nox-solar-10.7B-v4: benchmark results
davidkim205's nox-solar 10.7B v4, an SFT and DPO fine-tune of Edentns' DataVortexS-10.7B-dpo-v1.8 using Korean conversation and comparison datasets. Provider: davidkim205. Released 2024-03-16. Access: Open.
Aggregate Rating 1387 ± 18, rank #1265 of 1656 rated models, from 25 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-Winogrande | 73.88 | Accuracy (%) | 98.9 |
| Open Korean LLM - KorNAT-Helpful | 67.5 | Normalized accuracy (%) | 98.7 |
| Open Korean LLM - KorNAT-Harmless | 78.26 | Normalized accuracy (%) | 98.4 |
| Open Korean LLM - Ko-GPQA-Diamond | 26.77 | Normalized accuracy (%) | 88.9 |
| Open LLM Leaderboard - MuSR | 42.98 | Score | 70.2 |
| Open LLM Leaderboard - GPQA | 30.7 | Score | 64.2 |
| Open Ko-LLM Leaderboard | 37.58 | Average (%) | 63.2 |
| Open Korean LLM Leaderboard | 37.58 | Average Score (%) | 63.2 |
| Open Korean LLM - KorNAT-CKA | 25.85 | Normalized accuracy (%) | 61.1 |
| LogicKor - Grammar | 5.07 | Score (0-10) | 52.9 |
| Open Korean LLM - Ko-IFEval | 28.73 | Strict accuracy, prompt/instruction mean (%) | 49.3 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 17.97 | Exact match, flexible extract (%) | 49.1 |
Interactive version: theaggregate.ai/model?slug=nox-solar-10-7b-v4 · How It Works · Data refreshed daily, snapshot 2026-10-20.