Natural Language Annotation

[cpeters]

Hello!

I did experiment with locally run llm too and experienced the opposite what you're stating:

Not bad as far as the commentary goes.

, as it is totally off (like in this game/commentary). Simply bollocks. The bloody obvious is getting much coverage, but the big elephants (or "Helffandt" (A. Dürer)) won't get addressed. There are plenty of mistakes here.

Maybe it's a prompt thingy/technical issue though (because the cloud-services produce precise results).


greetings

[AdminX]

Hello!

I did experiment with locally run llm too and experienced the opposite what you're stating:

Not bad as far as the commentary goes.

, as it is totally off (like in this game/commentary). Simply bollocks. The bloody obvious is getting much coverage, but the big elephants (or "Helffandt" (A. Dürer)) won't get addressed. There are plenty of mistakes here.

Maybe it's a prompt thingy/technical issue though (because the cloud-services produce precise results).


greetings

In general I agree about local models, still so far Gemma gave me the best result locally if I compare it to Muse. My best results came from Cloud models.

[cpeters]

From this list I did try the two first ones (you can fiddle with the prompt..):

https://github.com/shauryasri14/Chess-Move-Analyzer
https://github.com/rishavray/chess-analyzer

https://github.com/Nomka0/Chess-Analyzer
https://github.com/Atharva-Jayappa/chess-analyzer
https://github.com/Bosslibra/ShashGuru


https://github.com/M2002HR/LLM_Based_Ch ... t_Analyzer
https://github.com/skysyaz/Chess
https://github.com/n1ops/chess_analyzer
https://github.com/ggolani/chess-analyzer
https://github.com/nolansvt/chess-analyzer
https://github.com/alcow5/Chess-Analyzer
https://github.com/Niklas7400/lichess-chess-analyzer

with these models:

Code: Select all

a-llama3.2:3b
b-dolphin-mistral:7b
c-dolphin-llama3:8b
d-llama3.1:8b 

(because of the mouse-RAM on my macbook) and got 'mixed' results. I'll have a look at it again in half a year or so.

greetings