A 27M parameter model beating LLMs on reasoning tasks6 points·by SteadySurfdom·vor 8 Monaten·1 comments