Pulpie: Open-Source Model Cleans Web Content at 20x Lower Cost

Pulpie, a family of Pareto optimal encoder models, strips boilerplate from raw HTML, matching state-of-the-art extraction quality while being 20x cheaper. Cleaning 1 billion webpages costs $7,900 with Pulpie versus $159,000 with Dripper. The models are open source on Hugging Face and aim to improve data quality for LLM context.