Total noob Leela question

[fohristiwhirl]

Well I can't really explain (nor reproduce, really) this Ne4 move.

I took the fast TC because I am really interested in what happens at very low node counts. I heard stories that a one-node Leela is rated over 2000 on some server, (playing against humans, no doubt), and I want to see that before I believe it. Does it support "go nodes"?

Yes, you can use "go nodes 1".

However, the net you've got (assuming you've stuck with the one provided) is rather weak (and is provided because it's small). If you want to play 1-node at strength, I'd suggest the SV-3010 net (which was used at TCEC) which you can find here:

https://www.comp.nus.edu.sg/~sergio-v/t60/384x30/

I've seen it beat a > 2650 GM at 1 node (admittedly at blitz time controls) though he may have been giving it too much respect. Using CPU only, I (with a mid-range CPU) get only 1 node per second for this net though; as opposed to 5000 or so on GPU.

[yurikvelo]

I heard stories that a one-node Leela is rated over 2000 on some server, (playing against humans, no doubt), and I want to see that before I believe it.

probably not with bundled 10x128 network

Does it support "go nodes"?

it is full UCI-compliant

NN evaluation must have learned that it is in general better to have a Pawn than not have it, and that having passers is even better?!?

first - you run 10x128, very basic network.
all NN tend to play worth in late midgame/endgame, even big networks with good GPU
with extremely low node count and basic network you cannot expect 3000+ elo play in late-mid/EG
switching to AB engine at move #35...40 provide big ELO boost

[Ovyron]

I want to suggest trying the network that I found to be best for my CPU, though this is with 23 nps (ten times slower than yours), and it hasn't been updated since December.

http://data.lczero.org/files/networks-c ... -821.pb.gz

You download it, unzip it, and rename the pb file to weights and replace the weights file Leela comes with. Or just set it to this net in engine settings, or set it to discover and just delete the original weights file so it uses this one.