Maybe not the best diversity of strongest chess engines under development

[connor_mcmonigle]

What about Seer, Halogen 8, and Minic 3?

I seem to be unable to run Seer 1.1 and Halogen 8.1 with Sim, even if I messed with Sim.tcl file. I am not sure what's the matter, maybe they are not fully UCI compliant. I was curious about them, as they seem to be original NNUE implementations.

Do you have any logs that I could dig through to maybe find the issue with Seer's UCI compliance? I've been trying to resolve all outstanding issues for the next release. A path to the Weights file must be specified as a UCI option for Seer to run so perhaps that was the issue. Seer 1.1 also has a known bug where it prints a junk bestmove to the console when stop is issued before go. It should support "go movetime <x>" and "go depth <x>", though it doesn't support pondering or "go mate in ..." commands.

[Laskos]

Unfortunate, and entirely unsurprising in regards to the NNUE similarities. Perhaps, some hope though. Even though Komodo and Stockfish are, it appears, trained on the same code base, the differences between their evals, and its usage in training, is enough to have at least some diversity. You don't get quite near the level of intra-engine play. So one can return back to the argument of, "Its unique if its trained on different data, even if the trainer is the same", which was the failed mantra of DeusX, but seemed to work for Leelenstein and Allie.

I was unable to run Halogen 8.1 in Sim, it seems the feeding of fen positions doesn't work properly with Halogen. I was curious about this engine, as if I understood correctly, it uses your trainer and there are very few engines using a different trainer (3?).

[Laskos]

What about Seer, Halogen 8, and Minic 3?

I seem to be unable to run Seer 1.1 and Halogen 8.1 with Sim, even if I messed with Sim.tcl file. I am not sure what's the matter, maybe they are not fully UCI compliant. I was curious about them, as they seem to be original NNUE implementations.

Do you have any logs that I could dig through to maybe find the issue with Seer's UCI compliance? I've been trying to resolve all outstanding issues for the next release. A path to the Weights file must be specified as a UCI option for Seer to run so perhaps that was the issue. Seer 1.1 also has a known bug where it prints a junk bestmove to the console when stop is issued before go. It should support "go movetime <x>" and "go depth <x>", though it doesn't support pondering or "go mate in ..." commands.

Unfortunately I don't have logs with the Sim tool. Sim tcl file has these UCI commands:

"go depth 1" ---> for depth=1 similarity
"go depth 50" and after 100ms "stop" ---> for time=100ms/position similarity.

Sim, Save.bin and Seer 1.1 are in the same directory.
Without "Weights" option, Seer 1.1 runs and freezes while running after the first position.
With "Weights = ./save.bin" I am getting the weird "Invalid option: Weights".
It seems a Sim problem, but no other engine refuses an UCI option fed by Sim. Maybe I will manage to have some logs.