paper2sw

Paper2SW

Predict high‑impact neural network weights (“super‑weights”) directly from research papers — no model build required.

Get Started{ .md-button .md-button–primary } CLI{ .md-button } GitHub{ .md-button }

Why Paper2SW?

Feature Highlights

- :rocket: **Fast CLI** — Predict in seconds with a single command - :floppy_disk: **Caching** — Avoid recomputation across runs - :scissors: **Long-context selection** — Keep only the most relevant text - :gear: **Configurable** — Tweak model, device, precision, and more - :memo: **Mermaid + GLightbox** — Rich diagrams and image lightbox - :zap: **Minified docs** — Snappy GitHub Pages site

Quick Demo

```bash title=”Predict top-5 super-weights” paper2sw predict
–paper https://arxiv.org/abs/2411.07191
–out sw.jsonl
–top_k 5


```python title="Via Python API"
from paper2sw import Predictor

predictor = Predictor.from_pretrained(enable_cache=True, selection_keep_ratio=0.5)
preds = predictor.predict("./README.md", top_k=5)
for p in preds:
    print(p)

Learn More

This site is open source. Improve this page.