Sync from GitHub via hub-sync

README.md

@@ -9,44 +9,59 @@ pinned: false
 
 # UV Scripts
 
-**Run ML workflows on GPU with a single command - no setup required**
+**Run a data or ML task over a Hugging Face dataset in one command — for humans and agents.**
 
-Self-contained Python scripts powered by [UV](https://docs.astral.sh/uv/guides/scripts/) and [HF Jobs](https://huggingface.co/docs/huggingface_hub/guides/jobs). Pick a task, copy the command, and you're running on cloud GPUs in seconds.
+Each recipe is a single self-contained [UV script](https://docs.astral.sh/uv/guides/scripts/): dependencies are declared inline, so you run it straight from a URL with no clone, no virtualenv, no `pip install`. Run it locally with `uv run` where you have the hardware, or hand it to [Hugging Face Jobs](https://huggingface.co/docs/hub/jobs) for a managed GPU. Most recipes read a Hub dataset and write a new one, so they chain into pipelines.
 
-## Get Started
+## See every recipe (no GPU, no token)
+
+The lowest-friction start — just [uv](https://docs.astral.sh/uv/getting-started/installation/), runs locally in seconds:
+
+```bash
+uv run https://huggingface.co/datasets/uv-scripts/jobs-utils/raw/main/list-recipes.py
+```
+
+It prints a runnable URL for every recipe in the org. Run any of them the same way: `uv run <url>` locally, or `hf jobs uv run <url>` on a GPU.
+
+## Run one for real
+
+The flagship is **[OCR](https://huggingface.co/datasets/uv-scripts/ocr)** — turn an image dataset into text & structured data, 30+ models. On a managed GPU (no hardware of your own; pay-per-second):
 
 ```bash
-hf jobs uv run --flavor l4x1 \
-  https://huggingface.co/datasets/uv-scripts/ocr/raw/main/nanonets-ocr.py \
-  your-images your-extracted-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, no setup. Works locally too with just `uv run`.
+One command → a new dataset with a `markdown` column.
+
+## For your coding agent
 
-## Browse Scripts
+Recipes are built to be agent-driven — same `input output` arg order, runnable from a URL, self-describing headers. Two prompts to paste into Claude Code, Cursor, or similar:
 
-**Document Processing**
-- [ocr](https://huggingface.co/datasets/uv-scripts/ocr) - Extract text from images with VLMs (LaTeX, tables, forms, handwriting)
-- [dataset-creation](https://huggingface.co/datasets/uv-scripts/dataset-creation) - Create HF datasets from PDFs and local files (CPU only)
+**Try it now** — runs a real OCR job and hands back a dataset:
 
-**Computer Vision**
-- [sam3](https://huggingface.co/datasets/uv-scripts/sam3) - Zero-shot object detection with natural language prompts using SAM3
+```
+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.
+```
 
-**Text Generation & Classification**
-- [vllm](https://huggingface.co/datasets/uv-scripts/vllm) - High-performance GPU inference with vLLM (classification, VLM tasks)
-- [classification](https://huggingface.co/datasets/uv-scripts/classification) - Text classification with structured outputs and guaranteed valid labels
-- [openai-oss](https://huggingface.co/datasets/uv-scripts/openai-oss) - Generate responses using OpenAI's open-source reasoning models
-- [synthetic-data](https://huggingface.co/datasets/uv-scripts/synthetic-data) - Generate synthetic training data with chain-of-thought reasoning
+**Put it to work** — when you need data for a task:
 
-**Model Training**
-- [unsloth/jobs](https://huggingface.co/datasets/unsloth/jobs) - Fine-tune LLMs and VLMs with Unsloth (LFM2.5, Qwen3-VL, Gemma3, continued pretraining)
+```
+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.
+```
 
-**Analysis & Visualization**
-- [dataset-stats](https://huggingface.co/datasets/uv-scripts/dataset-stats) - Analyze dataset statistics with streaming and Polars (CPU only)
-- [build-atlas](https://huggingface.co/datasets/uv-scripts/build-atlas) - Generate interactive embedding visualizations with Apple's Atlas
+Prefer a packaged setup? The cookbook ships an **agent skill** for discovering and running recipes — see the [GitHub repo](https://github.com/davanstrien/uv-scripts-for-ai). Hugging Face also ships an [`hf` CLI skill for agents](https://huggingface.co/docs/hub/agents-cli). _(We'll refine these prompts over time.)_
 
-## Learn More
+## More
 
-- [HF Jobs Documentation](https://huggingface.co/docs/hub/jobs) - run `hf jobs hardware` to see available GPUs and pricing
-- [Available GPU hardware](https://huggingface.co/docs/hub/en/spaces-gpus#hardware-specs)
-- [UV Script Documentation](https://docs.astral.sh/uv/guides/scripts/)
+Every other recipe is in the list below — detection & segmentation, audio transcription, NER & classification, embeddings & atlas maps, batch LLM/VLM inference, synthetic data, and dataset creation. Or browse on **[GitHub](https://github.com/davanstrien/uv-scripts-for-ai)** · run `hf jobs hardware` for GPU flavors & pricing.