New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Test "hybrid" 6B weights on CGOS #954
Comments
@gcp Maybe we can try "hybrid" in the end of 6 block, after hybrid -> self-play -> train several weights -> hybrid -> ..., maybe we can see the true limit of 6 block. |
Now I run W25-PO1600 and W25-NOLIMIT on CGOS, the W25 file is inherited from 3 weights after No.89(53f4cc79) , win rate to No.89 is 40% - 50%, weight file and po 200 match log is below(Thanks 飞 to hybrid) We can wait the CGOS result for a while... |
This is cool. I hope we try promoting a hybrid one at end of 6 blocks. When you make w26 I'll run no limit on a 1080ti on Windows if you want. |
@jjoshua2 welcome~ If we find a new stronger hybrid weight, I will post here . |
I tried to use it on Windows and haven't gotten it to work yet. Gogui
requires java and there's weird character encoding errors
…On Mar 1, 2018 12:40 AM, "pangafu" ***@***.***> wrote:
@jjoshua2 <https://github.com/jjoshua2> welcome~ If we find a new
stronger hybrid weight, I will post here .
also you can hybrid by yourself with tools:
https://github.com/pangafu/Hybrid_LeelaZero
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#954 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AO6INOYIyqACSDIC3R86J3T0kzEAL_wmks5tZ4nhgaJpZM4SXpqy>
.
|
I have run an even game of w24 and AQ and W24 won. This hybrid weight is like always created two major battle fields at same time and made opponents confused. |
@jjoshua2 maybe you should download java to run gogui. |
To see the true power of the hybrid, I hybrid 2 weights and run it on cgos.
6w24_081&aed&a91&92d&9ac&8ee&f66&4d5&5e4.zip
6w23_f66&081&9ac&8ee&4d5&5e4_1&0.5&0.5&0.5&0.5&0.5.zip
W23 is inherited from No.86 weight (9ac646d7) and other 5 weight after it , W24 is inherited from No.86(9ac646d7) and No.87(92dd0397) and other 7 weights between them.
And you can see the PO 200 match log in the zip file.
LZ-HY-W23-PO1600: -g -p 1600 --noponder -t 1 -q -d -r 1 -w
LZ-HY-W24-PO1600: -g -p 1600 --noponder -t 1 -q -d -r 1 -w
LZ-HY-W24-NOLIMIT: -g --noponder -t 4 -q -d -r 1 -w
My test machine is 1070 + ubuntu, there is many other leela zero weights run on cgos, such as LZ-0xx-p1600-t1-r1, StellarLZ-p6400, LeelaZero-test... I think LZ-HY-W23-PO1600 's parameter is same as LZ-0xx-p1600-t1-r1, so the result of them can compare.
LZ-HY-W24-PO1600 3030
LZ-HY-W23-PO1600 3011
And other LZ-0XX-p1600-t1-r1 is not close to static(<100 games), the score of them are all bellow the 2 hybrid weights, and to my experiment, the score of LZ-0XX-p1600-t1-r1 will continue to drop during 50 to 100 games process.
So after several test, in my opinion:
The text was updated successfully, but these errors were encountered: