It drew against me. It had an extra R but couldn't convert (it was RB vs. Bpp, with Bs of opposite colors and my B securely guarding the pawns, which stood on a5 and b4. I played white).
Leela Zero( A Neural Network engine similar to Alpha Zero)
Sort:
For voluntary contribution on testing, it is possible to run both Leela and python(stockfish testing framework) simultaneously. Oops! 
I wonder about the hard coded number of used threads. You can see the client.exe calls Leela with -t1, but with one thread the CPU and GPU are only used partly (50% GPU in my case). With two or three threads more training games can be played in the same time.
(A Leela programmer's answer to your question)
If you're talking about the network training, just run another instance of the training (clone the existing folder to keep the instances separated).
If you're talking about letting the engine play (against human or artificial opponents), increase the threads assigned to the engine with the -t command line parameter (default value is 2). Note that this method doesn't work for the training, because the thread count is limited to 1 for that.
https://github.com/glinscott/leela-chess/wiki/Running-LCZero-in-a-Chess-GUI
Introduction to Leela and how to install Leela.
Leela is getting stronger and stronger day by day, she is close to 2000 rating!! 
Mar 31, 2018
0
#48
drmrboss wrote:
Leela is getting stronger and stronger day by day, she is close to 2000 rating!!
This graph was posted above: it looks like a plateau at the end. Is is not? Click to enlarge
There was plateau in a few days ago due to bug , but the trend is going up again today! This updated graph is 4 hours later from the last graph I mentioned above,
Elroch wrote:
drmrboss wrote:
Leela is getting stronger and stronger day by day, she is close to 2000 rating!!
This graph was posted above: it looks like a plateau at the end. Is is not? Click to enlarge
In the graph of drmrboss the line goes linearly up now. Later it will become harder to make progress. You can compare it with zero.sjeng.org were you can see multiple plateaus. (They looked more than a plateau at the time any progress needed a lot of time.)
I expect the same in LCZ, but maybe I am wrong.
After 850,000 games a bug was fixed.
After 3,000,000 games another bug was fixed, it was switched to a larger network and there were some performance optimizations.
After 4,800,000 games the network size was increased again.
The single two blue dots at the top are results from even larger networks.
And than there are hurdles which must be taken, for example because it becomes clear that a tactic no longer works and must be forgotten or replaced by a better one. Maybe at 1,360,000 there was some.
Both projects doing a test run to be able to optimize and find bugs. In LZ it was not planned to continue the test run for such a long time, but the result is just to good and retraining a larger network with the same old data seems to work quite well. And I am amazed about the fast progress of LCZ, cannot wait for 3,000 (real) ELO and some full strength matches with a single GPU against SF. Some weeks or months … (or some hours for Google 
)
Leela is comparable to 2700 engines positional understanding in opening. This test suit showed (copy from talkchess forum) Leela did 45% correction
or 2700 rating in positional understanding. As usual, Neural Networks's weakness is tactics but very strong in positional understanding like human(meanwhile stockfish and minimax engines are strong in tactics but weak in positional understanding). Leela will scale better if there is more time for tactical calculation.
Hi, has anyone here have a mac they were able to get Leela working on? I am stuck :/
Mar 31, 2018
0
#53
drmrboss wrote:
Leela is comparable to 2700 engines positional understanding in opening. This test suit showed (copy from talkchess forum) Leela did 45% correction or 2700 rating in positional understanding. As usual, Neural Networks's weakness is tactics but very strong in positional understanding like human(meanwhile stockfish and minimax engines are strong in tactics but weak in positional understanding). Leela will scale better if there is more time for tactical calculation.
Brilliant!
Also very interested to see your comments on Leela's rise.
Leela Zero perfectly found Scott opening by her own up to 5. Be3. 5. ---Ne5 was bad move according to SF . . 3500+ rated Daddy Fish will immediately punish for such inaccuracy!!
http://lczero.org/match_game/27947
@naviary. Could you post screenshots or steps you did? (I dont use mac though)
wow
I downloaded the source code & so far towards compiling I've installed Homebrew, boost, openBLAS, & zlib... But TensorFlow fails to install (Python quits unexpectedly) with this left in the terminal window: 
Attempting to clone the GitHub repo yields at:
... Attempting to run a pre-compiled (cpu-version?) of lczero in terminal (someone sent to me) gives error "Segmentation fault: 11".
And attempting to run the downloaded "client_mac" in terminal says "Permission Denied" (and if I use sudo command I only get "command not found") 
EDIT: After figuring out that I needed an SSH key, now here's what happens when I clone the GitHub repo: 
and: 
Not sure if that step is done cause nothing happened after entering that last line 
Are you doing for network contribution (testing) or play in GUI?
Hmm, what GUI do you use? This might be a good GUI. https://sourceforge.net/projects/scidvspc/files/mac/
Afaik. You just need to download mac version of leela and latest network to play Leela..
To play Leela in GUI you dont need to download client.
You need dedicated graphic card to play Leela. ( there is window version that dont need GPU but it will be weak)
There were bad problems recently. Network shutdown cause unknown bugs introduced into the network. Network is now updated to version 4 and it is working fine again with gaining elo.