Ling 3.0 Reasoning Models Support Merged into llama.cpp
Support for the Ling 3.0 model family, including the Ling-3.0-tiny (8B) and Ling-3.0-flash (124B) reasoning models, has been officially merged into the llama.cpp repository. This integration allows users to run these native hybrid reasoning models locally. This update enables the local LLM community to run advanced, resource-efficient hybrid reasoning models on consumer hardware. By supporting these models, llama.cpp expands the options for developers seeking low-latency, cost-effective local AI reasoning. The Ling 3.0 models utilize a hybrid architecture, where Ling-3.0-tiny features 1.3B activated parameters (out of 8B total) and Ling-3.0-flash activates 5.1B parameters (out of 124B total). The integration was completed via pull request #26608 on the llama.cpp GitHub repository.
## BACKGROUND
llama.cpp is a highly popular open-source C/C++ inference engine designed to run large language models efficiently on local hardware, serving as the backbone for tools like Ollama. Ling 3.0, developed by Ant Bailing, is a family of native hybrid reasoning models designed for low-latency, cost-effective execution and real-time interaction.