bagel-dpo-34B-v0.5: benchmark results
Jon Durbin's bagel v0.5 DPO tune of Yi-34B-200K, SFT'd on 40+ datasets in four prompt formats then refined with preference DPO. Provider: Jon Durbin. Released 2024-04-01. Access: Open.
Aggregate Rating 1456 ± 23, rank #962 of 1656 rated models, from 25 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - MMLU | 75.78 | Accuracy (%) (5-shot) | 97.1 |
| Open Chinese LLM - C-Eval Semantic | 90.06 | Accuracy (%) | 92.9 |
| Open Chinese LLM - ARC Challenge | 62.37 | Accuracy (%) | 92.8 |
| Open Chinese LLM - CMMLU | 71.31 | Accuracy (%) | 91.4 |
| Open Portuguese LLM - BLUEX | 66.76 | Accuracy (%) | 89.3 |
| Open Portuguese LLM - OAB Exams | 54.17 | Accuracy (%) | 85.3 |
| Open Chinese LLM - HellaSwag | 67.78 | Accuracy (%) | 84.5 |
| Open LLM Leaderboard v1 - WinoGrande | 82.4 | Accuracy (%) (5-shot) | 83.9 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 65.08 | MC2 (%) (0-shot) | 83 |
| Open Chinese LLM - WinoGrande | 67.32 | Accuracy (%) | 81.8 |
| Open Portuguese LLM - ENEM | 71.66 | Accuracy (%) | 81.1 |
| Open LLM Leaderboard v1 - ARC Challenge | 68.34 | Normalized accuracy (%) (25-shot) | 80.1 |
Interactive version: theaggregate.ai/model?slug=bagel-dpo-34b-v0-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.