Products/Web Content Extraction/Pulpie – Models for Cleaning the Web

Pulpie – Models for Cleaning the Web

Pareto-Optimal Models for Cleaning the Web

Web Content Extraction
Visit Pulpie – Models for Cleaning the Web →

Our Take

Pulpie is an open-source family of Pareto-optimal models for extracting main content from HTML pages — stripping navigation, ads, sidebars, and footers to isolate core content. Built by Feyn, Pulpie's smallest model (210M parameters) matches the leading extractor Dripper (600M) on ROUGE-5 F1 (0.862 vs 0.864) at one-third the size and one-twentieth the cost. Processing speed: 13.7 pages/sec on NVIDIA L4 vs Dripper's 0.68 pages/sec. Open-source models available on Hugging Face (feyninc/pulpie-orange-small). Targets AI/ML teams building web scraping pipelines, pre-training data pipelines, and context management systems.

Extracts main content from HTML pages

Key Features
Pareto-optimal models for web content extraction, Encoder architecture that labels every HTML block as content or boilerplate in a single forward pass, Matches SOTA extraction quality at one twentieth the cost, Open source and available on Hugging Face, Fast processing speed (e.g., pulpie-orange-small processes 13.7 pages/sec on an NVIDIA L4 GPU)
Problem It Solves
Addresses the bottleneck of extraction in language models by cleaning web content to improve data quality for both pre-training and inference
Target Customer
Organizations or individuals working with language models and web data extraction
Use Cases
Pre-training language models with cleaner data, Improving context management in language models, High-quality web extraction at scale
Pricing Details
Cost of cleaning 1 billion pages is $7,900 with Pulpie compared to $159,000 with Dripper
Differentiator
Significantly cheaper and faster than competitors while maintaining high quality
Why Now
The need for high-quality, efficient web content extraction is growing due to the increasing reliance on language models for various applications

Key Facts

Category
Web Content Extraction
Discovered via
hacker-news

Links

WebsiteSource: hacker-news

Browse by category

Ai & machine learning (151)Devtools (56)Developer tools (47)Other (32)Productivity & collaboration (29)Security (23)Infrastructure (15)Finance & payments (12)Health (11)Fintech (10)Education (10)Data & analytics (10)Marketing (9)Finance (8)Productivity (7)Utilities (6)Saas (6)Media & entertainment (6)Communication (6)E-commerce & retail (5)

Want products like this in your inbox every morning?

Five products. Every morning. Written by someone who actually cares whether they're good or not. Free forever, unsubscribe whenever.

Pulpie – Models for Cleaning the Web — SLAYREPORT