Skip to content

Benchmark

Bridging the Gap Winogrande Sesotho (st) human translation back-translated to English

Multilingual benchmarks compare model performance across languages. Scores are percentages. Higher is better.

Models measured
6
Scores recorded
6
Sources
1
IQ weight
No weight published

Score distribution

58.3%
Mean
54.9%
Median
6.7%
Std dev
53.8%
10th percentile
68.1%
90th percentile
51.3%
Min
51.3%68.1%
The band covers the middle 80% of scores. The line marks the median.

Results

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

ModelScore
GPT 4otable: A.11, section: back-translation-accuracy, task: Winogrande, language: st, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data68.1%
GPT-3.5 Turbotable: A.11, section: back-translation-accuracy, task: Winogrande, language: st, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data54.6%
GPT-4 Turbotable: A.11, section: back-translation-accuracy, task: Winogrande, language: st, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data67.2%
Llama 3 70B Instructtable: A.11, section: back-translation-accuracy, task: Winogrande, language: st, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data53.8%
Llama 3 8B Instructtable: A.11, section: back-translation-accuracy, task: Winogrande, language: st, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data51.3%
Phi-3 Mini 4K Instructtable: A.11, section: back-translation-accuracy, task: Winogrande, language: st, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data54.9%

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)