README.md

metadata

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

UV Scripts

Run a data or ML task over a Hugging Face dataset in one command — for humans and agents.

Each recipe is a single self-contained UV script: dependencies are declared inline, so you run it straight from a URL — no clone, no virtualenv, no pip install. Run it locally with uv run, or hand it to Hugging Face Jobs for a managed GPU. Most recipes read a Hub dataset and write a new one, so they chain into pipelines.

Quickstart

See every recipe — locally, no GPU or token:

uv run https://huggingface.co/datasets/uv-scripts/jobs-utils/raw/main/list-recipes.py

Run one on a GPU — the flagship, OCR an image dataset to text:

hf jobs uv run --flavor l4x1 --secrets HF_TOKEN \
  https://huggingface.co/datasets/uv-scripts/ocr/raw/main/glm-ocr.py \
  davanstrien/ufo-ColPali your-username/ufo-ocr --max-samples 10

One command → a new dataset with a markdown column. Pay-per-second, no hardware of your own.

Drive it with your coding agent

Recipes take their arguments in the same input output order and run from a URL, so an agent (Claude Code, Cursor, …) can pick one and run it with no setup. The simplest start — paste this so it discovers what's available:

List the uv-scripts recipes and tell me which fit my task:
uv run https://huggingface.co/datasets/uv-scripts/jobs-utils/raw/main/list-recipes.py
For context on how these work, read the org page https://huggingface.co/uv-scripts
and the GitHub repo https://github.com/davanstrien/uv-scripts-for-ai.

More prompts — run a job, build a dataset →

Try it now — runs a real OCR job and hands back a dataset:

Using uv-scripts, OCR a sample dataset on Hugging Face Jobs:
  hf jobs uv run --flavor l4x1 --secrets HF_TOKEN \
    https://huggingface.co/datasets/uv-scripts/ocr/raw/main/glm-ocr.py \
    davanstrien/ufo-ColPali $MY_HF_USERNAME/ufo-ocr-test --max-samples 10
Then open the output dataset and show me the `markdown` column.

Put it to work — when you need data for a task:

I need a dataset for <my task>. uv-scripts has recipes that create, OCR,
transcribe, classify, deduplicate, and embed datasets on Hugging Face. List them:
  uv run https://huggingface.co/datasets/uv-scripts/jobs-utils/raw/main/list-recipes.py
Pick the one that fits, read its script header for the arguments, and run it with:
  hf jobs uv run --flavor l4x1 --secrets HF_TOKEN <script-url> INPUT_DATASET OUTPUT_DATASET
Each recipe reads a Hub dataset and writes a new one, so chain them as needed.
Background: https://huggingface.co/uv-scripts and https://github.com/davanstrien/uv-scripts-for-ai

The cookbook also ships a ready-made agent skill for discovering and running recipes — see the GitHub repo, and Hugging Face's own hf CLI skill for agents. (We'll refine these prompts over time.)

Browse

Every recipe is in the list below — OCR, detection & segmentation, audio transcription, NER & classification, embeddings & atlas maps, batch LLM/VLM inference, synthetic data, and dataset creation. Or browse on GitHub · run hf jobs hardware for GPU flavors & pricing.