Datalab Says Indra OCR 1.1 Reads Handwriting and 80+ Languages on CPU
Summary
Datalab releases Indra OCR 1.1, a 1M-parameter model that reads handwriting, OCRs 80+ languages and straightens pages tilted up to 45 degrees, all on CPU. Datalab reports an OmniParseBench headline of 66.7, up from 62.3, and 71.4 with --math, with handwriting tests rising from 8.3% to 25.2%.
Key Points
- Datalab reports a median page of 212 milliseconds on 4 vCPUs, about 16% slower than version 1.0.
- On 32 cores without a GPU, Datalab says 1.1 reads about 44 pages per second, 15 times Tesseract's rate.
- Datalab says character F1 reaches 0.90 or higher on 83 of 95 held-out printed languages.