according to"MilletCompany" official Weibo, today XiaomiLarge Language Model MiLM The large model has been officially approved for filing, and the relevant model will reportedly be gradually applied to Xiaomi cars, mobile phones, smart homes and other products, and will later "be open to more users for experience."
Xiaomi's MiLM large model first appeared on the C-Eval and CMMLU large model evaluation lists in August last year.At that time, this model ranked 10th in the C-Eval overall list and ranked 1st in the same parameter level.
According to the information given on the GitHub project page, MiLM-6B is a large-scale pre-trained language model developed by Xiaomi with a parameter size of 6.4 billion.
▲ Image source: MiLM-6B's GitHub page
According to the information provided by C-Eval, the MiLM-6B model achieved a high accuracy rate in the scores of specific subjects, including metrology, physics, chemistry, biology and other projects among all 20 STEM (science, technology, engineering and mathematics education) subjects.
It is reported that the MiLM-6B model has currently shown relatively good accuracy in most liberal arts subjects, but there is still room for improvement in subjects involving "abstract thinking" such as "law, mathematics, programming, probability theory, and discrete mathematics."
Among the 10 "social science" test subjects, the model achieved relatively ideal accuracy in 8 subjects except "education and geography". Among the 11 subjects of humanities, MiLM-6B had a good accuracy performance based on "history and law".