llama.cpp Release b10875 Deprecates Legacy Memory Management CLI Arguments
llama.cpp release b10875 officially deprecates the `--mmap`, `--mlock`, and `--dio` command-line flags. This change streamlines CLI flags following earlier refactoring of model loading options. Users and developers relying on legacy scripts or automated workflows using `--mmap`, `--mlock`, or `--dio` will need to update their command-line options. Transitioning to consolidated memory loading flags simplifies configuration and reduces maintenance overhead across multi-platform builds. The deprecation was merged via PR #28334 contributed by Aaron Teo. The release continues to support a wide range of binary distributions across Linux, Windows, macOS, Android, and iOS environments.
## BACKGROUND
llama.cpp is a popular open-source C/C++ inference framework designed for running Large Language Models locally across diverse hardware platforms. Techniques like memory mapping (`mmap`), memory locking (`mlock`), and direct I/O (`dio`) govern how model weights are accessed and paged into physical memory from storage. In earlier releases, these individual flags were consolidated into a single load mode parameter to simplify memory configurations.