06 Dec 2025
The Nanbeige4-3B model family from the Nanbeige LLM Lab at Boss Zhipin introduces a 3-billion-parameter language model that consistently outperforms much larger open-source models, setting new state-of-the-art averages in mathematical and scientific reasoning. This performance is achieved through a multi-stage training pipeline incorporating advanced data filtering, a fine-grained learning rate scheduler, dual-level preference distillation, and multi-stage reinforcement learning.
There are no more papers matching your filters at the moment.
