Upload README.md with huggingface_hub

README.md

@@ -6,39 +6,6 @@ tags:
 - markdown
 - uv-script
 - generated
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-dataset_info:
-  features:
-  - name: document_id
-    dtype: string
-  - name: page_number
-    dtype: string
-  - name: image
-    dtype: image
-  - name: text
-    dtype: string
-  - name: alto_xml
-    dtype: string
-  - name: has_image
-    dtype: bool
-  - name: has_alto
-    dtype: bool
-  - name: markdown
-    dtype: string
-  - name: inference_info
-    dtype: string
-  - name: fix-size
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 115623205
-    num_examples: 512
-  download_size: 84896742
-  dataset_size: 115623205
 ---
 
 # Document OCR using LightOnOCR-1B-1025
@@ -51,21 +18,21 @@ This dataset contains OCR results from images in [davanstrien/handbooks-lighton-
 - **Model**: [lightonai/LightOnOCR-1B-1025](https://huggingface.co/lightonai/LightOnOCR-1B-1025)
 - **Vocabulary Size**: 151k tokens
 - **Number of Samples**: 512
-- **Processing Time**: 2.5 min
-- **Processing Date**: 2025-10-27 11:57 UTC
+- **Processing Time**: 9.2 min
+- **Processing Date**: 2025-10-27 12:11 UTC
 
 ### Configuration
 
 - **Image Column**: `image`
 - **Output Column**: `markdown`
 - **Dataset Split**: `train`
-- **Batch Size**: 64
+- **Batch Size**: 32
 - **Target Image Size**: 1540px (longest dimension)
 - **Max Model Length**: 8,192 tokens
 - **Max Output Tokens**: 6,500
 - **Temperature**: 0.2
 - **Top P**: 0.9
-- **GPU Memory Utilization**: 97.0%
+- **GPU Memory Utilization**: 80.0%
 
 ## Model Information
 
@@ -120,12 +87,12 @@ uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/lighton-ocr.py \
     <output-dataset> \
     --vocab-size 151k \
     --image-column image \
-    --batch-size 64
+    --batch-size 32
 ```
 
 ## Performance
 
-- **Processing Speed**: ~3.41 images/second
+- **Processing Speed**: ~0.93 images/second
 - **Benchmark Score**: 76.1% overall (across diverse document types)
 - **Optimization**: Native resolution ViT + lightweight decoder