making LLMs sparse at inference time

visit shbcf.ru