Llama 3 8B SFR Iterative DPO R: benchmark results
Salesforce's Llama-3-8B-SFR-Iterative-DPO-R, trained with an iterative online DPO-based RLHF recipe on open-source datasets only, from Salesforce's RLHF research initiative. Provider: Salesforce. Released 2024-05-09. Access: Open.
Aggregate Rating 1428 ± 17, rank #1109 of 1656 rated models, from 28 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 66.39 | Accuracy (%) (5-shot) | 87.1 |
| Open LLM Leaderboard v1 - GSM8K | 67.25 | Accuracy (%) (5-shot) | 83.3 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 61.47 | MC2 (%) (0-shot) | 77.8 |
| Open Portuguese LLM - ASSIN2 RTE | 91.91 | Macro F1 (%) | 72.2 |
| Open LLM Leaderboard v1 - WinoGrande | 79.79 | Accuracy (%) (5-shot) | 71 |
| Open LLM Leaderboard v1 - ARC Challenge | 64.93 | Normalized accuracy (%) (25-shot) | 68.5 |
| Open CoT - LSAT Analytical Reasoning | 5.22 | CoT Gain (%) | 68.3 |
| Open Portuguese LLM - OAB Exams | 47.43 | Accuracy (%) | 66.1 |
| Open Portuguese LLM - ENEM | 66.27 | Accuracy (%) | 63.9 |
| Open LLM Leaderboard v1 - HellaSwag | 83.6 | Normalized accuracy (%) (10-shot) | 60 |
| Open Portuguese LLM - FaQuAD NLI | 69.87 | Macro F1 (%) | 59.9 |
| Open Portuguese LLM - BLUEX | 53.41 | Accuracy (%) | 55.9 |
Interactive version: theaggregate.ai/model?slug=llama-3-8b-sfr-iterative-dpo-r · How It Works · Data refreshed daily, snapshot 2026-10-20.