davanstrien

Run self-contained UV-script recipes over Hugging Face datasets on Hugging Face Jobs (`hf jobs uv run <url>`): OCR & document extraction, audio transcription, object detection & segmentation, NER/entity extraction, text & image classification, embeddings & dataset maps, synthetic data, and batch LLM/VLM inference. Each recipe reads a Hub dataset and writes a new one, so recipes chain into pipelines. Use when the user wants to batch-process a dataset at scale, pre-label data for human review, run a model over many images/audio/documents/rows, or build a data pipeline on Hugging Face — locally with `uv run` or on a managed GPU with `hf jobs uv run`. Recipes live at huggingface.co/uv-scripts (one dataset repo per task family); discover them and read a recipe's header before running.

Internal, dev-only: smoke-test and bump the pinned vLLM recipes in this repo. Use periodically (~quarterly) or after a vLLM release to confirm each pinned vLLM script still runs on Jobs and to propose pin / image-tag bumps to a newer known-good. Tests on l4x1 via `hf jobs uv run`; reports a diff of proposed bumps; never auto-merges.

Audit the shippable `uv-recipes` skill against the live Hugging Face Hub and `hf` CLI to catch drift. Use periodically — monthly, after an `hf`/`huggingface_hub`/vLLM release, or when adding a recipe family — to confirm the run pattern (INPUT_DATASET OUTPUT_DATASET, Hub-in/Hub-out), the `hf jobs uv run` flags, the hardware flavors, the default image, the discovery commands, and the task-family list in `skills/uv-recipes/SKILL.md` still match reality. Reports what is stale and proposes minimal edits; does not change the shippable skill without review.

Add a model to the GLAM extraction benchmark and run it, then refresh the leaderboard. Handles three model shapes — schema-native extractor Spaces (NuExtract), general VLMs via the HF Inference Providers router, and any OpenAI-compatible endpoint (including a model you served on Jobs). Use when adding or scoring a new model on this benchmark.

Serve any Hugging Face Hub model as a temporary OpenAI-compatible endpoint using HF Jobs + vLLM, so you can evaluate/benchmark it, iterate prompts, or hit it from an agent — without a permanent deployment. Use when a model is NOT on Inference Providers (e.g. a research model like datalab-to/lift) and you just need an endpoint for an afternoon. Always cancel the job when done so billing stops.

Work with the Hugging Face `datasets` Python library the right way — load and stream datasets, inspect schema/size without downloading, cast Features, and build or publish datasets to the Hub. Use when loading a dataset (load_dataset), streaming a large one (IterableDataset), previewing rows, or pushing local files/folders to the Hub (push_to_hub). Pairs with the `hf ds` CLI extension for safe one-shot operations.

Deploy and use Label Studio on Hugging Face Spaces. Use when the user wants to stand up an LS instance on Spaces, push tasks from a Hub dataset, add ML predictions as prelabels, export reviewed annotations back to a Hub dataset, or needs a human-in-the-loop labelling workflow for ML training or evaluation data (image bounding boxes, text NER, classification, etc.).