veloxquant_mlx.cli.estimate_memory module¶
CLI: analytical per-method KV-cache memory estimates for a model+workload.
- veloxquant_mlx.cli.estimate_memory.main(argv=None)[source]¶
Parse model/workload CLI args and print analytic per-method KV-memory estimates.
Builds a ModelProfile and WorkloadProfile from the given flags (or a –model-config JSON file), estimates every registered method’s KV footprint, optionally overrides with measured benchmark-database records (–benchmark-dir), and prints the smallest-footprint –top methods as text or JSON (–json).