llama-3-Korean-Bllossom-8B: benchmark results
MLP Lab (Seoultech), Teddysum and Yonsei University's Bllossom-8B, a Korean-English bilingual Llama 3 8B fine-tune with expanded Korean vocabulary, culture-specific instruction tuning and DPO alignment. Provider: Bllossom. Released 2024-04-25. Access: Open.
Aggregate Rating 1436 ± 17, rank #1066 of 1656 rated models, from 25 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-IFEval | 47.89 | Strict accuracy, prompt/instruction mean (%) | 92.3 |
| Open Korean LLM - KorNAT-Helpful | 50.63 | Normalized accuracy (%) | 86.4 |
| Open Korean LLM Leaderboard | 41.08 | Average Score (%) | 80.5 |
| Open Ko-LLM Leaderboard | 41.08 | Average (%) | 80.4 |
| LogicKor - Writing | 9.07 | Score (0-10) | 78.2 |
| Open Korean LLM - KorNAT-Harmless | 64.7 | Normalized accuracy (%) | 76.3 |
| LogicKor - Reasoning | 7.14 | Score (0-10) | 74.4 |
| Open Korean LLM - Ko-EQ-Bench | 25.62 | EQ-Bench score | 71.8 |
| LogicKor - Multi-Turn | 7 | Score (0-10) | 68.3 |
| LogicKor | 6.97 | Score (0-10) | 67.6 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 33.28 | Exact match, flexible extract (%) | 67 |
| LogicKor - Single-Turn | 6.95 | Score (0-10) | 62.2 |
Interactive version: theaggregate.ai/model?slug=llama-3-korean-bllossom-8b · How It Works · Data refreshed daily, snapshot 2026-10-20.