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "917029": "As the training proceeds, epoch 1 2 3 is fine,but with more epochs, notebook will failed.\nYour notebook tried to allocate more memory than is available. It has restarted.\nI found that RAM memory increased with the epochs until it exploded.\nDoes anyone has good suggestions to solve this....HELP!", "917111": "To debug the issue:\n1. Use smaller size images say 224x224\n2. Lesser number of Augmentations\n3. Smaller batch size (not greater than 64)\n4. Look at earlier posts about [data loading pipeline](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/158520#885869)", "917379": "Use colab if you want TPU with pytorch. Colab gives 35GB of RAM and 40 core processor while Kaggle offers only 13GB RAM and 4 core processor. The benefits of more RAM and processors outweighs the difference between v2 and v3 TPU. Also check [this link](https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification/discussion/159723) regarding pytorch TPU usage.", "918283": "Thank you for reply! i will check some great notebook to solve", "918287": "Thank you for telling me such information.😃 I've tried colab, but it didnt work in my area.i think i should try more a few times.Thank you", "920814": "You can try Colab Pro. With 35GB RAM. Tell me if you need help! 👐" }, "source": "meta" }