Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- ocr
|
| 4 |
+
- document-processing
|
| 5 |
+
- glm-ocr
|
| 6 |
+
- markdown
|
| 7 |
+
- uv-script
|
| 8 |
+
- generated
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# Document OCR using GLM-OCR
|
| 12 |
+
|
| 13 |
+
This dataset contains OCR results from images in [willwhim/wattsocr](https://huggingface.co/datasets/willwhim/wattsocr) using GLM-OCR, a compact 0.9B OCR model achieving SOTA performance.
|
| 14 |
+
|
| 15 |
+
## Processing Details
|
| 16 |
+
|
| 17 |
+
- **Source Dataset**: [willwhim/wattsocr](https://huggingface.co/datasets/willwhim/wattsocr)
|
| 18 |
+
- **Model**: [zai-org/GLM-OCR](https://huggingface.co/zai-org/GLM-OCR)
|
| 19 |
+
- **Task**: text recognition
|
| 20 |
+
- **Number of Samples**: 20
|
| 21 |
+
- **Processing Time**: 3.1 min
|
| 22 |
+
- **Processing Date**: 2026-02-20 01:44 UTC
|
| 23 |
+
|
| 24 |
+
### Configuration
|
| 25 |
+
|
| 26 |
+
- **Image Column**: `image`
|
| 27 |
+
- **Output Column**: `markdown`
|
| 28 |
+
- **Dataset Split**: `train`
|
| 29 |
+
- **Batch Size**: 16
|
| 30 |
+
- **Max Model Length**: 8,192 tokens
|
| 31 |
+
- **Max Output Tokens**: 8,192
|
| 32 |
+
- **Temperature**: 0.01
|
| 33 |
+
- **Top P**: 1e-05
|
| 34 |
+
- **GPU Memory Utilization**: 80.0%
|
| 35 |
+
|
| 36 |
+
## Model Information
|
| 37 |
+
|
| 38 |
+
GLM-OCR is a compact, high-performance OCR model:
|
| 39 |
+
- 0.9B parameters
|
| 40 |
+
- 94.62% on OmniDocBench V1.5
|
| 41 |
+
- CogViT visual encoder + GLM-0.5B language decoder
|
| 42 |
+
- Multi-Token Prediction (MTP) loss for efficiency
|
| 43 |
+
- Multilingual: zh, en, fr, es, ru, de, ja, ko
|
| 44 |
+
- MIT licensed
|
| 45 |
+
|
| 46 |
+
## Dataset Structure
|
| 47 |
+
|
| 48 |
+
The dataset contains all original columns plus:
|
| 49 |
+
- `markdown`: The extracted text in markdown format
|
| 50 |
+
- `inference_info`: JSON list tracking all OCR models applied to this dataset
|
| 51 |
+
|
| 52 |
+
## Reproduction
|
| 53 |
+
|
| 54 |
+
```bash
|
| 55 |
+
uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/glm-ocr-v2.py \
|
| 56 |
+
willwhim/wattsocr \
|
| 57 |
+
<output-dataset> \
|
| 58 |
+
--image-column image \
|
| 59 |
+
--batch-size 16 \
|
| 60 |
+
--task ocr
|
| 61 |
+
```
|
| 62 |
+
|
| 63 |
+
Generated with [UV Scripts](https://huggingface.co/uv-scripts) (glm-ocr-v2.py)
|