https://github.com/allenai/OLMoE is a very good MOE structure, with 7B parameter and 1B activate parameter. It is very suitable for academic research. Whether the litgpt team considers to support this model?