gemma-2-27B-it-SimPO-37K-100steps: benchmark results
AALF's SimPO-aligned Gemma 2 27B IT, a 100-step checkpoint from applying on-policy Simple Preference Optimization to Gemma 2 27B IT on 37K UltraFeedback preference pairs. Provider: AALF. Released 2024-08-13. Access: Open.
Aggregate Rating 1556 ± 20, rank #597 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-GSM8K (flexible extract) | 78.24 | Exact match, flexible extract (%) | 100 |
| Open Ko-LLM Leaderboard | 51.66 | Average (%) | 98 |
| Open Korean LLM Leaderboard | 51.66 | Average Score (%) | 98 |
| Open Korean LLM - Ko-IFEval | 63.51 | Strict accuracy, prompt/instruction mean (%) | 97.7 |
| Open Korean LLM - Ko-EQ-Bench | 57.78 | EQ-Bench score | 97.2 |
| Open Korean LLM - KorNAT-CKA | 36.63 | Normalized accuracy (%) | 96.9 |
| Open Korean LLM - Ko-Winogrande | 70.24 | Accuracy (%) | 96.8 |
| Open Portuguese LLM - BLUEX | 71.63 | Accuracy (%) | 92.7 |
| Open Korean LLM - Ko-GPQA-Diamond | 27.27 | Normalized accuracy (%) | 92.4 |
| Open Portuguese LLM - ENEM | 78.17 | Accuracy (%) | 91.8 |
| Open Portuguese LLM - OAB Exams | 56.63 | Accuracy (%) | 90.1 |
| Open Portuguese LLM - FaQuAD NLI | 70.43 | Macro F1 (%) | 61.2 |
Interactive version: theaggregate.ai/model?slug=gemma-2-27b-it-simpo-37k-100steps · How It Works · Data refreshed daily, snapshot 2026-10-23.