Benchmark
Bridging the Gap MMLU College Medicine English (en) original
Multilingual benchmarks compare model performance across languages. Scores are percentages. Higher is better.
- Models measured
- 6
- Scores recorded
- 19
- Sources
- 1
- IQ weight
- 0.89
Score distribution
- 72.3%
- Mean
- 76.9%
- Median
- 8.9%
- Std dev
- 60.1%
- 10th percentile
- 84.4%
- 90th percentile
- 60.1%
- Min
This benchmark has a weight of 0.89 for chat models, measured over 6 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | GPT 4otable: 2, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 84.4% |
| 2 | GPT-4 Turbotable: 2, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 78% |
| 3 | Llama 3 70B Instructtable: 2, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 76.9% |
| 4 | Phi-3 Mini 4K Instructtable: 2, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 66.5% |
| 5 | GPT-3.5 Turbotable: 2, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 63.6% |
| 6 | Llama 3 8B Instructtable: 2, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 60.1% |
Not ranked
The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.
| Model | Score |
|---|---|
| GPT 4otable: 1, section: baseline, task: MMLU College Medicine· Bridging the Gap (arXiv:2412.12417v1) data | 84.4% |
| GPT 4otable: A.12, section: back-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 84.4% |
| GPT 4otable: A.12, section: machine-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 84.4% |
| GPT-3.5 Turbotable: A.12, section: back-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 63.6% |
| GPT-3.5 Turbotable: A.12, section: machine-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 63.6% |
| GPT-4 Turbotable: A.12, section: back-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 78% |
| GPT-4 Turbotable: A.12, section: machine-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 78% |
| Llama 3 70B Instructtable: A.12, section: back-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 76.9% |
| Llama 3 70B Instructtable: A.12, section: machine-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 76.9% |
| Llama 3 8B Instructtable: A.12, section: back-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 60.1% |
| Llama 3 8B Instructtable: A.12, section: machine-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 60.1% |
| Phi-3 Mini 4K Instructtable: A.12, section: back-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 66.5% |
| Phi-3 Mini 4K Instructtable: A.12, section: machine-translation-accuracy, task: MMLU College Medicine, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data | 66.5% |
Data credits: Tuka Alhanai, Adam Kasumovic, Mohammad Ghassemi, Aven Zitzelberger, Jessica Lundin and Guillaume Chabot-Couture. Bridging the Gap: Enhancing LLM Performance for Low-Resource African Languages with New Benchmarks, Fine-Tuning, and Cultural Adjustments, arXiv:2412.12417v1. CC BY 4.0 article numeric/protocol extract, restructured with values unchanged. No task or model rights asserted. (licence)