Ling-lite

About

Ling-lite is the smaller of inclusionAI's two open Ling mixture-of-experts language models, with 16.8 billion total parameters and 2.75 billion activated per token; the larger Ling-Plus has 290 billion total and 28.8 billion activated. Activating only a fraction of the weights keeps inference cost near that of a much smaller dense model, and the shared MoE structure is designed to scale up or down for different deployment budgets. Weights are on Hugging Face under the MIT license and can be served with vLLM or SGLang.