Skip to content

Rewrite README + add LICENSE#13

Merged
chonknick merged 1 commit into
mainfrom
readme-rewrite
Jun 30, 2026
Merged

Rewrite README + add LICENSE#13
chonknick merged 1 commit into
mainfrom
readme-rewrite

Conversation

@chonknick

Copy link
Copy Markdown
Contributor

Overhauls the README (modeled on chonkie's, clean & technical voice) and adds the missing Apache-2.0 LICENSE.

  • Hero: banner + badges (PyPI · Python · License · Downloads · Blog · stars) + nav links
  • Sections: feature bullets, install (pip/uv + markdown extra), usage (Extractor + Pipeline), model zoo, how-it-works, benchmarks, acknowledgements, citation, Chonkie/Feyn footer
  • Numbers grounded in the published blog — corrects stale figures (Small 0.862 / Base 0.863 / Large 0.873; 13.7 p/s on L4; ~$7,900/1B pages, 20x faster/cheaper than Dripper)
  • No fabricated badges/links (no docs site/Discord/codecov yet)
  • Adds LICENSE (Apache-2.0 was declared in pyproject but the file was missing)
  • Reconciles BENCHMARKS.md cost/throughput tables to the blog's L4 numbers

- New README modeled on chonkie's: centered hero (banner + badges + nav),
  feature bullets, install, usage (Extractor + Pipeline), model zoo, how-it-works,
  benchmarks, acknowledgements, citation, Feyn/Chonkie footer
- Numbers grounded in the published blog + BENCHMARKS.md (corrects stale figures:
  Small 0.862, Base 0.863, Large 0.873; L4 13.7 p/s; ~$7,900/1B pages, 20x)
- Badges: PyPI, Python, License, Downloads, Blog, stars (no fabricated docs/Discord)
- Add Apache-2.0 LICENSE file (was declared in pyproject but missing)
- Reconcile BENCHMARKS.md cost/throughput tables to the blog's L4 numbers
Copilot AI review requested due to automatic review settings June 30, 2026 04:43
@chonknick
chonknick merged commit b4a5295 into main Jun 30, 2026
8 checks passed
@chonknick
chonknick deleted the readme-rewrite branch June 30, 2026 04:44

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Updates the project’s top-level documentation to reflect current positioning and benchmark numbers, and adds the missing Apache-2.0 license file to match the declared package license.

Changes:

  • Overhaul README.md with a more structured hero section, installation/usage guidance, model zoo, and benchmark summary.
  • Add an Apache-2.0 LICENSE file (previously declared but missing).
  • Reconcile BENCHMARKS.md throughput/cost tables and notes to align with the published blog figures.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File Description
README.md Rewritten README structure (badges/sections/usage/models/benchmarks) and updated headline numbers.
LICENSE Adds Apache License 2.0 text to the repo.
BENCHMARKS.md Updates notes and adds L4/A100 throughput + L4 cost projection tables aligned to the blog.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Comment thread README.md
Comment on lines +28 to +32
**⚡ Fast** — an encoder labels every block in one forward pass (13.7 pages/sec on an L4) </br>
**🎯 Accurate** — matches SOTA quality: 0.862–0.873 ROUGE-5 F1 on WebMainBench </br>
**🪶 Small** — the recommended model is 210M params, fits on any GPU </br>
**💸 Cheap** — clean 1 billion pages for ~$7,900 vs ~$159,000 for the leading decoder </br>
**📦 Simple** — `pip install pulpie`, then `Extractor().extract(html)` </br>
Comment thread README.md
**🪶 Small** — the recommended model is 210M params, fits on any GPU </br>
**💸 Cheap** — clean 1 billion pages for ~$7,900 vs ~$159,000 for the leading decoder </br>
**📦 Simple** — `pip install pulpie`, then `Extractor().extract(html)` </br>
**🔌 Batched** — overlapped CPU+GPU pipeline scales across multiple GPUs </br>
Comment thread BENCHMARKS.md
Comment on lines +27 to 28
- **Dripper** ([MinerU-HTML v1.1](https://huggingface.co/opendatalab/MinerU-HTML-v1.1-hunyuan0.5B-compact)) uses vLLM with guided regex decoding for batched inference.
- **Dripper** ([MinerU-HTML v1.1](https://huggingface.co/opendatalab/MinerU-HTML-v1.1-hunyuan0.5B-compact)) uses vLLM with guided regex decoding for batched inference.
Comment thread BENCHMARKS.md
- **pg/s** (this table): pages/second during the full WebMainBench quality run on a
single A100 80GB, with Pulpie using a naive sequential loop (no batching). The
batched throughput numbers in the section below are the ones to use for
speed/cost — batching roughly 6x's Pulpie Small on A100 (4.0 → 25.7).
Comment thread LICENSE
Comment on lines +1 to +2

Apache License
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants