tulu-2-7B: benchmark results
Allen AI's Tulu 2 7B, a supervised fine-tune of Llama 2 7B on a mix of public, synthetic and human-created datasets, part of the Tulu V2 suite. Provider: Allen AI. Released 2023-11-13. Access: Open.
Aggregate Rating 1331 ± 27, rank #1429 of 1656 rated models, from 23 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LLM Trustworthy - Stereotype | 96.6 | Trust Score (%) | 66.7 |
| LLM Trustworthy - Out-of-Distribution | 69.3 | Trust Score (%) | 58.3 |
| Open Portuguese LLM - FaQuAD NLI | 64.37 | Macro F1 (%) | 48.6 |
| Open CoT - LogiQA | 3.67 | CoT Gain (%) | 43.1 |
| Open Portuguese LLM - ASSIN2 RTE | 87.94 | Macro F1 (%) | 42.9 |
| AI-Secure LLM Trustworthy Leaderboard | 0.64 | Trustworthy average | 41.7 |
| LLM Trustworthy - Adversarial | 44.62 | Trust Score (%) | 41.7 |
| LLM Trustworthy - Adversarial Demo | 60.49 | Trust Score (%) | 41.7 |
| LLM Trustworthy Leaderboard | 63.56 | Average Trust Score (%) | 41.7 |
| Finetuning with Scientific Data Increases Hall | 62.7 | OFS (self-reported) | 41.2 |
| Open CoT - LSAT Analytical Reasoning | 3.04 | CoT Gain (%) | 41.2 |
| LLM Trustworthy - Ethics | 49 | Trust Score (%) | 37.5 |
Interactive version: theaggregate.ai/model?slug=tulu-2-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.