-
How can I get this to work with the very strong (3500+ ELO) 256*20 network trained with Leela Zero that was published by someone a while back?
-
Hi,
(I did not see any piece of information about this neither in the README or in open issues)
The training games are played with training specific settings:
-Nb playouts (1600)
-Noise (for explo…
-
跑的时候GPU内存占满了,但GPU运算部件的利用率只有12%,不知道可以增大利用率吗
-
https://github.com/suragnair/alpha-zero-general/blob/38fc2983a1dd85104348d72420284daa6a2df469/Coach.py#L69
What is the main goal of collecting all examples from all iterations? On one hand we get m…
-
Currently, each move is given remaining_time/moves_remaining in main time. See:
https://github.com/gcp/leela-zero/blob/next/src/TimeControl.cpp#L186-L190
and
https://github.com/gcp/leela-zero/blob/…
-
1. https://github.com/zakki/Ray/releases
rlo-0.1.1 (formly rn-4.32)
@zakki zakki released this 20 hours ago
Assets
83.2 MB
Rn.zip
Source code (zip)
Source code (tar.gz)
R…
-
Would it be worth applying the same kind of symmetry-exploiting that goes on in AlphaGo (Zero) with respect to king- and queenside? Castling would not be an issue, because it can still be considered a…
-
Now that its shown NN is scale invariant at this size, I imagine it wouldnt siphone too many client cycles away from the main 19 by 19 and could easily be done within a few weeks and we can get a supe…
-
一直不懂这个参数的意思。有什么作用?
-
Serious offer, $500 USD, half up-front and the second half when work is completed and results published online to the open public community. Basically it will be more or less a redo of bjiyxo's 20b 25…