{ "id": 232180, "title": "Time taking for inference", "url": "/competitions/hubmap-kidney-segmentation/discussion/232180", "author_name": "", "post_date": "2021-04-12T14:47:13.886159600Z", "votes": 3, "comment_count": 13, "views": 0, "content": "

Can someone share how much time their notebook took for inference after submitting ?
\nCurrently mine is over 2 hours and still showing status as \"Notebook Running\".

", "messages": [ { "id": "1271370", "postDate": "04/12/2021 14:47:13", "content": "

Can someone share how much time their notebook took for inference after submitting ?
\nCurrently mine is over 2 hours and still showing status as \"Notebook Running\".

", "rawMarkdown": "Can someone share how much time their notebook took for inference after submitting ?\nCurrently mine is over 2 hours and still showing status as \"Notebook Running\".", "votes": null }, { "id": "1271482", "postDate": "04/12/2021 16:46:02", "content": "

Mine can take up to 2h without GPU acceleration, i will use gpu on the next commit to see the changes and let you know

", "rawMarkdown": "Mine can take up to 2h without GPU acceleration, i will use gpu on the next commit to see the changes and let you know", "votes": null }, { "id": "1271483", "postDate": "04/12/2021 16:47:38", "content": "

mine takes roughly 20 - 30 mins using gpu for 5 folds of a single model

", "rawMarkdown": "mine takes roughly 20 - 30 mins using gpu for 5 folds of a single model", "votes": null }, { "id": "1271489", "postDate": "04/12/2021 16:54:45", "content": "

I just realised I was also not using gpu, but still on cpu it is over 4 hours now. I don't know what I am doing wrong. I'll try with gpu on this time.

", "rawMarkdown": "I just realised I was also not using gpu, but still on cpu it is over 4 hours now. I don't know what I am doing wrong. I'll try with gpu on this time.", "votes": null }, { "id": "1271500", "postDate": "04/12/2021 17:02:54", "content": "

Ok i just tried on GPU and now its going pretty fast (<15min), i guess i just have to get used to this kaggle env, to be honest i find it a little bit confusing.

", "rawMarkdown": "Ok i just tried on GPU and now its going pretty fast (<15min), i guess i just have to get used to this kaggle env, to be honest i find it a little bit confusing.", "votes": null }, { "id": "1271529", "postDate": "04/12/2021 17:17:35", "content": "

Yeah, I also tried on GPU and it took only 35 minutes while my CPU one is still running.

", "rawMarkdown": "Yeah, I also tried on GPU and it took only 35 minutes while my CPU one is still running.", "votes": null }, { "id": "1271533", "postDate": "04/12/2021 17:19:26", "content": "

I also ran it on GPU but it took 35 minutes for a single fold, can you suggest some tips to reduce this time ?

", "rawMarkdown": "I also ran it on GPU but it took 35 minutes for a single fold, can you suggest some tips to reduce this time ?", "votes": null }, { "id": "1271539", "postDate": "04/12/2021 17:21:47", "content": "

but why you guys running your inference code on cpu ?? any specific reason for that ?

", "rawMarkdown": "but why you guys running your inference code on cpu ?? any specific reason for that ?", "votes": null }, { "id": "1271544", "postDate": "04/12/2021 17:25:27", "content": "

Well my case is that i didn't know any better 😂

\n

I won't do it again now that i know a little better my ways around this kaggle notebooks environment

", "rawMarkdown": "Well my case is that i didn't know any better 😂\n\nI won't do it again now that i know a little better my ways around this kaggle notebooks environment", "votes": null }, { "id": "1271547", "postDate": "04/12/2021 17:26:16", "content": "

I am new to this kaggle env, hence when I saw the comments I realized that the version that I submitted didn't had GPU turned on.

", "rawMarkdown": "I am new to this kaggle env, hence when I saw the comments I realized that the version that I submitted didn't had GPU turned on.", "votes": null }, { "id": "1271550", "postDate": "04/12/2021 17:27:47", "content": "

it depends upon various factors , whether you are doing TTA or not , what is your batch size , image_size, how big is your model etc etc . for me i am using 256*256 image size , batch_size = 64 , with no TTA

", "rawMarkdown": "it depends upon various factors , whether you are doing TTA or not , what is your batch size , image_size, how big is your model etc etc . for me i am using 256*256 image size , batch_size = 64 , with no TTA", "votes": null }, { "id": "1272780", "postDate": "04/13/2021 18:43:45", "content": "

I have 2 types of submission. The heavy one, when including all post processing, takes near 3 hours, so it's a close call when running on the private dataset. The other one is under one hour.

", "rawMarkdown": "I have 2 types of submission. The heavy one, when including all post processing, takes near 3 hours, so it's a close call when running on the private dataset. The other one is under one hour.", "votes": null }, { "id": "1272822", "postDate": "04/13/2021 19:36:37", "content": "

I wonder what type of post processing you would be doing 😲 !!

", "rawMarkdown": "I wonder what type of post processing you would be doing 😲 !!", "votes": null }, { "id": "1273170", "postDate": "04/14/2021 06:25:05", "content": "

My heavy one takes over 8 hours! But is this a problem? I assumed those 3 hours for you (8 for me) includes private test set inference, but we only have the public one displayed at this stage?

\n

Are you saying we need to aim for our notebooks to finish within 4 hours, because later they will be run on the private set separately?

", "rawMarkdown": "My heavy one takes over 8 hours! But is this a problem? I assumed those 3 hours for you (8 for me) includes private test set inference, but we only have the public one displayed at this stage?\n\nAre you saying we need to aim for our notebooks to finish within 4 hours, because later they will be run on the private set separately?", "votes": null } ], "comments": [ { "id": 1271482, "author_name": "victorasso", "author_url": "", "post_date": "04/12/2021 16:46:02", "content": "

Mine can take up to 2h without GPU acceleration, i will use gpu on the next commit to see the changes and let you know

", "votes": null, "replies": [ { "id": 1271489, "author_name": "deveshdarshan", "author_url": "", "post_date": "04/12/2021 16:54:45", "content": "

I just realised I was also not using gpu, but still on cpu it is over 4 hours now. I don't know what I am doing wrong. I'll try with gpu on this time.

", "votes": null, "replies": [] }, { "id": 1271500, "author_name": "victorasso", "author_url": "", "post_date": "04/12/2021 17:02:54", "content": "

Ok i just tried on GPU and now its going pretty fast (<15min), i guess i just have to get used to this kaggle env, to be honest i find it a little bit confusing.

", "votes": null, "replies": [] }, { "id": 1271529, "author_name": "deveshdarshan", "author_url": "", "post_date": "04/12/2021 17:17:35", "content": "

Yeah, I also tried on GPU and it took only 35 minutes while my CPU one is still running.

", "votes": null, "replies": [] }, { "id": 1271539, "author_name": "trooperog", "author_url": "", "post_date": "04/12/2021 17:21:47", "content": "

but why you guys running your inference code on cpu ?? any specific reason for that ?

", "votes": null, "replies": [] }, { "id": 1271544, "author_name": "victorasso", "author_url": "", "post_date": "04/12/2021 17:25:27", "content": "

Well my case is that i didn't know any better 😂

\n

I won't do it again now that i know a little better my ways around this kaggle notebooks environment

", "votes": null, "replies": [] }, { "id": 1271547, "author_name": "deveshdarshan", "author_url": "", "post_date": "04/12/2021 17:26:16", "content": "

I am new to this kaggle env, hence when I saw the comments I realized that the version that I submitted didn't had GPU turned on.

", "votes": null, "replies": [] } ] }, { "id": 1271483, "author_name": "trooperog", "author_url": "", "post_date": "04/12/2021 16:47:38", "content": "

mine takes roughly 20 - 30 mins using gpu for 5 folds of a single model

", "votes": null, "replies": [ { "id": 1271533, "author_name": "deveshdarshan", "author_url": "", "post_date": "04/12/2021 17:19:26", "content": "

I also ran it on GPU but it took 35 minutes for a single fold, can you suggest some tips to reduce this time ?

", "votes": null, "replies": [] }, { "id": 1271550, "author_name": "trooperog", "author_url": "", "post_date": "04/12/2021 17:27:47", "content": "

it depends upon various factors , whether you are doing TTA or not , what is your batch size , image_size, how big is your model etc etc . for me i am using 256*256 image size , batch_size = 64 , with no TTA

", "votes": null, "replies": [] } ] }, { "id": 1272780, "author_name": "andrasferenczi", "author_url": "", "post_date": "04/13/2021 18:43:45", "content": "

I have 2 types of submission. The heavy one, when including all post processing, takes near 3 hours, so it's a close call when running on the private dataset. The other one is under one hour.

", "votes": null, "replies": [ { "id": 1272822, "author_name": "deveshdarshan", "author_url": "", "post_date": "04/13/2021 19:36:37", "content": "

I wonder what type of post processing you would be doing 😲 !!

", "votes": null, "replies": [] }, { "id": 1273170, "author_name": "jamesphoward", "author_url": "", "post_date": "04/14/2021 06:25:05", "content": "

My heavy one takes over 8 hours! But is this a problem? I assumed those 3 hours for you (8 for me) includes private test set inference, but we only have the public one displayed at this stage?

\n

Are you saying we need to aim for our notebooks to finish within 4 hours, because later they will be run on the private set separately?

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1271370": "Can someone share how much time their notebook took for inference after submitting ?\nCurrently mine is over 2 hours and still showing status as \"Notebook Running\".", "1271482": "Mine can take up to 2h without GPU acceleration, i will use gpu on the next commit to see the changes and let you know", "1271483": "mine takes roughly 20 - 30 mins using gpu for 5 folds of a single model", "1271489": "I just realised I was also not using gpu, but still on cpu it is over 4 hours now. I don't know what I am doing wrong. I'll try with gpu on this time.", "1271500": "Ok i just tried on GPU and now its going pretty fast (<15min), i guess i just have to get used to this kaggle env, to be honest i find it a little bit confusing.", "1271529": "Yeah, I also tried on GPU and it took only 35 minutes while my CPU one is still running.", "1271533": "I also ran it on GPU but it took 35 minutes for a single fold, can you suggest some tips to reduce this time ?", "1271539": "but why you guys running your inference code on cpu ?? any specific reason for that ?", "1271544": "Well my case is that i didn't know any better 😂\n\nI won't do it again now that i know a little better my ways around this kaggle notebooks environment", "1271547": "I am new to this kaggle env, hence when I saw the comments I realized that the version that I submitted didn't had GPU turned on.", "1271550": "it depends upon various factors , whether you are doing TTA or not , what is your batch size , image_size, how big is your model etc etc . for me i am using 256*256 image size , batch_size = 64 , with no TTA", "1272780": "I have 2 types of submission. The heavy one, when including all post processing, takes near 3 hours, so it's a close call when running on the private dataset. The other one is under one hour.", "1272822": "I wonder what type of post processing you would be doing 😲 !!", "1273170": "My heavy one takes over 8 hours! But is this a problem? I assumed those 3 hours for you (8 for me) includes private test set inference, but we only have the public one displayed at this stage?\n\nAre you saying we need to aim for our notebooks to finish within 4 hours, because later they will be run on the private set separately?" }, "source": "meta" }