Oriel — AI & technology worth noticing, every day.

xiaomi-research/Xiaomi-OCR-0

AI model · https://huggingface.co/xiaomi-research/Xiaomi-OCR-0

First seen by Oriel on .

AI summary

Xiaomi-OCR-0 is a unified 0.8B vision-language model for document parsing, OCR-centric visual question answering, and key information extraction.

  • Starting from Qwen3.5-0.8B-Base, it uses approximately 170M OCR-centric samples, Q-Mask text anchoring, continued pretraining, and Mix-RL with verifiable rewards.
  • It combines expert-consensus annotation, render-guided correction, multi-factor mining, and coverage- or failure-driven synthesis for text, tables, and formulas.
  • The model scores 95.24 on Real5-OmniDocBench, 96.83 on OmniDocBench v1.6, and 83.2 averaged across five OCR-centric VQA benchmarks.
  • Scaling from 0.8B to 4B improves understanding by about 4.9 points, while parsing improves by about 0.15 points; Mix-RL peaks higher than faster-converging MOPD.
  • Transformers usage supports whole-page parsing with Markdown, OTSL tables, LaTeX formulas, plain text regions, and JSON-schema-based key information extraction.

#ocr #document parsing #vision-language models #multimodal ai

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Models: 311, latest observation ; +154 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Models: 24,621, latest observation ; +8,612 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Modelsxiaomi-research/Xiaomi-OCR-0https://huggingface.co/xiaomi-research/Xiaomi-OCR-0 Read discussion