Home /Benchmarks, evals and calibration /jev-spam-eval

jev-spam-eval

Zero-shot spam filtering with Noul questions against TF-IDF baselines.

From the readme

Zero-shot email classification with TypeSafe’s Jev TypeSafe’s Jev reached 98.64% accuracy on a 5,733-email ham/spam/phishing test using written category definitions and email context, without task-specific fine-tuning or labeled examples in its requests. A TF-IDF logistic regression classifier trained on roughly 4,600 labeled messages per fold reached …

Details

Section
Benchmarks, evals and calibration
owner
bitnovus
stars
0
forks
0
pushed
2026-09-18
Language
Jupyter Notebook
License
MIT

What it is

Form
Dataset or benchmark
Host agent
Standalone
Audience
Researchers
Maturity
Docs
reports measured numbers

Useful for

Best intent matches