Configurable-Hermes-2-Pro-Llama-3-8B: benchmark results
vicgalle's configurable-safety-tuning (CST) research fine-tune of Hermes 2 Pro Llama 3 8B, letting the system prompt set safety behavior at inference. Provider: vicgalle. Released 2024-05-02. Access: Open.
Aggregate Rating 1457 ± 15, rank #954 of 1656 rated models, from 41 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM Leaderboard | 45.29 | Average Score (%) | 91.5 |
| Open Ko-LLM Leaderboard | 45.29 | Average (%) | 91.4 |
| Open Korean LLM - KorNAT-Harmless | 67.84 | Normalized accuracy (%) | 90.8 |
| Open Korean LLM - Ko-GPQA-Diamond | 26.77 | Normalized accuracy (%) | 88.9 |
| Open LLM Leaderboard v1 - GSM8K | 68.46 | Accuracy (%) (5-shot) | 85.9 |
| Open Korean LLM - Ko-EQ-Bench | 35.66 | EQ-Bench score | 85.8 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 52.01 | Exact match, flexible extract (%) | 85.2 |
| Open LLM Leaderboard v1 - MMLU | 65.23 | Accuracy (%) (5-shot) | 83.2 |
| Open Korean LLM - Ko-IFEval | 39.32 | Strict accuracy, prompt/instruction mean (%) | 80.8 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 62.51 | MC2 (%) (0-shot) | 79.3 |
| Open Portuguese LLM - ASSIN2 RTE | 92.24 | Macro F1 (%) | 76.6 |
| Open CoT - LSAT Logical Reasoning | 15.69 | CoT Gain (%) | 75.2 |
Interactive version: theaggregate.ai/model?slug=configurable-hermes-2-pro-llama-3-8b · How It Works · Data refreshed daily, snapshot 2026-10-20.