{ "id": 441930, "title": "Question about training time of baseline models", "url": "/competitions/predict-ai-model-runtime/discussion/441930", "author_name": "", "post_date": "2023-09-20T16:04:53.605408Z", "votes": 2, "comment_count": 1, "views": 0, "content": "

From the paper that describes the dataset, it said that it takes 2-3 days to train the model on a powerful GPU. How can I afford to train the same model when tinkering with the same model on an old laptop? Thank you
\n\"\"

", "messages": [ { "id": "2448490", "postDate": "09/20/2023 16:04:53", "content": "

From the paper that describes the dataset, it said that it takes 2-3 days to train the model on a powerful GPU. How can I afford to train the same model when tinkering with the same model on an old laptop? Thank you
\n\"\"

", "rawMarkdown": "From the paper that describes the dataset, it said that it takes 2-3 days to train the model on a powerful GPU. How can I afford to train the same model when tinkering with the same model on an old laptop? Thank you\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F16790356%2Fd4a57cfdf0649b2bf7ab2f93ff337a38%2Fggg.jpg?generation=1695225886634980&alt=media)", "votes": null }, { "id": "2455962", "postDate": "09/25/2023 20:04:50", "content": "

The reported number is from training the GST model on a GPU. The layout baseline model in tpu_graphs repo can be trained on a CPU. Although we haven't trained that model long enough to say how long it takes to converge.

", "rawMarkdown": "The reported number is from training [the GST model](https://github.com/kaidic/GST) on a GPU. The layout baseline model in [tpu_graphs repo](https://github.com/google-research-datasets/tpu_graphs) can be trained on a CPU. Although we haven't trained that model long enough to say how long it takes to converge.", "votes": null } ], "comments": [ { "id": 2455962, "author_name": "mangpophothilimthana", "author_url": "", "post_date": "09/25/2023 20:04:50", "content": "

The reported number is from training the GST model on a GPU. The layout baseline model in tpu_graphs repo can be trained on a CPU. Although we haven't trained that model long enough to say how long it takes to converge.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2448490": "From the paper that describes the dataset, it said that it takes 2-3 days to train the model on a powerful GPU. How can I afford to train the same model when tinkering with the same model on an old laptop? Thank you\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F16790356%2Fd4a57cfdf0649b2bf7ab2f93ff337a38%2Fggg.jpg?generation=1695225886634980&alt=media)", "2455962": "The reported number is from training [the GST model](https://github.com/kaidic/GST) on a GPU. The layout baseline model in [tpu_graphs repo](https://github.com/google-research-datasets/tpu_graphs) can be trained on a CPU. Although we haven't trained that model long enough to say how long it takes to converge." }, "source": "meta" }