README.md

---
title: README
emoji: 📚
colorFrom: red
colorTo: indigo
sdk: static
pinned: false
---

# UV Scripts

**Ready-to-run ML tools powered by UV - zero setup, maximum power**

Run state-of-the-art ML workflows with a single command. From OCR to classification, all scripts work instantly with `uv run`.

## What are UV scripts?

UV scripts are self-contained Python scripts that use [inline metadata](https://docs.astral.sh/uv/guides/scripts/) to specify dependencies. Just `uv run script.py` and everything installs automatically.

Perfect for:

- 🚀 **GPU workflows** on [HF Jobs](https://huggingface.co/docs/huggingface_hub/guides/jobs)
- 💻 **Local processing** on your machine
- 🔄 **Reproducible pipelines** that work anywhere

## 🚀 Quick Example

```bash
# Extract text from images with state-of-the-art OCR (no local GPU needed!)
hf jobs uv run --flavor l4x1 \
  https://huggingface.co/datasets/uv-scripts/ocr/raw/main/nanonets-ocr.py \
  your-images your-extracted-text
```

## 📚 Browse Scripts

| Script Collection                                                               | Description                                               | GPU Required |
| ------------------------------------------------------------------------------- | --------------------------------------------------------- | ------------ |
| [ocr](https://huggingface.co/datasets/uv-scripts/ocr)                           | Extract text from images with VLMs (LaTeX, tables, forms) | ✅           |
| [classification](https://huggingface.co/datasets/uv-scripts/classification)     | Text classification with guaranteed valid outputs         | ✅           |
| [dataset-creation](https://huggingface.co/datasets/uv-scripts/dataset-creation) | Create datasets from PDFs and files                       | ❌           |
| [vllm](https://huggingface.co/datasets/uv-scripts/vllm)                         | High-performance inference with vLLM                      | ✅           |
| [synthetic-data](https://huggingface.co/datasets/uv-scripts/synthetic-data)     | Generate high-quality synthetic data with CoT reasoning   | ✅           |
| [deduplication](https://huggingface.co/datasets/uv-scripts/deduplication)       | Remove duplicates using semantic similarity               | ❌           |
| [openai-oss](https://huggingface.co/datasets/uv-scripts/openai-oss)             | Generate responses with visible reasoning traces          | ✅           |

## 🎯 Why UV Scripts?

### Zero Setup

No virtual environments, no dependency conflicts, no installation steps. UV handles everything automatically when you run the script.

### GPU Optimized

Seamlessly run on local GPUs or scale to cloud with [HF Jobs](https://huggingface.co/docs/huggingface_hub/guides/jobs). Same script, different compute.

## 🌟 Featured Scripts

### OCR Any Document Dataset

Extract text from images with state-of-the-art accuracy:

```bash
# Handles LaTeX, tables, forms, handwriting
hf jobs uv run --flavor l4x1 \
  https://huggingface.co/datasets/uv-scripts/ocr/raw/main/nanonets-ocr.py \
  your-images extracted-text
```

### Deduplicate Datasets (CPU-Friendly!)

Remove duplicates using semantic similarity - no GPU needed:

```bash
# Fast semantic deduplication on CPU
uv run https://huggingface.co/datasets/uv-scripts/deduplication/raw/main/semantic-dedupe.py \
  your-dataset text your-dataset-clean \
  --method duplicates --threshold 0.9
```

### Generate Synthetic Training Data

Create high-quality synthetic data with chain-of-thought reasoning:

```bash
# Generate synthetic math problems with reasoning
hf jobs uv run --flavor l4x1 \
  https://huggingface.co/datasets/uv-scripts/synthetic-data/raw/main/cot-self-instruct.py \
  --seed-dataset math-examples --output-dataset synthetic-math \
  --task-type reasoning --num-samples 1000
```

## 🚀 Getting Started with HF Jobs

Run any UV script on GPU infrastructure:

```bash
hf jobs uv run --flavor l4x1 \
  https://huggingface.co/datasets/uv-scripts/[collection]/raw/main/[script].py \
  [args]
```

Choose your GPU flavor:
- `l4x1` - Good balance for most tasks
- `a10g-large` - More memory for larger models
- `a100-large` - Maximum performance

## 📖 Learn More

- [UV Documentation](https://docs.astral.sh/uv/)
- [HF Jobs Guide](https://huggingface.co/docs/huggingface_hub/guides/jobs)
- [Script Examples](https://github.com/astral-sh/uv/tree/main/scripts)

---

_UV Scripts is a community project showcasing the power of [UV](https://github.com/astral-sh/uv) for ML workflows._