KataGo b60c320-20210424 playouts 1, komi 0 - CS Zero 9d:
| Attachments: |
CS Zero - KataGo.sgf [1.7 KiB]
Downloaded 2045 times |
| Life In 19x19 http://www.lifein19x19.com/ |
|
| KataGo Distributed Training and new networks http://www.lifein19x19.com/viewtopic.php?f=18&t=17944 |
Page 6 of 22 |
| Author: | Zeed [ Wed Mar 17, 2021 5:23 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
Where do u take those 60b nets? I searched all over https://katagotraining.org/networks/ and didn't find any. And how much stronger this 60b in comparison with 40b 697? |
|
| Author: | And [ Wed Mar 17, 2021 8:04 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
while there are only two networks 60 blocks: https://media.katagotraining.org/upload ... 221.bin.gz https://media.katagotraining.org/upload ... 314.bin.gz |
|
| Author: | Zeed [ Wed Mar 17, 2021 12:01 pm ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
Yeah, I know, but those links are not to the site, but download links. And where do you find them? I couldn't find anything about 60b nets on the katago site (https://katagotraining.org/networks/). Does someone upload them on github or what? Is there any other source where I can look for new ones when they eventually come, other than you and go4thewin who uploaded previous 60b net? |
|
| Author: | lightvector [ Wed Mar 17, 2021 4:04 pm ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
Join the discord server: https://discord.gg/bqkZAz3 60b is not ready to officially take over 40b because it's not strong enough yet given the extra compute cost, these are nets that are posted to see how it's progressing. Join discord if you care about news or discussions of these things, or you want to know when cool stuff is posted (nets, experimental branches of code, stats, test results, games) |
|
| Author: | Zeed [ Thu Mar 18, 2021 9:33 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
That's awesome, thanks! |
|
| Author: | And [ Sat Apr 03, 2021 4:32 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
New b60 net https://media.katagotraining.org/upload ... 898.bin.gz |
|
| Author: | And [ Mon Apr 19, 2021 6:06 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
KataGo v1.8.2 "This is a minor release mainly of interest to contributors to KataGo's distributed run, or to users who run KataGo self-play training on their own GPUs. This release doesn't make any changes to the KataGo engine itself, but it does fix an issue that was believed to be limiting the diversity in KataGo's self-play games. Switching to this version should, over the long term of training, improve KataGo's learning, particularly on small boards, as well as enable a few further parameter changes in the future once most people have upgraded, which should also further improve opening diversity." https://github.com/lightvector/KataGo/releases |
|
| Author: | And [ Sun Apr 25, 2021 4:58 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
New b60 nets http://media.katagotraining.org/uploade ... 932.bin.gz https://media.katagotraining.org/upload ... 424.bin.gz https://discord.com/channels/4170221623 ... 8804732928 |
|
| Author: | And [ Mon Apr 26, 2021 8:09 am ] | ||
| Post subject: | Re: KataGo Distributed Training and new networks | ||
KataGo b60c320-20210424 playouts 1, komi 0 - CS Zero 9d:
|
|||
| Author: | And [ Mon Apr 26, 2021 2:45 pm ] | ||
| Post subject: | Re: KataGo Distributed Training and new networks | ||
KataGo b60c320-s275 playouts 1 - CS Zero 9d H2:
|
|||
| Author: | ez4u [ Wed Apr 28, 2021 5:34 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
And wrote: KataGo v1.8.2 "This is a minor release mainly of interest to contributors to KataGo's distributed run, or to users who run KataGo self-play training on their own GPUs. This release doesn't make any changes to the KataGo engine itself, but it does fix an issue that was believed to be limiting the diversity in KataGo's self-play games. Switching to this version should, over the long term of training, improve KataGo's learning, particularly on small boards, as well as enable a few further parameter changes in the future once most people have upgraded, which should also further improve opening diversity." https://github.com/lightvector/KataGo/releases The new v1.8.2 seems to run much slower than 1.8.1 on my machine. I am running contribute on windows 10 with an Nvidia GTX1650 super (no tensor cores, no FP16) using the opencl version. Moving from 1.8.1 to 1.8.2, I am seeing: -33% on plays per second, -30% on nn evals per second, and +61% on seconds per game. This is from the information in the log files. |
|
| Author: | And [ Wed Apr 28, 2021 6:27 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
I compared versions of eigenavx2 - 1.8.2 10% slower, several moves 20% slower, watched over 100 moves. The default_gtp.cfg files are the same, network 20b, 10 playouts 
|
|
| Author: | And [ Wed Apr 28, 2021 6:50 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
OpenCL version 1.8.2 is 2.5 times slower! but there are differences in opencltuning files (GT 610) EDIT I replaced the opencltuning file, nothing changed. 1.8.2 is also 2.5x slower! The default_gtp.cfg files are the same, network 20b, 10 playouts |
|
| Author: | ez4u [ Wed Apr 28, 2021 6:58 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
When I run the cuda version on google colab, I do not see any difference. If anything, 1.8.2 might be a little faster. However, that was with quite short runs versus over 24 hours when I looked at the opencl versions on my machine. |
|
| Author: | And [ Wed Apr 28, 2021 7:01 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
obviously something is wrong 
|
|
| Author: | lightvector [ Thu Apr 29, 2021 4:06 pm ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
We've figured out what's going on - it's not a matter of the version, but apparently on some GPUs KataGo's tuning is noisy enough that it it has an occasional chance of picking to not use FP16 storage when FP16 storage is beneficial, so on those GPUs, every time you manually re-tune you have a chance of getting different performance. I'm not entirely sure how to fix it without making the OpenCL tuning process heavier and slower. But definitely something I'll think about for a future release - @ez4u thanks for the report and being so careful to check and notice things.  Also, is there a place in the docs where I can write or fix to avoid giving the impression that people have to manually retune each version? Although, I guess it was really good that you did anyways, otherwise we wouldn't have known about this quirk about tuning.
|
|
| Author: | And [ Mon May 03, 2021 6:58 am ] | ||
| Post subject: | Re: KataGo Distributed Training and new networks | ||
KataGo s790 playouts 1 - CS Zero 9d H2:
|
|||
| Author: | And [ Tue May 11, 2021 12:10 am ] | ||
| Post subject: | Re: KataGo Distributed Training and new networks | ||
KataGo s800 playouts 1 - CS Zero 9d H3:
|
|||
| Author: | And [ Thu May 20, 2021 3:49 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
New b60 net https://media.katagotraining.org/upload ... 170.bin.gz |
|
| Author: | And [ Sat Jun 19, 2021 5:57 am ] |
| Post subject: | Re: KataGo Distributed Training and new networks |
New b60 net https://media.katagotraining.org/upload ... 705.bin.gz |
|
| Page 6 of 22 | All times are UTC - 8 hours [ DST ] |
| Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |
|