SoMark open-sources the 1.3B document parsing model OXR, ranking first on three benchmarks
SoMark has open-sourced the 1.3B document parsing model OXR, which ranks first on olmOCR-Bench, OmniDocBench, and DocParsingBench, achieves 3.05x the throughput of GLM-OCR on a single RTX PRO 6000, and is the fastest open-source result in the speed comparison. OXR introduces RSI during post-training: a VLM evaluates its own outputs, rollout divergence flags uncertainty, and a separate clustering model attributes it, based on which the next round of targeted data is generated. Rohan Paul calls OXR the first model to use RSI in document parsing, with the GitHub link in the comments.
Why read this
AI score 53
The 1.3B OXR ranks first on three document parsing benchmarks and reaches 3.05x the single-card throughput of GLM-OCR, offering a comparable open-source document parsing solution.
Topics
View original postBeyond OCR. Meet OXR. Only 1.3B. Open source. #1 on olmOCR-Bench, OmniDocBench & DocParsingBench🏆 #1 in parsing speed among open-source models⚡ RSI-powered. A first in document parsing. Turn documents into Agent-Friendly data. If OXR helps, give us a star on GitHub! 🌟 https://t.co/kQlsq8oIdN
View original
Beyond OCR. Meet OXR. Only 1.3B. Open source. #1 on olmOCR-Bench, OmniDocBench & DocParsingBench🏆 #1 in parsing speed among open-source models⚡ RSI-powered. A first in document parsing. Turn documents into Agent-Friendly data. If OXR helps, give us a star on GitHub! 🌟 https://t.co/kQlsq8oIdN
Body
OXR makes parsing failures part of the training process.
During post-training (OXR is the first document parser to use RSI here), a VLM evaluates its outputs, rollout disagreements flag uncertainty, and a separate clustering model groups the causes;
those findings then guide targeted data generation for the next round, and the 1.3B model that comes out of it sits first on olmOCR-Bench, OmniDocBench and DocParsingBench while running 3.05× GLM-OCR's throughput on a single RTX PRO 6000, the fastest open-source result in the speed comparison.
Worth studying (Github link in comment)