olmOCR 2: Unit Test Rewards for Document OCR

TL;DR

olmOCR 2 is an advanced OCR system that uses reinforcement learning with binary unit test rewards to improve document digitization, especially in complex layouts like math formulas and tables.

Contribution

Introduction of olmOCR 2, a specialized vision language model trained with RLVR using unit test rewards, achieving state-of-the-art OCR performance.

Findings

State-of-the-art performance on olmOCR-Bench

Largest improvements in math formula conversion

Enhanced table parsing and multi-column layout handling

Abstract

We present olmOCR 2, the latest in our family of powerful OCR systems for converting digitized print documents, like PDFs, into clean, naturally ordered plain text. olmOCR 2 is powered by olmOCR-2-7B-1025, a specialized, 7B vision language model (VLM) trained using reinforcement learning with verifiable rewards (RLVR), where our rewards are a diverse set of binary unit tests. To scale unit test creation, we develop a pipeline for generating synthetic documents with diverse and challenging layouts, known ground-truth HTML source code, and extracted test cases. We show that RL training on these test cases results in state-of-the-art performance on olmOCR-Bench, our English-language OCR benchmark, with the largest improvements in math formula conversion, table parsing, and multi-column layouts compared to previous versions. We release our model, data and code under permissive open licenses.