Configurable-Llama-3-8B-v0.3: benchmark results
vicgalle's configurable-safety-tuning research fine-tune of Meta's Llama 3 8B, letting the system prompt set the model's safety level. Provider: vicgalle. Released 2024-04-20. Access: Open.
Aggregate Rating 1435 ± 15, rank #1076 of 1656 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-IFEval | 50.57 | Strict accuracy, prompt/instruction mean (%) | 93.9 |
| Open Korean LLM - KorNAT-Helpful | 52.4 | Normalized accuracy (%) | 92.8 |
| Open LLM Leaderboard v1 - GSM8K | 69.9 | Accuracy (%) (5-shot) | 90.8 |
| Open LLM Leaderboard v1 - MMLU | 67.2 | Accuracy (%) (5-shot) | 89 |
| Open Korean LLM - KorNAT-Harmless | 66.24 | Normalized accuracy (%) | 85 |
| Open Korean LLM - Ko-GPQA-Diamond | 25.76 | Normalized accuracy (%) | 81.3 |
| Open Korean LLM - Ko-EQ-Bench | 30.92 | EQ-Bench score | 79.9 |
| Open Korean LLM Leaderboard | 40.66 | Average Score (%) | 78.2 |
| Open Ko-LLM Leaderboard | 40.66 | Average (%) | 78 |
| Open Portuguese LLM - OAB Exams | 49.98 | Accuracy (%) | 75.4 |
| Open Portuguese LLM - ENEM | 69.35 | Accuracy (%) | 73 |
| Open Portuguese LLM - ASSIN2 RTE | 91.91 | Macro F1 (%) | 72.2 |
Interactive version: theaggregate.ai/model?slug=configurable-llama-3-8b-v0-3 · How It Works · Data refreshed daily, snapshot 2026-10-20.