[Most-ai-contest] The deadline of final version FGC QA system

日 4月 5 12:33:06 CST 2020

If you need 國網中心 GPUs, please send your 國網中心 account to me. I will join your account to us (then you can create your container).

If you didn’t have a 國網中心 account, please register it first (as https://iservice.nchc.org.tw/nchc_service/index.php)

---ChaoChun

寄件者: kysu <kysu at iis.sinica.edu.tw>
日期: 2020年4月5日 星期日 下午12:27
收件者: 'Menphis Chen' <kychen at mail.ntust.edu.tw>, '范正忠' <jjfan at iis.sinica.edu.tw>
副本: 'Most-ai Contest' <Most-ai-contest at iis.sinica.edu.tw>, 'kysu' <kysu at iis.sinica.edu.tw>, 'ChaoChun Liang' <ccliang at iis.sinica.edu.tw>
主旨: RE: [Most-ai-contest] The deadline of final version FGC QA system

Sorry for not explaining my request clearly enough. What we want to know is the amount of required GPU time to train a RoBERTA-Large model (In my memory, Mr. Luo/Kuo/Liao had done it before, and should have a rough idea). For example, if we need 2 days to train a RoBERTA-Large model on a RTX2080TI-11G GPU card, then the person who wants to adopt ensemble approach must freeze his code now and train various classifiers in parallel. Since 國網中心has more advanced GPU cards with 32G memory each, it is strongly suggested that the member who wants to adopt ensemble approach to train his model via 國網中心 GPUs in parallel. The person who wants to use 國網中心 GPUs but has not got its account, please directly contact Mr. Chao-Chun Liang (I have attached his email to the CC-list), and do it asap. We don’t have much time left before freezing the version. Thanks for your effort and contribution in advance!

KY

From: Menphis Chen [mailto:kychen at mail.ntust.edu.tw] 
Sent: Sunday, April 5, 2020 12:02 PM
To: 范正忠 <jjfan at iis.sinica.edu.tw>
Cc: kysu <kysu at iis.sinica.edu.tw>; Most-ai Contest <Most-ai-contest at iis.sinica.edu.tw>
Subject: Re: [Most-ai-contest] The deadline of final version FGC QA system

Dear all,

According to Prof. Su’s idea, We would like to set a deadline of further improving the model, and need to know how much training-time would be required to train a RoBERT-Large model (say, for the task of single-span extraction). 

Based on Dr. Fan’s schedule, we should definitely complete each model before the next Tuesday (4/7) morning 8:00AM.

Hope we can further fine tune the joint model, and achieve a better result!

Best, Menphis.

范正忠 <jjfan at iis.sinica.edu.tw> 於 2020年4月5日 09:02 寫道：

Dear all,

Before you start train all FGC data set ( train/dev/test), please send me your final best model trained on training set only (i.e. best testing performance).

Thanks.

Best,

jjfan

From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "kysu" <kysu at iis.sinica.edu.tw>
Cc: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Friday, April 3, 2020 5:45:51 PM
Subject: Re: [Most-ai-contest] The deadline of final version FGC QA system

Dear all,

The next Monday (4/6) please merge all FGC data-set (train/dev/test) for fine-tuning to generate their final version, and keep the training-set performance about 90 to 95%. Thanks.

Best,

jjfan

From: "范正忠" <jjfan at iis.sinica.edu.tw>
To: "kysu" <kysu at iis.sinica.edu.tw>
Cc: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Sent: Friday, April 3, 2020 12:25:21 PM
Subject: [Most-ai-contest] The deadline of final version FGC QA system

Dear all,

To get as much as time we have to improve the system, I will  freeze the code/model of each module before the next Tuesday (4/7) morning 8:00AM.

Best,

jjfan

From: "kysu" <kysu at iis.sinica.edu.tw>
To: "Most-ai Contest" <Most-ai-contest at iis.sinica.edu.tw>
Cc: "范正忠" <jjfan at iis.sinica.edu.tw>, "kysu" <kysu at iis.sinica.edu.tw>
Sent: Wednesday, April 1, 2020 11:55:35 AM
Subject: We need your help/effort during the coming long weekend

Dear all,

Thanks a lot for your great effort and contribution in the past few weeks. However, according to Dr. JJ Fan, our performance is still behind another team. Therefore, we would like to encourage you to try your best to further improve the performance during this coming long weekend. 

Dr. Fan and I will check/watch the performance, and then inform you the bottleneck of our system. Please check your email/Line for Dr. Fan’s question/request from time to time. Thanks a lot for your help in advance. Look forward to getting your quick response/action later.

KY

---------------------------------------------------

Dr. Keh-Yih Su

Institute of Information Science, Academia Sinica

128 Academia Road, Section 2, Nankang 

Taipei 11529, Taiwan

(O) 02-2788-3799 ext. 1801

(M) 0963-823052

Email:  kysu at iis.sinica.edu.tw (or bdc.kysu at gmail.com)

---------------------------------------------------

_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest

_______________________________________________
Most-ai-contest mailing list
Most-ai-contest at iis.sinica.edu.tw
https://www.iis.sinica.edu.tw/mailman/listinfo/most-ai-contest

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.iis.sinica.edu.tw/pipermail/most-ai-contest/attachments/20200405/6fa1a6e2/attachment-0001.html>