Llama-3-Base-8B-SFT-KTO: benchmark results
Princeton NLP's Llama-3-Base-8B-SFT-KTO, a Llama 3 8B base model with SFT and KTO alignment, released alongside Princeton's SimPO preference-optimization paper. Provider: Princeton NLP. Released 2024-05-17. Access: Open.
Aggregate Rating 1449 ± 19, rank #1008 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - KorNAT-Helpful | 54.7 | Normalized accuracy (%) | 95.3 |
| Open Korean LLM - KorNAT-Harmless | 67.04 | Normalized accuracy (%) | 87.7 |
| Open Korean LLM Leaderboard | 42.78 | Average Score (%) | 86.9 |
| Open Korean LLM - Ko-EQ-Bench | 36.39 | EQ-Bench score | 86.8 |
| Open Ko-LLM Leaderboard | 42.78 | Average (%) | 86.7 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 41.85 | Exact match, flexible extract (%) | 76.8 |
| Open Korean LLM - Ko-IFEval | 33.79 | Strict accuracy, prompt/instruction mean (%) | 68.7 |
| Open Portuguese LLM - ASSIN2 RTE | 91.5 | Macro F1 (%) | 67.3 |
| Open Portuguese LLM - OAB Exams | 46.65 | Accuracy (%) | 64.3 |
| Open LLM Leaderboard - GPQA | 30.54 | Score | 62.3 |
| Open Korean LLM - KorNAT-CKA | 26.05 | Normalized accuracy (%) | 61.9 |
| Open Portuguese LLM - ENEM | 65.22 | Accuracy (%) | 60.8 |
Interactive version: theaggregate.ai/model?slug=llama-3-base-8b-sft-kto · How It Works · Data refreshed daily, snapshot 2026-10-23.