Configurable-Llama-3.1-8B-Instruct: benchmark results
vicgalle's configurable-safety-tuning research fine-tune of Llama 3.1 8B Instruct, letting system prompts set safety behavior at inference time. Provider: vicgalle. Released 2024-07-24. Access: Open.
Aggregate Rating 1488 ± 20, rank #815 of 1656 rated models, from 25 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - IFEval | 83.12 | Score | 99.3 |
| Open Korean LLM - Ko-IFEval | 54.92 | Strict accuracy, prompt/instruction mean (%) | 96.4 |
| Open Ko-LLM Leaderboard | 46.55 | Average (%) | 93.4 |
| Open Korean LLM Leaderboard | 46.55 | Average Score (%) | 93.4 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 56.94 | Exact match, flexible extract (%) | 91.1 |
| Open Korean LLM - KorNAT-CKA | 33.12 | Normalized accuracy (%) | 90.7 |
| Open Korean LLM - Ko-GPQA-Diamond | 25.76 | Normalized accuracy (%) | 81.3 |
| Open Portuguese LLM - OAB Exams | 51.48 | Accuracy (%) | 80.3 |
| Open Korean LLM - Ko-EQ-Bench | 30.06 | EQ-Bench score | 78.7 |
| Open Portuguese LLM - ENEM | 70.33 | Accuracy (%) | 75.7 |
| Open Portuguese LLM - BLUEX | 58.69 | Accuracy (%) | 74.2 |
| Open Portuguese LLM - ASSIN2 RTE | 91.9 | Macro F1 (%) | 71.8 |
Interactive version: theaggregate.ai/model?slug=configurable-llama-3-1-8b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.