Note Primary OCR SFT training data: single-page PDFs with transcription and layout-aware targets.

tencent/HunyuanOCR

Image-Text-to-Text • 1B • Updated Aug 28 • 818k • 842

Note OCR/VL baseline and reference for document parsing, structured extraction, and serving design.

Note Held-out OCR/document benchmark only. Never use as training data.

Note Receipt OCR and structured JSON extraction. Domain-adaptation data after general OCR SFT.

Note Noisy English receipt OCR, bounding boxes, key fields, and line-item labels.

Note Document layout segmentation reference; later auxiliary layout work, not initial LFM SFT.

Note Receipt/invoice audit candidate; verify license, provenance, and overlap with CORD before training.

Note Electronic schematics and PCB layouts; symbols, labels, and circuit topology reference.

Note Industrial wiring and electrical-diagram benchmark; evaluation and graph-extraction reference.

Note Hand-drawn circuit diagrams with component, text, and port annotations.

Note Synthetic electrical/engineering diagram QA and structured visual reasoning.