llama.cpp Release b11624 Improves RPC Tensor Transfer Logs
llama.cpp release b11624 introduces improved RPC tensor transfer logging, formatting byte counts in KiB, MiB, and GiB while adding detailed tensor names, operations, types, and shapes to trace logs. It also fixes incorrect UNKNOWN command name mappings for various RPC functions and reorganizes log helper functions into a common ggml-rpc module. These updates enhance debugging visibility and transparency for developers running distributed large language model inference across remote procedure call (RPC) backends. Better diagnostic tooling helps engineers monitor resource consumption and troubleshoot performance bottlenecks more effectively. The update adds rpc_log_format_size() and rpc_log_format_tensor() to standardize trace logging across client, server, and transport layers. Additionally, log helper functions were cleanly relocated to ggml-rpc/log.{h,cpp} under a unified rpc_log_ prefix.
## BACKGROUND
llama.cpp is a popular open-source library that performs local inference for large language models using the GGML tensor computation engine. It supports remote procedure call (RPC) functionality to distribute computation tasks across multiple machines or devices.