datalab-to/indra-ocr — new model trending #30 on Hugging Face
Datalab released Indra OCR, a 1M-parameter model converting documents at 49 pages per second on CPU.
Datalab released Indra OCR, a roughly 1-million-parameter model trending on Hugging Face that converts PDFs and images to markdown, text, or JSON. It reports about 49 pages per second and roughly 230 ms per page on four x86 vCPUs, using an 18 MB ONNX Runtime package without PyTorch. On OmniParseBench it scores 41.3 on text, ahead of PaddleOCR PP-OCR at 35.8 and Tesseract at 19.3, with a 62.1 headline score versus 58.4 for PP-StructureV3. The code is Apache 2.0 and the weights use a modified OpenRAIL-M license.
- About 1 million parameters and 18 MB of ONNX models.
- Claims 49 pages per second and about 230 ms per page on four vCPUs.
- OmniParseBench text score 41.3 and headline score 62.1.
- Outputs markdown, text, or JSON with layout boxes and reading order.
- Code is Apache 2.0; weights use a modified OpenRAIL-M license.
- Vendors
- Datalab
- Organizations
- DatalabHugging Face
From the original article
12 min read at huggingface.coimage-to-text · 1,053 downloads · 161 likes · license openrail
By datalab-to · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.