Llama 3 8B Magpie Align V0.3: benchmark results
Magpie-Align's Llama-3-8B-Magpie-Align-v0.3, a Llama 3 8B aligned via SFT on Magpie-generated instruction/reasoning/Chinese data, then DPO on the UltraFeedback-ArmoRM dataset. Provider: Magpie. Released 2024-07-15. Access: Open.
Aggregate Rating 1425 ± 17, rank #1125 of 1656 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - KorNAT-Helpful | 53.2 | Normalized accuracy (%) | 94.2 |
| Open Korean LLM - KorNAT-Harmless | 71.43 | Normalized accuracy (%) | 94 |
| Open Korean LLM - KorNAT-CKA | 31.86 | Normalized accuracy (%) | 87.8 |
| Open Ko-LLM Leaderboard | 40.84 | Average (%) | 78.8 |
| Open Korean LLM Leaderboard | 40.67 | Average Score (%) | 78.5 |
| Open Korean LLM - Ko-Winogrande | 62.83 | Accuracy (%) | 70.8 |
| Open Korean LLM - Ko-EQ-Bench | 24.57 | EQ-Bench score | 70.3 |
| Open Korean LLM - Ko-IFEval | 33.02 | Strict accuracy, prompt/instruction mean (%) | 64.7 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 29.11 | Exact match, flexible extract (%) | 63.4 |
| Open Portuguese LLM - OAB Exams | 45.15 | Accuracy (%) | 60 |
| Open Portuguese LLM - ENEM | 64.52 | Accuracy (%) | 58.9 |
| Open Chinese LLM - TruthfulQA MC | 54.4 | Accuracy (%) | 56.5 |
Interactive version: theaggregate.ai/model?slug=llama-3-8b-magpie-align-v0-3 · How It Works · Data refreshed daily, snapshot 2026-10-20.