Update README.md
Browse files
README.md
CHANGED
|
@@ -31,9 +31,13 @@ tags:
|
|
| 31 |
|
| 32 |
**Best OCR model (recommended).** LightOnOCR-2-1B is our flagship OCR model, refined with RLVR training for maximum accuracy. We recommend this variant for most OCR tasks.
|
| 33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 34 |
## Highlights
|
| 35 |
|
| 36 |
-
* ⚡ **Speed:** 5× faster than dots.ocr, 2× faster than PaddleOCR-VL-0.9B, 1.73× faster than DeepSeekOCR
|
| 37 |
* 💸 **Efficiency:** Processes 5.71 pages/s on a single H100 (~493k pages/day) for **<$0.01 per 1,000 pages**
|
| 38 |
* 🧠 **End-to-End:** Fully differentiable, no external OCR pipeline
|
| 39 |
* 🧾 **Versatile:** Handles tables, receipts, forms, multi-column layouts, and math notation
|
|
@@ -41,7 +45,7 @@ tags:
|
|
| 41 |
|
| 42 |
---
|
| 43 |
|
| 44 |
-
📄 **[Paper](https://huggingface.co/papers/lightonocr-2)** | 📝 **[Blog Post](https://huggingface.co/blog/lightonai/lightonocr-2)** | 🚀 **[Demo](https://huggingface.co/spaces/lightonai/LightOnOCR-2-Demo)** | 📊 **[Dataset](https://huggingface.co/datasets/lightonai/LightOnOCR-mix-0126)**
|
| 45 |
|
| 46 |
---
|
| 47 |
|
|
@@ -61,7 +65,7 @@ tags:
|
|
| 61 |
## Benchmarks
|
| 62 |
|
| 63 |
<div align="center">
|
| 64 |
-
<img src="
|
| 65 |
</div>
|
| 66 |
|
| 67 |
*See the [paper](https://huggingface.co/papers/lightonocr-2) for full benchmark details and methodology.*
|
|
|
|
| 31 |
|
| 32 |
**Best OCR model (recommended).** LightOnOCR-2-1B is our flagship OCR model, refined with RLVR training for maximum accuracy. We recommend this variant for most OCR tasks.
|
| 33 |
|
| 34 |
+
## About LightOnOCR-2
|
| 35 |
+
|
| 36 |
+
LightOnOCR-2 is an efficient end-to-end 1B-parameter vision-language model for converting documents (PDFs, scans, images) into clean, naturally ordered text without relying on brittle pipelines. This second version is trained on a larger and higher-quality corpus with stronger French, arXiv, and scan coverage, improved LaTeX handling, and cleaner normalization. LightOnOCR-2 achieves state-of-the-art performance on OlmOCR-Bench while being ~9× smaller and significantly faster than competing approaches.
|
| 37 |
+
|
| 38 |
## Highlights
|
| 39 |
|
| 40 |
+
* ⚡ **Speed:** 3.3× faster than Chandra OCR, 1.7× faster than OlmOCR, 5× faster than dots.ocr, 2× faster than PaddleOCR-VL-0.9B, 1.73× faster than DeepSeekOCR
|
| 41 |
* 💸 **Efficiency:** Processes 5.71 pages/s on a single H100 (~493k pages/day) for **<$0.01 per 1,000 pages**
|
| 42 |
* 🧠 **End-to-End:** Fully differentiable, no external OCR pipeline
|
| 43 |
* 🧾 **Versatile:** Handles tables, receipts, forms, multi-column layouts, and math notation
|
|
|
|
| 45 |
|
| 46 |
---
|
| 47 |
|
| 48 |
+
📄 **[Paper](https://huggingface.co/papers/lightonocr-2)** | 📝 **[Blog Post](https://huggingface.co/blog/lightonai/lightonocr-2)** | 🚀 **[Demo](https://huggingface.co/spaces/lightonai/LightOnOCR-2-1B-Demo)** | 📊 **[Dataset](https://huggingface.co/datasets/lightonai/LightOnOCR-mix-0126)**
|
| 49 |
|
| 50 |
---
|
| 51 |
|
|
|
|
| 65 |
## Benchmarks
|
| 66 |
|
| 67 |
<div align="center">
|
| 68 |
+
<img src="benchmark.png" alt="OlmOCR-Bench Results" width="900"/>
|
| 69 |
</div>
|
| 70 |
|
| 71 |
*See the [paper](https://huggingface.co/papers/lightonocr-2) for full benchmark details and methodology.*
|