RYS-Llama-3.1-8B-Instruct: benchmark results
dnhkng's experimental Llama 3.1 8B Instruct variant built with his unpublished 'RYS' model-modification technique. Provider: dnhkng. Released 2024-08-08. Access: Open.
Aggregate Rating 1439 ± 19, rank #1061 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - IFEval | 76.85 | Score | 92.7 |
| Open Korean LLM - Ko-IFEval | 46.05 | Strict accuracy, prompt/instruction mean (%) | 90.6 |
| Open Korean LLM - KorNAT-Helpful | 50.13 | Normalized accuracy (%) | 84 |
| Open Portuguese LLM - ASSIN2 RTE | 92.77 | Macro F1 (%) | 83.4 |
| Open Korean LLM Leaderboard | 41.45 | Average Score (%) | 82.9 |
| Open Ko-LLM Leaderboard | 41.45 | Average (%) | 82.8 |
| Open Portuguese LLM - OAB Exams | 51.39 | Accuracy (%) | 79.7 |
| Open Portuguese LLM - ENEM | 70.54 | Accuracy (%) | 76.7 |
| Open Portuguese LLM - FaQuAD NLI | 76.01 | Macro F1 (%) | 75.2 |
| Open Portuguese LLM - BLUEX | 58 | Accuracy (%) | 71.7 |
| Open Korean LLM - Ko-GPQA-Diamond | 24.75 | Normalized accuracy (%) | 69.5 |
| Open Korean LLM - KorNAT-Harmless | 63.87 | Normalized accuracy (%) | 67.6 |
Interactive version: theaggregate.ai/model?slug=rys-llama-3-1-8b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-23.