Skip to content
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

Implement multi-contempt mode #17

Open
skiminki opened this issue Apr 20, 2024 · 0 comments
Open

Implement multi-contempt mode #17

skiminki opened this issue Apr 20, 2024 · 0 comments
Milestone

Comments

@skiminki
Copy link
Owner

skiminki commented Apr 20, 2024

Implement multi-contempt analysis mode. The way this would work is as follows:

For every position:

  1. Do Lichess masters db query
  2. Clear tree when (a) more than one contempt level is specified; OR (b) both side analysis is used at non-zero contempt
  3. Run Lc0 analysis at first specified contempt level
  4. Run Lc0 analysis at second specified contempt level
  5. ...
  6. Produce: (a) PGN combining the contempt levels (if requested); AND (b) one PGN per contempt level (if requested)

This should speed up the analysis, since the NN evaluation cache is shared between Lc0 analyses. The NN evaluation is where the majority of Leela search time goes, even with a strong GPU such as RTX 4090.

Contempt level would be specified as something such as:
--analysis-pass=contempt,search-nodes:[double-check-nodes]

Further, we should get rid of separate white/black engines. Instead, one engine is always used, but the user could specify analysis sides white/black/both.

A warning should be produced if the user has not specified NN cache, or has specified a small NN cache than the maximum nodes per move at any contempt level.

@skiminki skiminki added this to the Release 0.2 milestone Apr 20, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant