I've currently submitted my notebook, but it's still scoring. It is almost 6h from the time I submit it. Does anyone have the same situation as me? Or maybe my submission comes to error?
", "messages": [ { "id": "1816766", "postDate": "06/10/2022 14:32:26", "content": "I've currently submitted my notebook, but it's still scoring. It is almost 6h from the time I submit it. Does anyone have the same situation as me? Or maybe my submission comes to error?
", "rawMarkdown": "I've currently submitted my notebook, but it's still scoring. It is almost 6h from the time I submit it. Does anyone have the same situation as me? Or maybe my submission comes to error?", "votes": null }, { "id": "1817074", "postDate": "06/10/2022 21:08:34", "content": "It pretty much depends (on the backbone, how many folds, your model architecture, settings, etc.). In my case, inference with a 1-fold model with codebert-base takes nearly 2 hours. So, if you are using a bigger backbone, more folds, or complicated architecture, 6 hours is possible. But, you could check if you are using GPU or CPU first because I got that silly mistake once.
", "rawMarkdown": "It pretty much depends (on the backbone, how many folds, your model architecture, settings, etc.). In my case, inference with a 1-fold model with codebert-base takes nearly 2 hours. So, if you are using a bigger backbone, more folds, or complicated architecture, 6 hours is possible. But, you could check if you are using GPU or CPU first because I got that silly mistake once.", "votes": null }, { "id": "1817202", "postDate": "06/11/2022 03:46:00", "content": "For reference simple pointwise model only take 20min.
", "rawMarkdown": "For reference simple pointwise model only take 20min.", "votes": null }, { "id": "1822195", "postDate": "06/16/2022 06:19:29", "content": "In my experience, it took about 70min when I checked out a model that was fine-tuned model using code-bert with 20,000 train data.
", "rawMarkdown": "In my experience, it took about 70min when I checked out a model that was fine-tuned model using code-bert with 20,000 train data.", "votes": null }, { "id": "1822514", "postDate": "06/16/2022 13:14:23", "content": "If your code isn't optimized submissions can take a very long time. Check the 'notebook exceeded allowed compute!' thread for some ideas. Also you can feed your submission code 20K of training data files in a GPU Kaggle Notebook to get an idea how long it needs and where the bottlenecks are.
", "rawMarkdown": "If your code isn't optimized submissions can take a very long time. Check the 'notebook exceeded allowed compute!' thread for some ideas. Also you can feed your submission code 20K of training data files in a GPU Kaggle Notebook to get an idea how long it needs and where the bottlenecks are.", "votes": null } ], "comments": [ { "id": 1817074, "author_name": "shinomoriaoshi", "author_url": "", "post_date": "06/10/2022 21:08:34", "content": "It pretty much depends (on the backbone, how many folds, your model architecture, settings, etc.). In my case, inference with a 1-fold model with codebert-base takes nearly 2 hours. So, if you are using a bigger backbone, more folds, or complicated architecture, 6 hours is possible. But, you could check if you are using GPU or CPU first because I got that silly mistake once.
", "votes": null, "replies": [] }, { "id": 1817202, "author_name": "goldenlock", "author_url": "", "post_date": "06/11/2022 03:46:00", "content": "For reference simple pointwise model only take 20min.
", "votes": null, "replies": [] }, { "id": 1822195, "author_name": "heuingttiang", "author_url": "", "post_date": "06/16/2022 06:19:29", "content": "In my experience, it took about 70min when I checked out a model that was fine-tuned model using code-bert with 20,000 train data.
", "votes": null, "replies": [] }, { "id": 1822514, "author_name": "happycube", "author_url": "", "post_date": "06/16/2022 13:14:23", "content": "If your code isn't optimized submissions can take a very long time. Check the 'notebook exceeded allowed compute!' thread for some ideas. Also you can feed your submission code 20K of training data files in a GPU Kaggle Notebook to get an idea how long it needs and where the bottlenecks are.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1816766": "I've currently submitted my notebook, but it's still scoring. It is almost 6h from the time I submit it. Does anyone have the same situation as me? Or maybe my submission comes to error?", "1817074": "It pretty much depends (on the backbone, how many folds, your model architecture, settings, etc.). In my case, inference with a 1-fold model with codebert-base takes nearly 2 hours. So, if you are using a bigger backbone, more folds, or complicated architecture, 6 hours is possible. But, you could check if you are using GPU or CPU first because I got that silly mistake once.", "1817202": "For reference simple pointwise model only take 20min.", "1822195": "In my experience, it took about 70min when I checked out a model that was fine-tuned model using code-bert with 20,000 train data.", "1822514": "If your code isn't optimized submissions can take a very long time. Check the 'notebook exceeded allowed compute!' thread for some ideas. Also you can feed your submission code 20K of training data files in a GPU Kaggle Notebook to get an idea how long it needs and where the bottlenecks are." }, "source": "meta" }