BriLLM-MetaPred achieves the first engineering validation of the core primitives of AGI theory

BriLLM-MetaPred enables the first core primitive derived from AGI theory to be implemented in the design of a language model. The model is built based on mathematical theory, with a parameter size of 175B, achieving a score of 78.9 on the MMLU benchmark, a 3.5x speedup, and supporting a context length of 8k tokens. It is currently provided as a closed-source API.