Show HN: JevPertus – Jev-style option scoring on Apertus

Category: ai-ml

Tags: llm, multiple-choice, scoring, lora, apertus

Score: 6.0/10 (Innovation: 6, Technical: 6, Documentation: 7, Utility: 5)

JevPertus implements a pointer-head scoring mechanism on top of the Apertus LLM to assign probabilities to answer options for multiple-choice, rating, and true/false questions. It uses LoRA adapters and a lightweight head to score all options in a single forward pass, providing an efficient alternative to generative evaluation. The project includes training scripts, inference code, and benchmark results on MMLU, ARC, and Global-MMLU.

Target audience: AI researchers and ML engineers

Repository: https://github.com/Jaluus/JevPertus · Python · 9 stars

View on Hacker News