Llama-3-Instruct-8B-SPPO-Iter2: benchmark results
UCLA-AGI's Llama-3-Instruct-8B-SPPO-Iter2, the second iteration of Self-Play Preference Optimization applied to Llama 3 8B Instruct, trained on synthetic UltraFeedback-derived responses. Provider: UCLA. Released 2024-06-25. Access: Open.
Aggregate Rating 1427 ± 17, rank #1117 of 1656 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-GSM8K (flexible extract) | 55.65 | Exact match, flexible extract (%) | 89.7 |
| Open Korean LLM - KorNAT-Helpful | 50.7 | Normalized accuracy (%) | 86.7 |
| Open Korean LLM - Ko-IFEval | 41.75 | Strict accuracy, prompt/instruction mean (%) | 85.1 |
| Open LLM Leaderboard - IFEval | 69.89 | Score | 82.8 |
| Open Korean LLM - Ko-EQ-Bench | 29.31 | EQ-Bench score | 77.7 |
| Open Portuguese LLM - OAB Exams | 50.07 | Accuracy (%) | 75.8 |
| Open Portuguese LLM - ENEM | 70.19 | Accuracy (%) | 75.2 |
| Open Chinese LLM - GSM8K | 52.31 | Accuracy (%) | 71.9 |
| Open Korean LLM Leaderboard | 39.25 | Average Score (%) | 70.8 |
| Open Ko-LLM Leaderboard | 39.25 | Average (%) | 70.7 |
| Open Portuguese LLM - BLUEX | 57.72 | Accuracy (%) | 70.7 |
| Open Korean LLM - Ko-GPQA-Diamond | 24.75 | Normalized accuracy (%) | 69.5 |
Interactive version: theaggregate.ai/model?slug=llama-3-instruct-8b-sppo-iter2 · How It Works · Data refreshed daily, snapshot 2026-10-20.