LightOnOCR-2 on Pinokio 🔥
Just uploaded: LightOnOCR-2 — Local OCR on Pinokio 🔥
Took the brand-new lightonai/LightOnOCR-2-1B (the current SOTA 1B OCR model on OlmOCR-Bench) and packaged it perfectly for Pinokio.
What you get:
Multiple variants (best overall OCR + bbox detection + soups)
Images + full PDF support (page selector + preview)
Streaming output (text appears as it’s generated)
Runs on NVIDIA / Apple Silicon / CPU (have not tested on Apple or CPU, please reach out if there is an issue)
Model caching + smart memory management
Clean Gradio UI with examples
~9× smaller and faster than the big OCR VLMs, handles tables, math, forms, handwriting, multi-column layouts… all locally.
Discussion (7)
Up to 10 files, 25MB each. Images are optimized; GIFs -> MP4; videos 720p (max 120s).
