Is 90 minutes CPU-only is for ALL the processing? Like, can I leave only fine-tune code part of model and load pre-trained model? As if I shrink my dataset gathered from input data files I get no satisfactory results. But if I train it in CPU-only mode it takes too long to fit in the 90 minutes.
\nI mean upload pre-trained model by me on my PC, using the code from Kaggle-notebook I upload for submission. So it was not published somewhere before, but I am fine if someone can use it as the competition finishes. It is about the definition of terms in competition description I am not really familiar with.
\nBecause no I am pressed to waste the time on optimizing training process to fit it in almost unrealistic calculation time instead of looking for a solutions for refining dataset and enhancing other things
Is 90 minutes CPU-only is for ALL the processing? Like, can I leave only fine-tune code part of model and load pre-trained model? As if I shrink my dataset gathered from input data files I get no satisfactory results. But if I train it in CPU-only mode it takes too long to fit in the 90 minutes.
\nI mean upload pre-trained model by me on my PC, using the code from Kaggle-notebook I upload for submission. So it was not published somewhere before, but I am fine if someone can use it as the competition finishes. It is about the definition of terms in competition description I am not really familiar with.
\nBecause no I am pressed to waste the time on optimizing training process to fit it in almost unrealistic calculation time instead of looking for a solutions for refining dataset and enhancing other things
Hi! The 90-minute limit applies only to the inference phase — that is, when the already trained model is used to generate the final predictions for evaluation.
\nYou're allowed to train and fine-tune your model beforehand on your own machine (or another environment), and then upload the trained model as part of your inference notebook. What matters is that loading the model and generating predictions must stay within the 90-minute limit during notebook execution on Kaggle.
\nI recommend checking some public notebooks focused only on inference — many participants train their models offline and simply load them in the notebook.
\nThis way, you can focus on improving your dataset, preprocessing, and model quality, without having to squeeze training into the time limit.
\nGood luck with the competition!
", "rawMarkdown": "Hi! The 90-minute limit applies only to the inference phase — that is, when the already trained model is used to generate the final predictions for evaluation.\n\nYou're allowed to train and fine-tune your model beforehand on your own machine (or another environment), and then upload the trained model as part of your inference notebook. What matters is that loading the model and generating predictions must stay within the 90-minute limit during notebook execution on Kaggle.\n\nI recommend checking some public notebooks focused only on inference — many participants train their models offline and simply load them in the notebook.\n\nThis way, you can focus on improving your dataset, preprocessing, and model quality, without having to squeeze training into the time limit.\n\nGood luck with the competition!", "votes": null }, { "id": "3216539", "postDate": "06/03/2025 17:58:51", "content": "Thanks a lot! Though it might seem ignorant from my side, but I truely tried to read and catch the details through the text and could not really understand the issue. But I have already made chatGPT to explain the rules with focus on the question. Only then it made the things clearer, but I have already tried to train the GT nns and dont trustem anymore =)
Hi! The 90-minute limit applies only to the inference phase — that is, when the already trained model is used to generate the final predictions for evaluation.
\nYou're allowed to train and fine-tune your model beforehand on your own machine (or another environment), and then upload the trained model as part of your inference notebook. What matters is that loading the model and generating predictions must stay within the 90-minute limit during notebook execution on Kaggle.
\nI recommend checking some public notebooks focused only on inference — many participants train their models offline and simply load them in the notebook.
\nThis way, you can focus on improving your dataset, preprocessing, and model quality, without having to squeeze training into the time limit.
\nGood luck with the competition!
", "votes": null, "replies": [ { "id": 3216539, "author_name": "pavelssuskis", "author_url": "", "post_date": "06/03/2025 17:58:51", "content": "Thanks a lot! Though it might seem ignorant from my side, but I truely tried to read and catch the details through the text and could not really understand the issue. But I have already made chatGPT to explain the rules with focus on the question. Only then it made the things clearer, but I have already tried to train the GT nns and dont trustem anymore =)