module
ion7-core
Functions
ion7.init
Initialise the llama.cpp + ggml backend and configure logging. Must be called once at process startup, BEFORE loading any model.
ion7.shutdown
Release every llama.cpp / ggml resource. Safe to call multiple times (the second call is a no-op). Pair with `init` at process exit to release VRAM cleanly.
ion7.capabilities
Snapshot of the build-time and runtime capabilities of the linked libllama / bridge. Downstream modules (`ion7-llm`, `ion7-embed`) consult this to adapt behaviour (e.g. enable mmap when supported).
ion7.time_us
Microsecond timestamp from llama.cpp's internal monotonic clock. Useful for precise inner-loop timing without `os.clock`'s platform-dependent resolution.
ion7.numa_init
Configure NUMA placement. Call BEFORE `ion7.init` for the policy to apply to llama.cpp's own thread spawning.