>>107081926
I lost motivation to keep updating the leaderboard because I don't think my ranking criteria were as good as I wanted. It was sufficient to show that big models = good and small models = usually worse, but sometimes there were outliers that scored badly but weren't actually that bad, or models that scored well but performed poorly in real use. That really didn't inspire confidence. I was investigating LLM as a judge but didn't get too far.
But anyway, I may add scores for the new Gemma models, I'm a bit curious about where Gemma 3n would fall.

>>107082013
>>107082038
I'm not that guy lol. I don't even have a Xwitter account.

>>107082176
I actually just took this license from the original Mikupad script that was released on pastebin and never cared to change it, because I don't care about attributions. I mean, "/lmg/ anon" means nothing, lol.
I don't know if using a license like GPL would be a good idea because it could pose problems for people who want to include Mikupad in MIT projects.
So for now I guess I'll keep it like that.

>>107080625
Heh I wish it was that simple, life just hasn't been gentle with me this year. So I'm catching up on the low hanging fruits now.

Btw, if anyone wants to suggest or report anything about Mikupad, feel free to do it, I promise to not take long to give a reply. I saw some anons saying logits were broken but I couldn't reproduce that, so I'm not sure if I already fixed that or if I just didn't find the problem.