Mistral7B-PairRM-SPPO-ExPO: benchmark results
A community ExPO (extrapolated) model from the 'Weak-to-Strong Extrapolation Expedites Alignment' paper, built by extrapolating from UCLA-AGI's Mistral7B-PairRM-SPPO and Mistral-7B-Instruct-v0.2 checkpoints. Provider: chujiezheng. Released 2024-05-04. Access: Open.
Aggregate Rating 1363 ± 17, rank #1359 of 1656 rated models, from 30 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-GPQA-Diamond | 28.79 | Normalized accuracy (%) | 98.1 |
| Open Korean LLM - KorNAT-Harmless | 68.54 | Normalized accuracy (%) | 92.2 |
| Open Chinese LLM - TruthfulQA MC | 61.07 | Accuracy (%) | 86.1 |
| AlpacaEval 2.0 | 31.82 | LC Win Rate (%) | 72.1 |
| Open Korean LLM - KorNAT-Helpful | 48 | Normalized accuracy (%) | 63.5 |
| Open Korean LLM - Ko-EQ-Bench | 7.84 | EQ-Bench score | 54.3 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 18.35 | Exact match, flexible extract (%) | 49.4 |
| Open Portuguese LLM - ASSIN2 RTE | 89.2 | Macro F1 (%) | 48.7 |
| Open Korean LLM - Ko-IFEval | 28.57 | Strict accuracy, prompt/instruction mean (%) | 48.6 |
| Open LLM Leaderboard - MuSR | 40.55 | Score | 46.5 |
| Open Portuguese LLM - BLUEX | 49.51 | Accuracy (%) | 45.3 |
| Open Portuguese LLM - FaQuAD NLI | 58.93 | Macro F1 (%) | 43 |
Interactive version: theaggregate.ai/model?slug=mistral7b-pairrm-sppo-expo · How It Works · Data refreshed daily, snapshot 2026-10-20.