German AI consortium releases Soofi S, an open 30B model

12 pointsposted 10 hours ago
by yogthos
(the-decoder.com)

3 Comments

porridgeraisin

7 hours ago

https://x.com/eliebakouch/status/2077428860639973471

> gpqa diamond alone makes up 20% of their "capability index" and is responsible for ~70% of the gap between their model and nemotron nano (10 point difference!)

> so what about contamination? i looked at the mixture and they train on a dataset "AIML-TUDA/QA-base" (for 10 epochs lol) that literally is a very very light rephrasing of the gpqa diamond test set

> tl;dr: they trained on 10 epochs of a "very light rephrasing" of EVERY gpqa diamond EVAL data :)