llama.cpp b11027 released with RPC device filtering fix
llama.cpp build b11027 has been released, introducing a patch that skips accelerator (ACCEL) devices in the Remote Procedure Call (RPC) server backend. This update improves stability during distributed LLM inference across networked devices, preventing crashes caused by exposing unsupported acceleration backends to the scheduler. Implemented via pull request #29020, the fix ensures that `ggml-rpc-server` filters out ACCEL devices, resolving issues when combining hardware like Apple Silicon Macs and NVIDIA PCs.
## BACKGROUND
llama.cpp is an open-source C/C++ framework optimized for running Large Language Models efficiently on consumer hardware. Its RPC backend enables distributed inference by sharing compute workload and memory across multiple computers connected over a local network.