Hi. I had a 5 hours 20 minut left on GPU quota and I've launched a commit with one last experimental training. I've tested it locally before and it should finished in about 4 hours according to tqdm ETA. However, it's still running and it consumed my GPU quota completely. Although the true scoring is executed externally it seems that is not possible to submit inference GPU codes without GPU quota remaining since the dummy local execution has to be done. Is that the case? Thanks.
", "messages": [ { "id": 2638156, "postDate": "2024-02-06T05:52:23.863Z", "content": "There was some discussion a few months ago in Product Feedback: Can't submit a quick saved notebook to a competition
\nwith the idea of creating a dummy submission.csv and do a QuickSave with Advanced Settings to save output and Run with GPU.
\nHowever this had an issue in the submission since Quick Save never uses GPU just CPU and then the submission did not use GPU. There was going to be internal discussion about whether that might change, not sure what the outcome was.
\nIf it was changed then in theory you could use Quick Save then do the submit with the submission.csv in its output.
In the past, submit to competition directly (in the notebook right panel) was a way people got around the GPU quota, but that was changed when it was abused. Would be good though just for the final day in a competition.
", "rawMarkdown": "There was some discussion a few months ago in Product Feedback: [Can't submit a quick saved notebook to a competition](https://www.kaggle.com/discussions/product-feedback/440750)\nwith the idea of creating a dummy submission.csv and do a QuickSave with Advanced Settings to save output and Run with GPU.\nHowever this had an issue in the submission since Quick Save never uses GPU just CPU and then the submission did not use GPU. There was going to be internal discussion about whether that might change, not sure what the outcome was.\nIf it was changed then in theory you could use Quick Save then do the submit with the submission.csv in its output. \n\nIn the past, submit to competition directly (in the notebook right panel) was a way people got around the GPU quota, but that was changed when it was abused. Would be good though just for the final day in a competition.\n", "votes": 1, "replies": [ { "id": 2638542, "postDate": "2024-02-06T11:04:39.370Z", "content": "Bad luck then. Good luck everyone with last submissions.
", "rawMarkdown": "Bad luck then. Good luck everyone with last submissions." } ] }, { "id": 2637459, "postDate": "2024-02-05T18:08:33.737Z", "content": "If you are completely out, I do not believe so. However if you have even a minute you can submit 2 at a time without issue.
", "rawMarkdown": "If you are completely out, I do not believe so. However if you have even a minute you can submit 2 at a time without issue.", "votes": 1 }, { "id": 2637360, "postDate": "2024-02-05T17:04:43.503Z", "content": "Hi. I had a 5 hours 20 minut left on GPU quota and I've launched a commit with one last experimental training. I've tested it locally before and it should finished in about 4 hours according to tqdm ETA. However, it's still running and it consumed my GPU quota completely. Although the true scoring is executed externally it seems that is not possible to submit inference GPU codes without GPU quota remaining since the dummy local execution has to be done. Is that the case? Thanks.
", "rawMarkdown": "Hi. I had a 5 hours 20 minut left on GPU quota and I've launched a commit with one last experimental training. I've tested it locally before and it should finished in about 4 hours according to tqdm ETA. However, it's still running and it consumed my GPU quota completely. Although the true scoring is executed externally it seems that is not possible to submit inference GPU codes without GPU quota remaining since the dummy local execution has to be done. Is that the case? Thanks." }, { "id": 2637992, "postDate": "2024-02-06T03:31:50.627Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ], "comments": [ { "id": 2638156, "author_name": "something4kag", "author_url": "", "post_date": "2024-02-06T05:52:23.863000", "content": "There was some discussion a few months ago in Product Feedback: Can't submit a quick saved notebook to a competition
\nwith the idea of creating a dummy submission.csv and do a QuickSave with Advanced Settings to save output and Run with GPU.
\nHowever this had an issue in the submission since Quick Save never uses GPU just CPU and then the submission did not use GPU. There was going to be internal discussion about whether that might change, not sure what the outcome was.
\nIf it was changed then in theory you could use Quick Save then do the submit with the submission.csv in its output.
In the past, submit to competition directly (in the notebook right panel) was a way people got around the GPU quota, but that was changed when it was abused. Would be good though just for the final day in a competition.
", "votes": 1, "replies": [ { "id": 2638542, "author_name": "Ángel Jacinto Sánchez Ruiz", "author_url": "", "post_date": "2024-02-06T11:04:39.370000", "content": "Bad luck then. Good luck everyone with last submissions.
", "votes": 0, "replies": [] } ] }, { "id": 2637459, "author_name": "Cody_Null", "author_url": "", "post_date": "2024-02-05T18:08:33.737000", "content": "If you are completely out, I do not believe so. However if you have even a minute you can submit 2 at a time without issue.
", "votes": 1, "replies": [] }, { "id": 2637992, "author_name": "", "author_url": "", "post_date": "2024-02-06T03:31:50.627000", "content": "", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "2638156": "There was some discussion a few months ago in Product Feedback: [Can't submit a quick saved notebook to a competition](https://www.kaggle.com/discussions/product-feedback/440750)\nwith the idea of creating a dummy submission.csv and do a QuickSave with Advanced Settings to save output and Run with GPU.\nHowever this had an issue in the submission since Quick Save never uses GPU just CPU and then the submission did not use GPU. There was going to be internal discussion about whether that might change, not sure what the outcome was.\nIf it was changed then in theory you could use Quick Save then do the submit with the submission.csv in its output. \n\nIn the past, submit to competition directly (in the notebook right panel) was a way people got around the GPU quota, but that was changed when it was abused. Would be good though just for the final day in a competition.\n", "2637459": "If you are completely out, I do not believe so. However if you have even a minute you can submit 2 at a time without issue.", "2637360": "Hi. I had a 5 hours 20 minut left on GPU quota and I've launched a commit with one last experimental training. I've tested it locally before and it should finished in about 4 hours according to tqdm ETA. However, it's still running and it consumed my GPU quota completely. Although the true scoring is executed externally it seems that is not possible to submit inference GPU codes without GPU quota remaining since the dummy local execution has to be done. Is that the case? Thanks.", "2637992": "" } }