Open-source algorithms enable AI developers worldwide to combine/ accelerate power of different AI models; LLMs' performance up to 2.8X faster

https://www.eurekalert.org/news-releases/1091432

"speculative decoding: small/ fast model, possessing limited knowledge, makes first guess while answering user’s query, and larger/ powerful but slower model reviews/ corrects, maintaining 100% accuracy... each LLM translates output from internal token language into shared format, then another algorithm prompts reliance on collaborative work that use tokens having same meaning across models... division of labor, especially benefiting edge devices: phones, drones, autonomous cars"