Upload README.md with huggingface_hub

README.md

@@ -5,9 +5,9 @@ tags: [uv-script, ocr, vision-language-model, document-processing]
 
 # OCR UV Scripts
 
-> Part of [uv-scripts](https://huggingface.co/uv-scripts) - ready-to-run ML tools powered by UV
+> Part of [uv-scripts](https://huggingface.co/uv-scripts) - ready-to-run ML tools powered by UV and HuggingFace Jobs.
 
-Ready-to-run OCR scripts that work with `uv run` - no setup required!
+Ready-to-run OCR scripts that work with `uv run` and HuggingFace Jobs - no setup required!
 
 ## 🚀 Quick Start with HuggingFace Jobs
 
@@ -31,6 +31,45 @@ That's it! The script will:
 
 ## 📋 Available Scripts
 
+### PaddleOCR-VL (`paddleocr-vl.py`) 🎯 Smallest model with task-specific modes!
+
+Ultra-compact OCR using [PaddlePaddle/PaddleOCR-VL](https://huggingface.co/PaddlePaddle/PaddleOCR-VL) with only 0.9B parameters:
+
+- 🎯 **Smallest model** - Only 0.9B parameters (even smaller than LightOnOCR!)
+- 📝 **OCR mode** - General text extraction to markdown
+- 📊 **Table mode** - HTML table recognition and extraction
+- 📐 **Formula mode** - LaTeX mathematical notation
+- 📈 **Chart mode** - Structured chart and diagram analysis
+- 🌍 **Multilingual** - Support for multiple languages
+- ⚡ **Fast initialization** - Tiny model size for quick startup
+- 🔧 **ERNIE-4.5 based** - Different architecture from Qwen models
+
+**Task Modes:**
+
+- `ocr`: General text extraction (default)
+- `table`: Table extraction to HTML
+- `formula`: Mathematical formula to LaTeX
+- `chart`: Chart and diagram analysis
+
+**Quick start:**
+
+```bash
+# Basic OCR mode
+hf jobs uv run --flavor l4x1 \
+    -s HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/paddleocr-vl.py \
+    your-input-dataset your-output-dataset \
+    --max-samples 100
+
+# Table extraction
+hf jobs uv run --flavor l4x1 \
+    -s HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/paddleocr-vl.py \
+    documents tables-extracted \
+    --task-mode table \
+    --batch-size 32
+```
+
 ### LightOnOCR (`lighton-ocr.py`) ⚡ Good one to test first since it's small and fast!
 
 Fast and compact OCR using [lightonai/LightOnOCR-1B-1025](https://huggingface.co/lightonai/LightOnOCR-1B-1025):
@@ -44,11 +83,13 @@ Fast and compact OCR using [lightonai/LightOnOCR-1B-1025](https://huggingface.co
 - 🚀 **Production-ready**: 76.1% benchmark score, used in production
 
 **Vocabulary sizes:**
+
 - `151k`: Full vocabulary, all languages (default)
 - `32k`: European languages, ~12% faster decoding
 - `16k`: European languages, ~12% faster decoding
 
 **Quick start:**
+
 ```bash
 # Test on 100 samples with English text (32k vocab is fastest for European languages)
 hf jobs uv run --flavor l4x1 \
@@ -69,7 +110,39 @@ hf jobs uv run --flavor a100-large \
     --temperature 0.0
 ```
 
-### DeepSeek-OCR (`deepseek-ocr-vllm.py`) ⭐ NEW
+### LightOnOCR-2 (`lighton-ocr2.py`) ⚡ Fastest OCR model!
+
+Next-generation fast OCR using [lightonai/LightOnOCR-2-1B](https://huggingface.co/lightonai/LightOnOCR-2-1B) with RLVR training:
+
+- ⚡ **7× faster than v1**: 42.8 pages/sec on H100 (vs 5.71 for v1)
+- 🎯 **Higher accuracy**: 83.2% on OlmOCR-Bench (+7.1% vs v1)
+- 🧠 **RLVR trained**: Eliminates repetition loops and formatting errors
+- 📚 **Better dataset**: 2.5× larger training data with cleaner annotations
+- 🌍 **Multilingual**: Optimized for European languages
+- 📐 **LaTeX formulas**: Mathematical notation support
+- 📊 **Table extraction**: Markdown table format
+- 💪 **Production-ready**: Outperforms models 9× larger
+
+**Quick start:**
+
+```bash
+# Test on 100 samples
+hf jobs uv run --flavor a100-large \
+    -s HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/lighton-ocr2.py \
+    your-input-dataset your-output-dataset \
+    --batch-size 32 \
+    --max-samples 100
+
+# Full production run
+hf jobs uv run --flavor a100-large \
+    -s HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/lighton-ocr2.py \
+    your-input-dataset your-output-dataset \
+    --batch-size 32
+```
+
+### DeepSeek-OCR (`deepseek-ocr-vllm.py`)
 
 Advanced document OCR using [deepseek-ai/DeepSeek-OCR](https://huggingface.co/deepseek-ai/DeepSeek-OCR) with visual-text compression:
 
@@ -84,6 +157,7 @@ Advanced document OCR using [deepseek-ai/DeepSeek-OCR](https://huggingface.co/de
 - ⚡ **Fast batch processing** - vLLM acceleration
 
 **Resolution Modes:**
+
 - `tiny` (512×512): Fast, 64 vision tokens
 - `small` (640×640): Balanced, 100 vision tokens
 - `base` (1024×1024): High quality, 256 vision tokens
@@ -91,6 +165,7 @@ Advanced document OCR using [deepseek-ai/DeepSeek-OCR](https://huggingface.co/de
 - `gundam` (dynamic): Adaptive multi-tile (default)
 
 **Prompt Modes:**
+
 - `document`: Convert to markdown with grounding (default)
 - `image`: OCR any image with grounding
 - `free`: Fast OCR without layout
@@ -176,7 +251,6 @@ High-quality document OCR using [allenai/olmOCR-2-7B-1025-FP8](https://huggingfa
 - 🧩 **YAML metadata** - Structured front matter (language, rotation, content type)
 - 🚀 **Based on Qwen2.5-VL-7B** - Fine-tuned with reinforcement learning
 
-
 ## 🆕 New Features
 
 ### Multi-Model Comparison Support
@@ -209,6 +283,7 @@ uv run nanonets-ocr.py dataset output --max-samples 100 --shuffle --seed 42
 ### Automatic Dataset Cards
 
 Every OCR run now generates comprehensive dataset documentation including:
+
 - Model configuration and parameters
 - Processing statistics
 - Column descriptions
@@ -221,6 +296,29 @@ Every OCR run now generates comprehensive dataset documentation including:
 No GPU? No problem! Run on HF infrastructure:
 
 ```bash
+# PaddleOCR-VL - Smallest model (0.9B) with task modes
+hf jobs uv run --flavor l4x1 \
+    --secrets HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/paddleocr-vl.py \
+    your-input-dataset your-output-dataset \
+    --task-mode ocr \
+    --max-samples 100
+
+# PaddleOCR-VL - Extract tables from documents
+hf jobs uv run --flavor l4x1 \
+    --secrets HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/paddleocr-vl.py \
+    documents tables-dataset \
+    --task-mode table
+
+# PaddleOCR-VL - Formula recognition
+hf jobs uv run --flavor l4x1 \
+    --secrets HF_TOKEN \
+    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/paddleocr-vl.py \
+    scientific-papers formulas-extracted \
+    --task-mode formula \
+    --batch-size 32
+
 # DeepSeek-OCR - Real-world example (National Library of Scotland handbooks)
 hf jobs uv run --flavor a100-large \
     -s HF_TOKEN \
@@ -337,6 +435,11 @@ uv run nanonets-ocr.py input-dataset output-dataset
 uv run https://huggingface.co/datasets/uv-scripts/ocr/raw/main/nanonets-ocr.py \
     input-dataset output-dataset
 
+# PaddleOCR-VL for task-specific OCR (smallest model!)
+uv run paddleocr-vl.py documents extracted --task-mode ocr
+uv run paddleocr-vl.py papers tables --task-mode table  # Extract tables
+uv run paddleocr-vl.py textbooks formulas --task-mode formula  # LaTeX formulas
+
 # RolmOCR for fast text extraction
 uv run rolm-ocr.py documents extracted-text
 uv run rolm-ocr.py images texts --shuffle --max-samples 100  # Random sample
@@ -354,25 +457,33 @@ Any HuggingFace dataset containing images - documents, forms, receipts, books, h
 
 ### Common Options (All Scripts)
 
-| Option                     | Default | Description                   |
-| -------------------------- | ------- | ----------------------------- |
-| `--image-column`           | `image` | Column containing images      |
-| `--batch-size`             | `32`/`16`* | Images processed together     |
-| `--max-model-len`          | `8192`/`16384`** | Max context length     |
-| `--max-tokens`             | `4096`/`8192`** | Max output tokens      |
-| `--gpu-memory-utilization` | `0.8`   | GPU memory usage (0.0-1.0)    |
-| `--split`                  | `train` | Dataset split to process      |
-| `--max-samples`            | None    | Limit samples (for testing)   |
-| `--private`                | False   | Make output dataset private   |
-| `--shuffle`                | False   | Shuffle dataset before processing |
-| `--seed`                   | `42`    | Random seed for shuffling     |
-
-*RolmOCR and DoTS use batch size 16
-**RolmOCR uses 16384/8192
+| Option                     | Default            | Description                       |
+| -------------------------- | ------------------ | --------------------------------- |
+| `--image-column`           | `image`            | Column containing images          |
+| `--batch-size`             | `32`/`16`\*        | Images processed together         |
+| `--max-model-len`          | `8192`/`16384`\*\* | Max context length                |
+| `--max-tokens`             | `4096`/`8192`\*\*  | Max output tokens                 |
+| `--gpu-memory-utilization` | `0.8`              | GPU memory usage (0.0-1.0)        |
+| `--split`                  | `train`            | Dataset split to process          |
+| `--max-samples`            | None               | Limit samples (for testing)       |
+| `--private`                | False              | Make output dataset private       |
+| `--shuffle`                | False              | Shuffle dataset before processing |
+| `--seed`                   | `42`               | Random seed for shuffling         |
+
+\*RolmOCR and DoTS use batch size 16
+\*\*RolmOCR uses 16384/8192
 
 ### Script-Specific Options
 
+**PaddleOCR-VL**:
+
+- `--task-mode`: Task type - `ocr` (default), `table`, `formula`, or `chart`
+- `--no-smart-resize`: Disable adaptive resizing (use original image size)
+- `--output-column`: Override default column name (default: `paddleocr_[task_mode]`)
+- Ultra-compact 0.9B model - fastest initialization!
+
 **DeepSeek-OCR**:
+
 - `--resolution-mode`: Quality level - `tiny`, `small`, `base`, `large`, or `gundam` (default)
 - `--prompt-mode`: Task type - `document` (default), `image`, `free`, `figure`, or `describe`
 - `--prompt`: Custom OCR prompt (overrides prompt-mode)
@@ -380,10 +491,12 @@ Any HuggingFace dataset containing images - documents, forms, receipts, books, h
 - ⚠️ **Important for HF Jobs**: Add `-e UV_TORCH_BACKEND=auto` for proper PyTorch installation
 
 **RolmOCR**:
+
 - Output column is auto-generated from model name (e.g., `rolmocr_text`)
 - Use `--output-column` to override the default name
 
 **DoTS.ocr**:
+
 - `--prompt-mode`: Choose `ocr` (default), `layout-all`, or `layout-only`
 - `--custom-prompt`: Override with custom prompt text
 - `--output-column`: Output column name (default: `markdown`)