SkyIsNotGreen/Scion-35B-A3B · Hugging Face - Ternary MoE

SkyIsNotGreen has released Scion-35B-A3B, a Ternary MoE model, on Hugging Face. Users can access a GitHub fork of llama.cpp at https://github.com/sky-is-green/prism-ml-llama.cpp, which includes prebuilt binaries for immediate use. A draft version of the model is also available at https://huggingface.co/SkyIsNotGreen/Scion-35B-A3B-mtp-drafter.

This is the first Ternary MoE model released on Hugging Face, offering a new approach compared to existing Mixture of Experts architectures.