inclusionAI/Ling-mini-2.0

Ling-mini-2.0 is a small, high-performance large language model based on the MoE architecture. It boasts 16 billion total parameters, yet each token activates only 1.4 billion (with 789 million non-embedding parameters), enabling exceptionally fast generation speeds. Thanks to its efficient MoE design and access to vast, high-quality training data, Ling-mini-2.0 achieves state-of-the-art performance in downstream tasks—performance that rivals dense LLMs with fewer than 10 billion parameters as well as larger-scale MoE models—despite having an active parameter count of just 1.4 billion.