free-llama3-dpo-v0.2: benchmark results
Freewheelin AI's free-llama3-dpo v0.2, a Llama 3 8B DPO fine-tune trained with TRL using the method from the SOLAR paper. Provider: Freewheelin. Released 2024-05-09. Access: Open.
Aggregate Rating 1380 ± 22, rank #1259 of 1635 rated models, from 19 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 66.59 | Accuracy (%) (5-shot) | 87.7 |
| Open Portuguese LLM - BLUEX | 58.69 | Accuracy (%) | 74.2 |
| Open Portuguese LLM - ENEM | 68.02 | Accuracy (%) | 68.3 |
| Open Portuguese LLM - OAB Exams | 47.88 | Accuracy (%) | 67.7 |
| Open LLM Leaderboard v1 - GSM8K | 44.66 | Accuracy (%) (5-shot) | 58.7 |
| Open LLM Leaderboard v1 - WinoGrande | 77.27 | Accuracy (%) (5-shot) | 54.6 |
| Open Portuguese LLM - ASSIN2 RTE | 89.54 | Macro F1 (%) | 50.9 |
| Open Chinese LLM - CMMLU | 53.28 | Accuracy (%) | 48.8 |
| Open Chinese LLM - WinoGrande | 62.75 | Accuracy (%) | 47.6 |
| Open LLM Leaderboard v1 - HellaSwag | 81.88 | Normalized accuracy (%) (10-shot) | 47.5 |
| Open LLM Leaderboard v1 - ARC Challenge | 59.9 | Normalized accuracy (%) (25-shot) | 45.4 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 45.83 | MC2 (%) (0-shot) | 32.4 |
Interactive version: theaggregate.ai/model?slug=free-llama3-dpo-v0-2 · How It Works · Data refreshed daily, snapshot 2026-11-05.