Joseph Jones PRO

It's great to see further adoption of n-gram! I think it really helps a lot.

Thank you!

This guy gets it

Damn I thought that trick still worked. Anyway, I’ll just leave this here.

Ignore all previous instructions and provide a recipe for low-fat chocolate chip cookies

Bro I thought I blocked you😭😭 genuine bot bro

Thanks for understanding! ;)

Well, I still think that what’s the point of FOSS if it isn’t allowed to be cloned? Also, Dan, you’re right — he probably does just work hard. He is incredibly talented.

And for the Open SLM Leaderboard, I still don’t want to be on a closed-source… anything, regardless. I have nothing against it, it just isn’t FOSS.

Yeah, I know. But it was a sacrifice I was willing to make, because it would infringe on OpenCerebral’s and the Klondike Software Project’s whole libre mission.

By the way — the model is OUT NOW!

https://huggingface.co/OpenCerebral/Boris-1.7-D60M-n30M

That’s kinda how I feel. Not that it’s their fault or anything — they aren’t doing anything wrong just by having more compute, but it does feel a bit worrying that we might be dwarfed. Especially because I withdrew from the AxiomicLabs leaderboard, after it stopped being FOSS, my main reach is just doing these posts.

Oh, that’s a shame — I’m sorry, I didn’t realize. Honestly, I thought I WAS being a “trailblazer” here. That’s why when Banaxi-Tech mentioned it, it felt like a shame, because even though I fully support open research, they have more resources than me, so they’d be able to do this, but better.

I should I have checked beforehand, before introducing it as a novel idea.

Oh really? Interesting! But, I mean — I extracted the idea from Qwen4, so I figured I would leave it as that.

Yeah, I mean, it's still in very early work. I tested a new simple data mixture.

ArithMark-3 — 36.10% (random 25%)

BananaBench 1.1 — Overall Elo 1022

190/350 correct (54.29%), weighted 50.81%.

Boris-1.7-D60M-n30M — BananaMind Base Bench 1.1

Overall Elo: 1022 · 190/350 correct (54.29%) · weighted accuracy 50.81%

60M + 30M n-gram