3 Commits

Author	SHA1	Message	Date
wangjue	7e9a014f4c	Use llama_chat_apply_template for generic model support; add --chat-template flag	2026-05-29 15:17:26 +08:00
wangjue	5180d7b508	feat: support reading image from clipboard when no image_path is given ... If <image_path> is omitted, reads image from clipboard using wl-paste (Wayland) or xclip (X11), then passes the raw data to mtmd_helper_bitmap_init_from_buf for decoding.	2026-05-29 14:59:22 +08:00
wangjue	cdcd1f7202	feat: implement OCR tool using llama.cpp multimodal API ... - Support GLM-OCR model with -m (model) and --mmproj (mmproj) options - Custom prompt support via -p (default: OCR) - Greedy/temperature sampling with --temp flag - JSON output with --json flag - GPU acceleration via -ngl option - Use correct GLM-OCR chat template format for prompt construction	2026-05-29 14:57:37 +08:00