{ "id": 207808, "title": "Last Week Extra Compute Resources? :)", "url": "/competitions/riiid-test-answer-prediction/discussion/207808", "author_name": "", "post_date": "2020-12-31T10:59:54.505275900Z", "votes": 9, "comment_count": 12, "views": 0, "content": "

Firstly I'd like to appreciate all the free compute resources that Kaggle has already provided. Couldn't have dreamed of training such deep learning models on my laptop. Kudos to the whole Kaggle team 😊

\n

However, as we are entering the competition's final week, can we get a new year's gift of some extra GPU/TPU compute time just for the final's week. Would help a lot with finalizing the models for submission.

", "messages": [ { "id": "1133587", "postDate": "12/31/2020 10:59:54", "content": "

Firstly I'd like to appreciate all the free compute resources that Kaggle has already provided. Couldn't have dreamed of training such deep learning models on my laptop. Kudos to the whole Kaggle team 😊

\n

However, as we are entering the competition's final week, can we get a new year's gift of some extra GPU/TPU compute time just for the final's week. Would help a lot with finalizing the models for submission.

", "rawMarkdown": "Firstly I'd like to appreciate all the free compute resources that Kaggle has already provided. Couldn't have dreamed of training such deep learning models on my laptop. Kudos to the whole Kaggle team 😊\n\nHowever, as we are entering the competition's final week, can we get a new year's gift of some extra GPU/TPU compute time just for the final's week. Would help a lot with finalizing the models for submission.", "votes": null }, { "id": "1133708", "postDate": "12/31/2020 13:05:53", "content": "

Maybe it won’t be extended, you can rent a server or find a teammate😄

", "rawMarkdown": "Maybe it won’t be extended, you can rent a server or find a teammate😄", "votes": null }, { "id": "1133719", "postDate": "12/31/2020 13:14:26", "content": "

congrats on breaking 0.79 mark :)

", "rawMarkdown": "congrats on breaking 0.79 mark :)", "votes": null }, { "id": "1133735", "postDate": "12/31/2020 13:29:44", "content": "

You can try google cloud platform, which provides up to 64GB memory for free use about 1 month. You can even extend more by creating swap memory.

", "rawMarkdown": "You can try google cloud platform, which provides up to 64GB memory for free use about 1 month. You can even extend more by creating swap memory.", "votes": null }, { "id": "1133748", "postDate": "12/31/2020 13:42:01", "content": "

Maybe in your kaggle e-mail or in your team merge request is your solution!

", "rawMarkdown": "Maybe in your kaggle e-mail or in your team merge request is your solution!", "votes": null }, { "id": "1133775", "postDate": "12/31/2020 14:17:33", "content": "

Thanks mate

", "rawMarkdown": "Thanks mate", "votes": null }, { "id": "1134136", "postDate": "12/31/2020 22:14:13", "content": "

In my case I needed more memory to make the ensemble…

", "rawMarkdown": "In my case I needed more memory to make the ensemble...", "votes": null }, { "id": "1134268", "postDate": "01/01/2021 04:59:00", "content": "

i get only 30 hrs of gpu but i remember i used o get 40-42 hours also some times not sure how is that decided.

\n

Overall there is lot things need to be reformed to make kaggle platform giving fair oppurtunity to all particiapants in terms of resources especially. Resource rich guys are able to do lot of things that much more capable people cant when they dont have. Kaggle needs to set up resource limit for RAM/GPU as per need of computing needs of competition.

\n

Some competition dont need much ram,gpu ,but some need accordingly resource limits should be set up. For this competition 18 gb should have been an APT ram ,48 hours of gpu also. if one is not using TPU they should have option to use those many GPUs hour or vice versa.Every week after first reset, one can chose GPU Only ,which should be irreversible option for the week so TPU hours go to GPU or viceversa. What diff does it makes

\n

Not sure such feedback just hit the Ignoring eyes or they are do taken in account for improvement.
\n@sohier @addisson

", "rawMarkdown": "i get only 30 hrs of gpu but i remember i used o get 40-42 hours also some times not sure how is that decided. \n\nOverall there is lot things need to be reformed to make kaggle platform giving fair oppurtunity to all particiapants in terms of resources especially. Resource rich guys are able to do lot of things that much more capable people cant when they dont have. Kaggle needs to set up resource limit for RAM/GPU as per need of computing needs of competition. \n\nSome competition dont need much ram,gpu ,but some need accordingly resource limits should be set up. For this competition 18 gb should have been an APT ram ,48 hours of gpu also. if one is not using TPU they should have option to use those many GPUs hour or vice versa.Every week after first reset, one can chose GPU Only ,which should be irreversible option for the week so TPU hours go to GPU or viceversa. What diff does it makes\n\nNot sure such feedback just hit the Ignoring eyes or they are do taken in account for improvement.\n@sohier @addisson", "votes": null }, { "id": "1134269", "postDate": "01/01/2021 05:00:39", "content": "

It is very expensive compared to just using GCP platform.While configuring upgrade you can check

", "rawMarkdown": "It is very expensive compared to just using GCP platform.While configuring upgrade you can check", "votes": null }, { "id": "1134306", "postDate": "01/01/2021 05:56:37", "content": "

Compute can give an advantage, not the wining solution always… And it's better this way only, Learning to do more with less :)

", "rawMarkdown": "Compute can give an advantage, not the wining solution always... And it's better this way only, Learning to do more with less :)", "votes": null }, { "id": "1135371", "postDate": "01/02/2021 07:45:37", "content": "

it's 40 hours GPU now,

", "rawMarkdown": "it's 40 hours GPU now,", "votes": null }, { "id": "1135377", "postDate": "01/02/2021 07:52:29", "content": "

Yeah. I just noticed as well. 👍

", "rawMarkdown": "Yeah. I just noticed as well. 👍", "votes": null }, { "id": "1135478", "postDate": "01/02/2021 09:29:38", "content": "

Actually its on usage ,if some one dsnt uses much he or she gets less next week ,if you exhaust this week ,you will get 42 hrs ,which is max so far I hqvevseen in other competitions .

\n

Eagerly wait for option to transfer usage hours of tpu to gpu at beginning of week .
\nBut perhaps Google will see loss of biz with this so kaggke will never provide that option

", "rawMarkdown": "Actually its on usage ,if some one dsnt uses much he or she gets less next week ,if you exhaust this week ,you will get 42 hrs ,which is max so far I hqvevseen in other competitions .\n\nEagerly wait for option to transfer usage hours of tpu to gpu at beginning of week .\nBut perhaps Google will see loss of biz with this so kaggke will never provide that option", "votes": null } ], "comments": [ { "id": 1133708, "author_name": "lzcabc123456", "author_url": "", "post_date": "12/31/2020 13:05:53", "content": "

Maybe it won’t be extended, you can rent a server or find a teammate😄

", "votes": null, "replies": [] }, { "id": 1133719, "author_name": "aman1391", "author_url": "", "post_date": "12/31/2020 13:14:26", "content": "

congrats on breaking 0.79 mark :)

", "votes": null, "replies": [ { "id": 1133775, "author_name": "abdurrafae", "author_url": "", "post_date": "12/31/2020 14:17:33", "content": "

Thanks mate

", "votes": null, "replies": [] } ] }, { "id": 1133735, "author_name": "louieshao", "author_url": "", "post_date": "12/31/2020 13:29:44", "content": "

You can try google cloud platform, which provides up to 64GB memory for free use about 1 month. You can even extend more by creating swap memory.

", "votes": null, "replies": [ { "id": 1134269, "author_name": "jaideepvalani", "author_url": "", "post_date": "01/01/2021 05:00:39", "content": "

It is very expensive compared to just using GCP platform.While configuring upgrade you can check

", "votes": null, "replies": [] } ] }, { "id": 1133748, "author_name": "coreacasa", "author_url": "", "post_date": "12/31/2020 13:42:01", "content": "

Maybe in your kaggle e-mail or in your team merge request is your solution!

", "votes": null, "replies": [] }, { "id": 1134136, "author_name": "felipefonte99", "author_url": "", "post_date": "12/31/2020 22:14:13", "content": "

In my case I needed more memory to make the ensemble…

", "votes": null, "replies": [] }, { "id": 1134268, "author_name": "jaideepvalani", "author_url": "", "post_date": "01/01/2021 04:59:00", "content": "

i get only 30 hrs of gpu but i remember i used o get 40-42 hours also some times not sure how is that decided.

\n

Overall there is lot things need to be reformed to make kaggle platform giving fair oppurtunity to all particiapants in terms of resources especially. Resource rich guys are able to do lot of things that much more capable people cant when they dont have. Kaggle needs to set up resource limit for RAM/GPU as per need of computing needs of competition.

\n

Some competition dont need much ram,gpu ,but some need accordingly resource limits should be set up. For this competition 18 gb should have been an APT ram ,48 hours of gpu also. if one is not using TPU they should have option to use those many GPUs hour or vice versa.Every week after first reset, one can chose GPU Only ,which should be irreversible option for the week so TPU hours go to GPU or viceversa. What diff does it makes

\n

Not sure such feedback just hit the Ignoring eyes or they are do taken in account for improvement.
\n@sohier @addisson

", "votes": null, "replies": [ { "id": 1134306, "author_name": "adityaecdrid", "author_url": "", "post_date": "01/01/2021 05:56:37", "content": "

Compute can give an advantage, not the wining solution always… And it's better this way only, Learning to do more with less :)

", "votes": null, "replies": [] } ] }, { "id": 1135371, "author_name": "yangxiaoshuai", "author_url": "", "post_date": "01/02/2021 07:45:37", "content": "

it's 40 hours GPU now,

", "votes": null, "replies": [ { "id": 1135377, "author_name": "abdurrafae", "author_url": "", "post_date": "01/02/2021 07:52:29", "content": "

Yeah. I just noticed as well. 👍

", "votes": null, "replies": [] }, { "id": 1135478, "author_name": "jaideepvalani", "author_url": "", "post_date": "01/02/2021 09:29:38", "content": "

Actually its on usage ,if some one dsnt uses much he or she gets less next week ,if you exhaust this week ,you will get 42 hrs ,which is max so far I hqvevseen in other competitions .

\n

Eagerly wait for option to transfer usage hours of tpu to gpu at beginning of week .
\nBut perhaps Google will see loss of biz with this so kaggke will never provide that option

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1133587": "Firstly I'd like to appreciate all the free compute resources that Kaggle has already provided. Couldn't have dreamed of training such deep learning models on my laptop. Kudos to the whole Kaggle team 😊\n\nHowever, as we are entering the competition's final week, can we get a new year's gift of some extra GPU/TPU compute time just for the final's week. Would help a lot with finalizing the models for submission.", "1133708": "Maybe it won’t be extended, you can rent a server or find a teammate😄", "1133719": "congrats on breaking 0.79 mark :)", "1133735": "You can try google cloud platform, which provides up to 64GB memory for free use about 1 month. You can even extend more by creating swap memory.", "1133748": "Maybe in your kaggle e-mail or in your team merge request is your solution!", "1133775": "Thanks mate", "1134136": "In my case I needed more memory to make the ensemble...", "1134268": "i get only 30 hrs of gpu but i remember i used o get 40-42 hours also some times not sure how is that decided. \n\nOverall there is lot things need to be reformed to make kaggle platform giving fair oppurtunity to all particiapants in terms of resources especially. Resource rich guys are able to do lot of things that much more capable people cant when they dont have. Kaggle needs to set up resource limit for RAM/GPU as per need of computing needs of competition. \n\nSome competition dont need much ram,gpu ,but some need accordingly resource limits should be set up. For this competition 18 gb should have been an APT ram ,48 hours of gpu also. if one is not using TPU they should have option to use those many GPUs hour or vice versa.Every week after first reset, one can chose GPU Only ,which should be irreversible option for the week so TPU hours go to GPU or viceversa. What diff does it makes\n\nNot sure such feedback just hit the Ignoring eyes or they are do taken in account for improvement.\n@sohier @addisson", "1134269": "It is very expensive compared to just using GCP platform.While configuring upgrade you can check", "1134306": "Compute can give an advantage, not the wining solution always... And it's better this way only, Learning to do more with less :)", "1135371": "it's 40 hours GPU now,", "1135377": "Yeah. I just noticed as well. 👍", "1135478": "Actually its on usage ,if some one dsnt uses much he or she gets less next week ,if you exhaust this week ,you will get 42 hrs ,which is max so far I hqvevseen in other competitions .\n\nEagerly wait for option to transfer usage hours of tpu to gpu at beginning of week .\nBut perhaps Google will see loss of biz with this so kaggke will never provide that option" }, "source": "meta" }