Should decoder only models like GPT used for classification?

I found a Llama model to be the best available among the ones I tried for Finnish text classification. So I think there might be a use case depending on the language and domain.