Bookworm-10.7B-v0.4-DPO: benchmark results
Yanolja's Bookworm-10.7B-v0.4-DPO, a DPO-aligned fine-tune of the Korean-vocabulary-extended KoSOLAR-10.7B (itself built on SOLAR-10.7B), trained on Korean-translated SlimOrca and UltraFeedback preference data. Provider: Yanolja. Released 2024-01-18. Access: Open.
Aggregate Rating 1477 ± 19, rank #860 of 1656 rated models, from 19 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-Winogrande | 70.4 | Accuracy (%) | 97.1 |
| Open Korean LLM - Ko-EQ-Bench | 45.07 | EQ-Bench score | 94 |
| Open Korean LLM - KorNAT-CKA | 33.64 | Normalized accuracy (%) | 92.7 |
| Open Ko-LLM Leaderboard | 45.54 | Average (%) | 92.1 |
| Open Korean LLM Leaderboard | 45.54 | Average Score (%) | 92.1 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 45.41 | Exact match, flexible extract (%) | 80.3 |
| Open Korean LLM - KorNAT-Helpful | 49.38 | Normalized accuracy (%) | 78.8 |
| LogicKor - Understanding | 8.07 | Score (0-10) | 58.8 |
| Open Korean LLM - Ko-IFEval | 30.67 | Strict accuracy, prompt/instruction mean (%) | 57.5 |
| LogicKor - Grammar | 5.28 | Score (0-10) | 56.9 |
| Open Korean LLM - KorNAT-Harmless | 62.83 | Normalized accuracy (%) | 50.6 |
| LogicKor - Single-Turn | 5.97 | Score (0-10) | 38 |
Interactive version: theaggregate.ai/model?slug=bookworm-10-7b-v0-4-dpo · How It Works · Data refreshed daily, snapshot 2026-10-20.