tulu-2-dpo-70B: benchmark results
Allen AI's Tulu 2 DPO fine-tune of Llama 2 70B, trained on the Tulu V2 mix and UltraFeedback as an open alternative to Llama 2 Chat. Provider: Allen AI. Released 2023-11-12. Access: Open.
Aggregate Rating 1439 ± 13, rank #1052 of 1656 rated models, from 62 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Portuguese LLM - FaQuAD NLI | 83.4 | Macro F1 (%) | 96.7 |
| Open LLM Leaderboard v1 - HellaSwag | 88.99 | Normalized accuracy (%) (10-shot) | 95.7 |
| AlpacaEval 1.0 | 95.03 | Win Rate (%) | 94.1 |
| JustEval | 4.82 | Avg Score (1-5) | 93.3 |
| Open LLM Leaderboard v1 - MMLU | 69.84 | Accuracy (%) (5-shot) | 93 |
| Open LLM Leaderboard v1 - ARC Challenge | 72.1 | Normalized accuracy (%) (25-shot) | 91 |
| RewardBench Chat | 97.49 | Accuracy (%) | 90.1 |
| JustEval - Depth | 4.57 | Score (1-5) | 90 |
| Open LLM Leaderboard v1 - WinoGrande | 83.27 | Accuracy (%) (5-shot) | 87.9 |
| Open Portuguese LLM - ASSIN2 RTE | 93.17 | Macro F1 (%) | 87.1 |
| JustEval - Factuality | 4.84 | Score (1-5) | 86.7 |
| AlpacaEval | 84.26 | Length-Controlled Win Rate (%) | 86 |
Interactive version: theaggregate.ai/model?slug=tulu-2-dpo-70b · How It Works · Data refreshed daily, snapshot 2026-10-20.