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