pulpie

Extract main content from HTML. One forward pass, not one token at a time.

EXAMPLE PAGE

▸OR PASTE RAW HTML

If set, this overrides the example dropdown. ⌘/Ctrl+Enter to run.

PULPIE 210M · ENCODER

—

× FASTER · ON THIS PAGE

DRIPPER 0.6B · DECODER

One encoder forward pass over every block.

WEBMAINBENCH · 6,647 PAGES · ROUGE-5 F1

pulpie orange-small210M

0.862

dripper (MinerU-HTML)0.6B

0.864

trafilatura—

0.619

Same quality. ~$7,900 to clean 1B pages, versus ~$159,000. — 13.7 pg/s on L4 vs 0.68 pg/s