![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3887924%2F78fa994834672e59112701539f064b6b%2FScreenshot%202024-06-11%20203713.png?generation=1718120358606445&alt=media\")
I am using my pretrained model for the inference. I have submitted the notebook and its been more than 2 hours and kaggle still process my notebook for submission. When I open the notebook, all the cells have run already.
\n
I am using my pretrained model for the inference. I have submitted the notebook and its been more than 2 hours and kaggle still process my notebook for submission. When I open the notebook, all the cells have run already.
\n
The number of test cases are greater during submission. You need to account for the runtime of your model on those cases also in the time it takes during submission.
", "rawMarkdown": "The number of test cases are greater during submission. You need to account for the runtime of your model on those cases also in the time it takes during submission.", "votes": null }, { "id": "2869922", "postDate": "06/13/2024 10:52:11", "content": "One of my submission took 16 hours and then I got the timeout error. Even though my notebook runs without any error and generates the submission file in the correct format. But still i am getting timeout error.
\nI am not training the model in my notebook. I am using my pretrained models to get the inference.
Yes, pretrained models to get the inference is the correct way to do a submission. But I am asking what is the average time taken by this inference code?
\nFor example, sample 200 study ids from train set and see what the is average time (per study). If this too long, see if there is any way to optimize it.
\nAlso make sure you are using GPU for submission and not CPU.
I have also had a scoring-time of 4 hours. The best and easiest thing, you can do is to set the accelerator to GPU T4*2. And resubmit.
", "rawMarkdown": "I have also had a scoring-time of 4 hours. The best and easiest thing, you can do is to set the accelerator to GPU T4*2. And resubmit.", "votes": null }, { "id": "2874366", "postDate": "06/16/2024 07:45:47", "content": "Would it utilize my GPU quota during submission?
", "rawMarkdown": "Would it utilize my GPU quota during submission?", "votes": null }, { "id": "2874446", "postDate": "06/16/2024 09:23:30", "content": "No, I have checked this in the past, where my GPU quota was almost running out. The re-run with private test data is unconnected to your existing GPU quota.
", "rawMarkdown": "No, I have checked this in the past, where my GPU quota was almost running out. The re-run with private test data is unconnected to your existing GPU quota.", "votes": null }, { "id": "2874475", "postDate": "06/16/2024 09:55:50", "content": "No, it will not
", "rawMarkdown": "No, it will not", "votes": null } ], "comments": [ { "id": 2868073, "author_name": "coderrkj", "author_url": "", "post_date": "06/12/2024 07:10:58", "content": "The number of test cases are greater during submission. You need to account for the runtime of your model on those cases also in the time it takes during submission.
", "votes": null, "replies": [ { "id": 2869922, "author_name": "hussamcheema", "author_url": "", "post_date": "06/13/2024 10:52:11", "content": "One of my submission took 16 hours and then I got the timeout error. Even though my notebook runs without any error and generates the submission file in the correct format. But still i am getting timeout error.
\nI am not training the model in my notebook. I am using my pretrained models to get the inference.
Yes, pretrained models to get the inference is the correct way to do a submission. But I am asking what is the average time taken by this inference code?
\nFor example, sample 200 study ids from train set and see what the is average time (per study). If this too long, see if there is any way to optimize it.
\nAlso make sure you are using GPU for submission and not CPU.
I have also had a scoring-time of 4 hours. The best and easiest thing, you can do is to set the accelerator to GPU T4*2. And resubmit.
", "votes": null, "replies": [ { "id": 2874366, "author_name": "hussamcheema", "author_url": "", "post_date": "06/16/2024 07:45:47", "content": "Would it utilize my GPU quota during submission?
", "votes": null, "replies": [ { "id": 2874446, "author_name": "coderrkj", "author_url": "", "post_date": "06/16/2024 09:23:30", "content": "No, I have checked this in the past, where my GPU quota was almost running out. The re-run with private test data is unconnected to your existing GPU quota.
", "votes": null, "replies": [] }, { "id": 2874475, "author_name": "devsya", "author_url": "", "post_date": "06/16/2024 09:55:50", "content": "No, it will not
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2867006": "I am using my pretrained model for the inference. I have submitted the notebook and its been more than 2 hours and kaggle still process my notebook for submission. When I open the notebook, all the cells have run already.\n\n\n", "2868073": "The number of test cases are greater during submission. You need to account for the runtime of your model on those cases also in the time it takes during submission.", "2869922": "One of my submission took 16 hours and then I got the timeout error. Even though my notebook runs without any error and generates the submission file in the correct format. But still i am getting timeout error.\nI am not training the model in my notebook. I am using my pretrained models to get the inference.", "2870351": "Yes, pretrained models to get the inference is the correct way to do a submission. But I am asking what is the average time taken by this inference code?\nFor example, sample 200 study ids from train set and see what the is average time (per study). If this too long, see if there is any way to optimize it.\nAlso make sure you are using GPU for submission and not CPU.", "2874239": "I have also had a scoring-time of 4 hours. The best and easiest thing, you can do is to set the accelerator to GPU T4*2. And resubmit.", "2874366": "Would it utilize my GPU quota during submission?", "2874446": "No, I have checked this in the past, where my GPU quota was almost running out. The re-run with private test data is unconnected to your existing GPU quota.", "2874475": "No, it will not" }, "source": "meta" }