My Trained Models Leaderboard

About the Benchmarks

This leaderboard evaluates fine-tuned language models using established, publicly available benchmarks. The goal is to measure model capabilities and identify potential regressions introduced during fine-tuning.

MMLU, HellaSwag, and GPQA: Evaluated using EleutherAI's lm-evaluation-harness. These benchmarks assess general knowledge, commonsense reasoning, and graduate-level reasoning.

StrongREJECT: Evaluated using Inspect AI. This benchmark measures model responses to harmful requests and evaluates refusal behavior.