Comment by Bolwin

Does anyone know if vlms suffer more from quantization? One thing I've noticed is practically every improvement in llms is already half there in quantization.