There is no single “best net” for Leela but there a few worth recommending for various purposes. The most important consideration in choosing a net is picking the right size for your hardware and time controls (TC). In general, if you have a weak GPU (or no GPU at all) and you want to only spend milliseconds per move, then you want a smaller net that evaluates positions more quickly, i.e. higher NPS (nodes per second). On the other hand, if you have an RTX card (or several) and you want to run an analysis on a single move for hours at a time, then the quality of the evaluation is more important than the speed and a larger (but slower) net is probably going to work best.
Size versus Recommended Purpose
- <10b: Recommended for sparring vs humans
- 10b: Recommended for running on CPU
- 20b: Recommended for running on non-RTX cards or TC on the order of seconds (with RTX)
- 24b: Recommended for TC > 1 minute per move with an RTX card
- >24b: Recommended for long analysis or when speed isn’t a major factor
Network List
For nets of the same size, the first net listed is likely the strongest.
| Size | Name | Source for Download | Notes |
|---|---|---|---|
| 5b x 48f | Good Gyal 5 | GitHub: dkappe Bad Gyal | Other sizes also here |
| 9b x 112f | ID11258-112x9-se | GitHub: dkappe Distilled Networks | Other sizes also here |
| 10b x 128f | Latest T59 | lczero.org run 2 networks 591226. | A previous test run |
| 10b x 128f | Little Demon 2 | data.lczero.org repository (LD2) | JH nets also here |
| 16b x 192f | J20-460 | GitHub: jhorthos Leela Training | Trained on T40 data |
| 20b x 256f | Leelenstein 14.2 | Patreon: jjosh | Early Access |
| 20b x 256f | SV-20b-t40-1541 | Sergio-V repository | Trained on T40 data |
| 20b x 256f | Leelenstein 14.0 | 14.0 Post | No account required |
| 20b x 256f | T40: #42850 | lczero.org run 1 networks | Last T40 net |
| 24b x 320f | Latest T60 | lczero.org run 1 networks | Current main run |
| 24b x 320f | J13B.2-136 | GitHub: jhorthos Leela Training | “Terminator 2” Net |
| 30b x 384f | 384x30-t60-3010, 384x30-t60-3044 | Sergio-V repository | Trained on T60 data |
| 30b x 384f | SV-30b-t40-1705 | Sergio-V repository | Trained on T40 data |
Note: The Sergio-V nets are also available on data.lczero.org in some cases.
“This is all too complicated. Just tell me what net to use!”
If you don’t care about squeezing out the very best performance for a particular situation and want a general-purpose net, pick a medium size 20b net, which should do reasonably well (if not optimal) under most common conditions.
As in the table above, the top-recommended 20b net is 256x20-t40-1541.pb.gz from Sergio Vieri’s repository.
If this page hasn’t been updated recently, check the Discord channels for a recommendation. Be sure to specify your hardware and use case so the helpful regulars know what to recommend.