Gemma-Ko-Merge: benchmark results
Gunulhona's breadcrumbs_ties merge of Gemma 2 9B tunes (Ataraxy, WPO-HB) with Korean ko-gemma-2-9b-it variants, aimed at Korean chat. Provider: Gunulhona. Released 2024-09-04. Access: Open.
Aggregate Rating 1521 ± 50, rank #2056 of 10916 rated models, from 15 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - KorNAT-CKA | 38.67 | Normalized accuracy (%) | 98 |
| Open Korean LLM - Ko-IFEval | 55.23 | Strict accuracy, prompt/instruction mean (%) | 96.7 |
| Open Korean LLM - KorNAT-Harmless | 68.72 | Normalized accuracy (%) | 92.5 |
| Open Korean LLM - KorNAT-Helpful | 51.03 | Normalized accuracy (%) | 88.7 |
| Open LLM Leaderboard - GPQA | 33.56 | Score | 82.7 |
| Open LLM Leaderboard - BBH | 58.13 | Score | 79.2 |
| Open Korean LLM Leaderboard | 40.68 | Average Score (%) | 78.6 |
| Open LLM Leaderboard - IFEval | 64.16 | Score | 76.6 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 41.32 | Exact match, flexible extract (%) | 75.9 |
| Open LLM Leaderboard - MMLU-Pro | 38.79 | Score | 69.6 |
| Open LLM Leaderboard - MATH Level 5 | 18.81 | Score | 68.8 |
| Open Korean LLM - Ko-Winogrande | 62.19 | Accuracy (%) | 67.2 |
Interactive version: theaggregate.ai/model?slug=gemma-ko-merge · How It Works · Data refreshed daily, snapshot 2026-11-08.