by Sachin Mehta, Mohammad Hossein Sekhavat, Qingqing Cao, Maxwell Horton, Yanzi Jin, Chenfan Sun, Iman Mirzadeh, Mahyar Najibi, Dmitry Belenko, Peter Zatloukal, and Mohammad Rastegari.
The tech giant said transparency and reproducibility are essential for advancing open research of LLMs. Four of the eight models have been pre-trained using the CoreNet library, and the other four are instruction-tuned models. “For example, with a parameter budget of approximately one billion parameters, OpenELM exhibits a 2.36% improvement in accuracy compared to OLMo while requiring 2× fewer pre-training tokens.”
Technology Technology Latest News, Technology Technology Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Source: mybroadband - 🏆 11. / 67 Read more »
Source: News24 - 🏆 4. / 80 Read more »