llama.cpp

ggml-rpc: chunk send()/recv() to avoid EINVAL for very large tensors over RPC (macOS & others)

#15188

Merged