![\"Time](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1001530%2Fabae983fb7e0c3d745995afa1bada930%2FTimeOutError.PNG?generation=1577024554356653&alt=media\")
This is the third time today I've faced this issue. I was able to submit a kernel (used resnet50 model for training) at first. But later, I got a timeout error twice while submitting another resnet50 model with the same notebook. Even I tried with resnet34 model but that didn't help. I am mostly using codes from Peter's inference notebook. Any sort of help would be appreciated. :)
\n\n
This is the third time today I've faced this issue. I was able to submit a kernel (used resnet50 model for training) at first. But later, I got a timeout error twice while submitting another resnet50 model with the same notebook. Even I tried with resnet34 model but that didn't help. I am mostly using codes from Peter's inference notebook. Any sort of help would be appreciated. :)
\n\n
The limit (GPU enabled) is 2 hours. My inference notebook is running ~2 hours, so any modification could end up with timeout. Try a bigger batch size, it should improve the inference time.
", "rawMarkdown": "The limit (GPU enabled) is 2 hours. My inference notebook is running ~2 hours, so any modification could end up with timeout. Try a bigger batch size, it should improve the inference time.", "votes": null }, { "id": "700758", "postDate": "12/22/2019 15:04:41", "content": "I was able to submit my kernel at first. Also, I've been using 64 as my batch size. But now I can't even submit kernels with shallower models using the same notebook.
", "rawMarkdown": "I was able to submit my kernel at first. Also, I've been using 64 as my batch size. But now I can't even submit kernels with shallower models using the same notebook.", "votes": null }, { "id": "700841", "postDate": "12/22/2019 17:41:19", "content": "By the way, I increased my batch size from 64 to 96 and it worked.
", "rawMarkdown": "By the way, I increased my batch size from 64 to 96 and it worked.", "votes": null }, { "id": "736415", "postDate": "02/04/2020 06:51:06", "content": "Facing the same issue. I was able to make the submission earlier with the model trained on batch size 128 but now I'm facing the issue with same batch size. Any other suggestions?
", "rawMarkdown": "Facing the same issue. I was able to make the submission earlier with the model trained on batch size 128 but now I'm facing the issue with same batch size. Any other suggestions?", "votes": null }, { "id": "736438", "postDate": "02/04/2020 07:19:58", "content": "Try converting pandas dataframe to numpy. It will speed up the inference.
", "rawMarkdown": "Try converting pandas dataframe to numpy. It will speed up the inference.", "votes": null }, { "id": "754476", "postDate": "02/23/2020 15:40:50", "content": "Can you tell me how to do it？Thanks.
", "rawMarkdown": "Can you tell me how to do it？Thanks.", "votes": null }, { "id": "754493", "postDate": "02/23/2020 16:20:49", "content": "Hey @tahsin Since my previous reply, I created an optimized inference notebook: https://www.kaggle.com/pestipeti/fast-ensemble-5-folds-20-minutes
\n\nWith this notebook, you can submit a 5-fold ensemble within 20 minutes. You can use it for one fold too if you update the configs at the beginning.
", "rawMarkdown": "Hey @tahsin Since my previous reply, I created an optimized inference notebook: https://www.kaggle.com/pestipeti/fast-ensemble-5-folds-20-minutes\n\nWith this notebook, you can submit a 5-fold ensemble within 20 minutes. You can use it for one fold too if you update the configs at the beginning.", "votes": null }, { "id": "881579", "postDate": "06/11/2020 07:20:47", "content": "Faced the same error for tweet sentiment extraction. I was using 5-fold-roberta and each fold was taking > 20 minutes to compile. In total, when 2 hours got exhausted, I got this error. With using only one-fold, it ran faster without error. So I had to compromise.. I feel, by using 5-fold, I could have got a higher score.
", "rawMarkdown": "Faced the same error for tweet sentiment extraction. I was using 5-fold-roberta and each fold was taking > 20 minutes to compile. In total, when 2 hours got exhausted, I got this error. With using only one-fold, it ran faster without error. So I had to compromise.. I feel, by using 5-fold, I could have got a higher score.", "votes": null } ], "comments": [ { "id": 700751, "author_name": "pestipeti", "author_url": "", "post_date": "12/22/2019 14:52:07", "content": "The limit (GPU enabled) is 2 hours. My inference notebook is running ~2 hours, so any modification could end up with timeout. Try a bigger batch size, it should improve the inference time.
", "votes": null, "replies": [ { "id": 700758, "author_name": "tahsin", "author_url": "", "post_date": "12/22/2019 15:04:41", "content": "I was able to submit my kernel at first. Also, I've been using 64 as my batch size. But now I can't even submit kernels with shallower models using the same notebook.
", "votes": null, "replies": [] }, { "id": 700841, "author_name": "tahsin", "author_url": "", "post_date": "12/22/2019 17:41:19", "content": "By the way, I increased my batch size from 64 to 96 and it worked.
", "votes": null, "replies": [] }, { "id": 736415, "author_name": "pawan28a95", "author_url": "", "post_date": "02/04/2020 06:51:06", "content": "Facing the same issue. I was able to make the submission earlier with the model trained on batch size 128 but now I'm facing the issue with same batch size. Any other suggestions?
", "votes": null, "replies": [] }, { "id": 736438, "author_name": "tahsin", "author_url": "", "post_date": "02/04/2020 07:19:58", "content": "Try converting pandas dataframe to numpy. It will speed up the inference.
", "votes": null, "replies": [] }, { "id": 754476, "author_name": "mnk812", "author_url": "", "post_date": "02/23/2020 15:40:50", "content": "Can you tell me how to do it？Thanks.
", "votes": null, "replies": [] } ] }, { "id": 754493, "author_name": "pestipeti", "author_url": "", "post_date": "02/23/2020 16:20:49", "content": "Hey @tahsin Since my previous reply, I created an optimized inference notebook: https://www.kaggle.com/pestipeti/fast-ensemble-5-folds-20-minutes
\n\nWith this notebook, you can submit a 5-fold ensemble within 20 minutes. You can use it for one fold too if you update the configs at the beginning.
", "votes": null, "replies": [] }, { "id": 881579, "author_name": "chandrimad31", "author_url": "", "post_date": "06/11/2020 07:20:47", "content": "Faced the same error for tweet sentiment extraction. I was using 5-fold-roberta and each fold was taking > 20 minutes to compile. In total, when 2 hours got exhausted, I got this error. With using only one-fold, it ran faster without error. So I had to compromise.. I feel, by using 5-fold, I could have got a higher score.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "700746": "This is the third time today I've faced this issue. I was able to submit a kernel (used resnet50 model for training) at first. But later, I got a timeout error twice while submitting another resnet50 model with the same notebook. Even I tried with resnet34 model but that didn't help. I am mostly using codes from Peter's inference notebook. Any sort of help would be appreciated. :)\n\n", "700751": "The limit (GPU enabled) is 2 hours. My inference notebook is running ~2 hours, so any modification could end up with timeout. Try a bigger batch size, it should improve the inference time.", "700758": "I was able to submit my kernel at first. Also, I've been using 64 as my batch size. But now I can't even submit kernels with shallower models using the same notebook.", "700841": "By the way, I increased my batch size from 64 to 96 and it worked.", "736415": "Facing the same issue. I was able to make the submission earlier with the model trained on batch size 128 but now I'm facing the issue with same batch size. Any other suggestions?", "736438": "Try converting pandas dataframe to numpy. It will speed up the inference.", "754476": "Can you tell me how to do it？Thanks.", "754493": "Hey @tahsin Since my previous reply, I created an optimized inference notebook: https://www.kaggle.com/pestipeti/fast-ensemble-5-folds-20-minutes\n\nWith this notebook, you can submit a 5-fold ensemble within 20 minutes. You can use it for one fold too if you update the configs at the beginning.", "881579": "Faced the same error for tweet sentiment extraction. I was using 5-fold-roberta and each fold was taking > 20 minutes to compile. In total, when 2 hours got exhausted, I got this error. With using only one-fold, it ran faster without error. So I had to compromise.. I feel, by using 5-fold, I could have got a higher score." }, "source": "meta" }