2025Nanbeige LLM Lab at Boss Zhipinopen source
Nanbeige4-3B-Thinking
v2511
Nanbeige4-3B-Thinking is a 3-billion parameter language model that leverages a 23 trillion token training pipeline to achieve reasoning capabilities comparable to larger models. It employs advanced data quality techniques, curriculum scheduling, distillation, and reinforcement learning to enhance performance.
3-billion parameter model
23 trillion token training pipeline
Advanced data quality techniques