Ed Lee

Explore every move twice before normal training self-play search

14

This is a bit of departure from AZ's Dirichlet noise, but it's still "zero" with an additional kind of "noise" with the hopes of better directing self-play's search by getting...

Adjust root magic numbers to help search find moves for training

1

As an alternative to #698 for those who don't like the inelegance of just forcing two visits per root move, there's existing numbers that affect root's search: epsilon, alpha, fpu,...

Like https://github.com/ggerganov/llama.cpp/pull/6554 to support http://localhost:8080/?q=prompt like https://chat.openai.com/?q=prompt starting a chat but without the index.html.hpp nor fix to api_url not applicable here. (cherry picked from commit ggerganov/llama.cpp@400d5d722d7edf7de0cf24a18c42b183c65047d2)