Skip to content

Benchmark

Bridging the Gap Winogrande English (en) original

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

Models and systems measured
46
Scores recorded
65
Sources
1
IQ weight
1.13

Score distribution

70.2%
Mean
69%
Median
9.7%
Std dev
60.2%
10th percentile
83.9%
90th percentile
52%
Min
52%88%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.13 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 or systemScore
1GPT 4otable: 2, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data83.9%
2GPT-4 Turbotable: 2, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data83.5%
3Phi-3 Mini 4K Instructtable: 2, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data64.7%
4Llama 3 70B Instructtable: 2, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data61.2%
5GPT-3.5 Turbotable: 2, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data59.6%
6Llama 3 8B Instructtable: 2, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data52%

Not ranked

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them. Showing 51–59 of 59 results.

Model or systemScore
Llama 3 70B ITLoRA, dataset: Winogrande, language: en, quality: High, quantity: 75%table: A.28, dataset: Winogrande, trainingLanguage: en, quality: High, quantity: 75%, task: Winogrande· Bridging the Gap (arXiv:2412.12417v1) data84.2%
Llama 3 70B ITLoRA, dataset: Winogrande, language: en, quality: Low, quantity: 100%table: A.28, dataset: Winogrande, trainingLanguage: en, quality: Low, quantity: 100%, task: Winogrande· Bridging the Gap (arXiv:2412.12417v1) data84.9%
Llama 3 70B ITLoRA, dataset: Winogrande, language: en, quality: Low, quantity: 25%table: A.28, dataset: Winogrande, trainingLanguage: en, quality: Low, quantity: 25%, task: Winogrande· Bridging the Gap (arXiv:2412.12417v1) data60.3%
Llama 3 70B ITLoRA, dataset: Winogrande, language: en, quality: Low, quantity: 50%table: A.28, dataset: Winogrande, trainingLanguage: en, quality: Low, quantity: 50%, task: Winogrande· Bridging the Gap (arXiv:2412.12417v1) data82.4%
Llama 3 70B ITLoRA, dataset: Winogrande, language: en, quality: Low, quantity: 75%table: A.28, dataset: Winogrande, trainingLanguage: en, quality: Low, quantity: 75%, task: Winogrande· Bridging the Gap (arXiv:2412.12417v1) data83.4%
Llama 3 8B Instructtable: A.11, section: back-translation-accuracy, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data52%
Llama 3 8B Instructtable: A.11, section: machine-translation-accuracy, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data52%
Phi-3 Mini 4K Instructtable: A.11, section: back-translation-accuracy, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data64.7%
Phi-3 Mini 4K Instructtable: A.11, section: machine-translation-accuracy, task: Winogrande, language: en, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data64.7%

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)