Apple SOTA OCR with Core ML and dots.ocr released

AI Impact Summary

Apple has released SOTA OCR with Core ML and dots.ocr, a 3B parameter OCR model from RedNote that surpasses Gemini 2.5 Pro in OmniDocBench. This capability enables on-device OCR processing, appealing to developers seeking to avoid API keys and network dependencies. The process involves converting a PyTorch model to Core ML using tools like CoreMLTools and MLX, highlighting the challenges of adapting models for on-device execution, particularly concerning data types and dynamic control flow. Developers can utilize a combination of CoreML and MLX to run the dots.ocr model on Apple devices, leveraging the Neural Engine for high-performance inference.