Gemma models could be run domestically on a laptop computer, and surpass equally sized Llama two models on several evaluated benchmarks. Generalized models can have equivalent effectiveness for language translation to specialized modest models The causal masked focus is fair within the enc