veloxquant_mlx.cli.auto_config module¶
CLI: pick a hardware-aware KV-cache config for a workload (issue #253).
- veloxquant_mlx.cli.auto_config.main(argv=None)[source]¶
Parse workload/hardware CLI args and print the selected KV-cache config.
Selects a method/bit-width/group-size from a small servable pool (turboquant_rvq, kivi, kvquant, gear) via select_kv_cache_config, using detected or overridden hardware memory pressure, and prints the result as text or JSON (–json).