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