llama.cppProduct update
llama.cpp · v0.5.0
llama.cpp v0.5.0 speeds up CUDA conv2d with implicit GEMM and brings MoE and SSM_CONV fusion optimizations to Metal. The server can listen on several addresses at once, with --host taking comma-separated TCP addresses and UNIX sockets. On the model side it adds HRM-Text (DFM Mimir 1B) and conversion support for MiMo-V2.6 and HunyuanOCR, and ggml moves to v0.25.0.
This release focuses on backend performance and correctness, broader model coverage, and more robust server/router operation.
Saved . A shortened excerpt from GitHub Releases; the original publisher retains authorship.