Adds host-memory figures for lean mode (Windows) and for a memory-limited Linux container, and states that no real small-memory computer was tested. The model file is unchanged.
Document the automatic memory selection and its measured thresholds
Quick start and at-a-glance table first; fidelity table unchanged; staged-memory result measured with this file; CPU versus GPU decoder comparison; comparison with a build lacking the denoiser calibration; text-prompt memory budget. Measurement records are linked from the idol.cpp repository.
2 days ago
Document full fidelity benchmark and verified Docker release
