Llama-3-Base-8B-SFT-DPO: benchmark results
Princeton NLP's Llama-3-Base-8B-SFT-DPO, a Llama 3 8B base model with SFT and DPO alignment, released alongside Princeton's SimPO preference-optimization paper. Provider: Princeton NLP. Released 2024-05-17. Access: Open.
Aggregate Rating 1442 ± 19, rank #1041 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - KorNAT-Helpful | 55.4 | Normalized accuracy (%) | 95.8 |
| Open Korean LLM - Ko-GPQA-Diamond | 27.78 | Normalized accuracy (%) | 95 |
| Open Korean LLM - KorNAT-Harmless | 67.34 | Normalized accuracy (%) | 89 |
| Open Korean LLM Leaderboard | 42.08 | Average Score (%) | 84.7 |
| Open Ko-LLM Leaderboard | 42.08 | Average (%) | 84.6 |
| Open Korean LLM - Ko-EQ-Bench | 34.21 | EQ-Bench score | 84 |
| Open Portuguese LLM - ASSIN2 RTE | 91.78 | Macro F1 (%) | 70.2 |
| Open LLM Leaderboard - GPQA | 31.04 | Score | 67.9 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 34.27 | Exact match, flexible extract (%) | 67.5 |
| Open Portuguese LLM - OAB Exams | 47.56 | Accuracy (%) | 66.6 |
| Open Portuguese LLM - ENEM | 65.64 | Accuracy (%) | 62.2 |
| Open Korean LLM - Ko-Winogrande | 61.25 | Accuracy (%) | 60.9 |
Interactive version: theaggregate.ai/model?slug=llama-3-base-8b-sft-dpo · How It Works · Data refreshed daily, snapshot 2026-10-23.