dolphin-2.9.3-mistral-7B-32k: benchmark results
Eric Hartford's Cognitive Computations fine-tune of Mistral-7B-v0.3, using ChatML formatting, a 32K context and a filtered dataset for reduced censorship in responses. Provider: Cognitive Computations. Released 2024-06-25. Access: Open.
Aggregate Rating 1375 ± 19, rank #1322 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - MuSR | 46.43 | Score | 89.4 |
| Open Korean LLM - Ko-IFEval | 41.3 | Strict accuracy, prompt/instruction mean (%) | 84.3 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 36.62 | Exact match, flexible extract (%) | 70.2 |
| Open Portuguese LLM - FaQuAD NLI | 74.05 | Macro F1 (%) | 69.9 |
| Open Portuguese LLM - ASSIN2 RTE | 90.73 | Macro F1 (%) | 60.2 |
| Open Korean LLM - Ko-EQ-Bench | 6.91 | EQ-Bench score | 52.8 |
| Open Korean LLM Leaderboard | 35.32 | Average Score (%) | 51.2 |
| Open Ko-LLM Leaderboard | 35.32 | Average (%) | 51.1 |
| Open Portuguese LLM - ENEM | 60.04 | Accuracy (%) | 46.8 |
| Open Portuguese LLM - BLUEX | 49.37 | Accuracy (%) | 44.9 |
| Open LLM Leaderboard - BBH | 48.13 | Score | 41.9 |
| Open LLM Leaderboard - IFEval | 41.26 | Score | 41.7 |
Interactive version: theaggregate.ai/model?slug=dolphin-2-9-3-mistral-7b-32k · How It Works · Data refreshed daily, snapshot 2026-10-23.