OCR AI performance controls
Local OCR AI can now be tuned for the balance of memory use, speed, and recognition quality that fits your Mac.
- Added smaller Q3_K_S and Q2_K model variants alongside the balanced Q4_K_M default.
- Added an Advanced options section for llama-server context size, GPU layers, CPU threads, batch sizing, and Flash Attention.
- Added memory mapping and RAM locking controls for local runtime tuning.
- Kept the app and command line OCR paths aligned with the selected model variant.