Researchers run high-performing LLM on the energy needed to power a lightbulb

UC Santa Cruz researchers show that it is possible to eliminate the most computationally expensive element of running large language models, called matrix multiplication, while maintaining performance.

Read more here: External Link