argilla
Argilla is a collaboration tool for AI engineers and domain experts to build high-quality datasets
What is argilla?
Argilla is a collaboration tool for AI engineers and domain experts to build high-quality datasets
Overview of argilla
What it is
Argilla is a collaboration tool for AI engineers and domain experts to build high-quality datasets. It is an open-source, community-driven project licensed under Apache-2.0. The original authors have moved on to new projects; the codebase is mature and stable, and the project is committed to bug fixes and patches but will not add new features going forward. The project is looking for dedicated contributors to take ownership of future development.
Key features
- Programmatic approach to build workflows for continuous evaluation and model improvement.
- Supports collecting human feedback for traditional NLP (text classification, NER), LLMs (RAG, preference tuning), and multimodal models (text to image).
- Provides tools to manage data and models, allowing users to own both their data and models.
- Enables interaction with data through filters, AI feedback suggestions, and semantic search for efficient labeling.
- Integrates with Hugging Face Spaces for easy deployment.
Use cases
- Creating open-source datasets and models, such as the cleaned UltraFeedback dataset used to fine-tune Notus and Notux models, and the distilabel Intel Orca DPO dataset used to fine-tune the improved OpenHermes model.
- AI for good: the Red Cross used Argilla to classify and redirect requests from refugees of the Ukrainian crisis.
- Customer support: Loris.ai used unsupervised and few-shot contrastive learning to validate and gain labeled samples for multi-label classifiers.
- Research studies: Prolific integrated with Argilla to distribute data collection projects among annotators for high-quality data collection.
Getting started
Install the SDK with pip:consolepip install argillaDeploy Argilla Server via Hugging Face Spaces. Then import the client:pythonimport argilla as rgclient = rg.Argilla(api_url=“https://[your-owner-name]-[your_space_name].hf.space”, api_key=“owner.apikey”)Create a dataset with settings and log records. For full details, see the repository README.
License and pricing
License: Apache-2.0. Pricing type: open_source. No pricing information is stated beyond the open-source license.
Based on the project's public README and repository metadata.