| title: Pulpie | |
| emoji: 🍊 | |
| colorFrom: red | |
| colorTo: yellow | |
| sdk: gradio | |
| sdk_version: 6.17.3 | |
| python_version: '3.13' | |
| app_file: app.py | |
| pinned: false | |
| license: apache-2.0 | |
| short_description: Clean HTML 20x faster — encoder vs decoder, live | |
| models: | |
| - feyninc/pulpie-orange-small | |
| - opendatalab/MinerU-HTML-v1.1-hunyuan0.5B-compact | |
| # Pulpie — encoder vs decoder | |
| Extract the main content from any web page, side by side: [pulpie](https://github.com/feyninc/pulpie) | |
| (`orange-small`, 210M encoder) races Dripper (MinerU-HTML, 0.6B decoder) on the same | |
| page. Pulpie labels every block in a single forward pass; the decoder emits its answer | |
| one token at a time. Same quality (~0.86 ROUGE-5 F1 on WebMainBench), ~20x the speed. | |
| - `pip install pulpie` | |
| - Code: https://github.com/feyninc/pulpie | |
| - Benchmarks: https://github.com/feyninc/pulpie/blob/main/BENCHMARKS.md | |
| - Blog: https://usefeyn.com/blog/pulpie-pareto-optimal-models-for-cleaning-the-web/ | |
| Built by [Feyn](https://usefeyn.com). | |