TL;DR: The latest Llama.cpp release enables local execution of Cohere2MoE architectures like TINY_AYA, expanding its model support ecosystem.
Summary: The ggerganov team released Llama.cpp build b9631. This update introduces cohere2moe vocabulary support specifically for the TINY_AYA model architecture. It also implements fixes for the CLI regarding preserved token copying and updates release checks to include SYCL.
Why it matters: Developers can now leverage Llama.cpp's efficient local inference engine to run Cohere Mixture-of-Experts models. Update your local binaries to build b9631 to experiment with the TINY_AYA architecture.
Source: github.com