KataGo
KataGo copied to clipboard
lightvector
GTP engine and self-play learning in Go
Experimental branch with an option (`useEvalCache=true`) that caches evals of nodes greater than a configurable number of visits (`evalCacheMinVisits`) and remembers them permanently for the duration of that running instance...
Hi, Does Katago can use new intel ultra CPU's NPU? Or, is there any advantage of NPU?
This is the phase where KataGo made the worst mistake by failing to search a mere 30~40 moves. Incidentally, FineArt also failed to come to the correct conclusion, so this...
Dear lightvector: I was just doing a selfplay between weight b28s7332 and b18s9996 with komi 3.5 and found this position  I set the engine only analyze N8 ...
         
see attached file: [OpenCL_error_KataGo_engine_died.txt](https://github.com/user-attachments/files/17134917/OpenCL_error_KataGo_engine_died.txt) Katrain 1.15.0 with included KataGo executable installed and updated with pipx Ubuntu 24.04 up to date Late 2018 Intel Macmini I have been using Katrain on...
Hello @lightvector, this is probably a wrong place but I'd like to share my work related to KataGo: https://github.com/maksimKorzh/kata-model-js/ It's a draft of querying bare KataGo network (quite an old...
I noticed a minor bug in play.py. At some point the arguments to the constructor for GameState were changed to take ruleset as well as board size, but a couple...
Using the cuda backend for distributed training, it always errors out when you set the maxSimultaneousGames to higher than 4000. Nvidia H100s and H200s can easily handle more than 4000...
Has 28b completely surpassed 18b in the same amount of time now? Is there an engine acceleration package like TensorRT that supports 28b?
