Xiaomi’s Large Language Model (LLM) called MiLM has officially passed the registration process for major models, according to a recent announcement on the company’s page. Weibo. With this, Xiaomi declares that MiLM is ready to be integrated into its products, including smartphones, smart home devices and even Xiaomi cars. The announcement also hinted at the possibility of expanding MiLM’s features to a broader user base in the future.

MiLM was first made publicly available on the C-Eval and CMMLU testbeds in August 2023, where it achieved impressive results.

The model ranked 1st in the parameter category in the C-Eval ranking and ranked 10th in the overall ranking. The information on the GitHub page of the project shows that the MiLM-6B model in question has 6.4 billion parameters.

C-Eval’s subject distribution highlights the strength of MiLM-6B in STEM (Science, Technology, Engineering and Mathematics) fields. Model; It achieved high accuracy across all 20 STEM subjects, including metrology, physics, chemistry and biology.

Although MiLM-6B shows promising performance in most humanities courses, fields that require “abstract thinking” such as law, mathematics, programming and probability theory seem to be areas that need improvement.

Social science is another field with mixed results. The MiLM-6B achieved high accuracy in eight out of ten social science subjects, except education and geography. Finally, when it comes to humanities disciplines, the model works well for history and law, while its accuracy for other subjects has yet to be determined.

Since MiLM-6B has overcome a major hurdle, it is now ready to be integrated into various Xiaomi products. Although its performance varies in different areas, it promises to improve user experience in different areas.