#56 · Primary category: Prompt Engineering Tools

promptsource

machine-learning natural-language-processing nlp

Toolkit for creating, sharing and using natural language prompts.

Project last updated:10/23/23

GitHub Stars

3.0K

Forks

375

Contributors

65

License

Apache-2.0

Why we included this project

People doing zero- or few-shot work with large language models usually end up hand-crafting prompts, and PromptSource was the toolkit the BigScience T0 model was trained with. It treats a prompt as a template written in Jinja that maps any example from a Hugging Face dataset into a natural-language input and target output. The bundled P3 collection holds roughly two thousand English prompts across more than 170 datasets, so you can pull a ready-made template for classification, entailment, or generation tasks, or write your own in the web interface and check the rendered text against sample examples right away. The Python API slots into evaluation or fine-tuning scripts, and the standalone prompt files make it simple to version, share, and inspect exactly what a model is being asked.

Articles for this project

No articles for this project yet.

To suggest a topic or contribute an article, contact us.

Related projects in this category