llama.cppProduct update
llama.cpp · v0.7.0
llama.cpp v0.7.0 adds --moe-cache-mib, a GPU cache for MoE experts kept in host memory that speeds decoding by up to about 2.2x when experts are offloaded. It also brings LiliCorr speculative decoding, MTP for GLM5-Next and new models including K2 Horizon, EmbeddingGemma-2 and MiniCPM-V 4.7. The server's default port changes from 8080 to 9931, though Docker images still use 8080 inside the container, and the web UI gains a full models manager.
ggml updated to v0.27.0: the backend scheduler gains user callbacks for split weight copying and compute-buffer reallocation, and the RPC protocol is bumped to v8 for tensor-parallel RPC servers.
Saved . A shortened excerpt from GitHub Releases; the original publisher retains authorship.