Skip to content

Benchmark

Bridging the Gap Winogrande Amharic (am) human-translated

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

Models and systems measured
30
Scores recorded
32
Sources
1
IQ weight
1.21

Score distribution

50.6%
Mean
50.4%
Median
1.7%
Std dev
49.7%
10th percentile
50.8%
90th percentile
49.1%
Min
49.1%59.4%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.21 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: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data59.4%
2GPT-3.5 Turbotable: 2, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data51.3%
3GPT-4 Turbotable: 2, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data51%
4Llama 3 70B Instructtable: 2, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.8%
5Llama 3 8B Instructtable: 2, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.3%
6Phi-3 Mini 4K Instructtable: 2, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.2%

Not ranked

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

Model or systemScore
Llama 3 70B Instructtable: A.20, row: base, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.8%
Llama 3 70B Instructtable: A.21, row: base, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.8%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: aftable: A.20, training: Winogrande train (small), trainingLanguage: af, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data49.4%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: amtable: A.20, training: Winogrande train (small), trainingLanguage: am, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.4%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: bmtable: A.20, training: Winogrande train (small), trainingLanguage: bm, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.5%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: entable: A.20, training: Winogrande train (small), trainingLanguage: en, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.1%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: igtable: A.20, training: Winogrande train (small), trainingLanguage: ig, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.2%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: nsotable: A.20, training: Winogrande train (small), trainingLanguage: nso, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.5%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: sntable: A.20, training: Winogrande train (small), trainingLanguage: sn, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.8%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: sttable: A.20, training: Winogrande train (small), trainingLanguage: st, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data49.8%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: tntable: A.20, training: Winogrande train (small), trainingLanguage: tn, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data49.9%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: tstable: A.20, training: Winogrande train (small), trainingLanguage: ts, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data49.3%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: xhtable: A.20, training: Winogrande train (small), trainingLanguage: xh, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data49.8%
Llama 3 70B ITLoRA, dataset: Winogrande train (small), language: zutable: A.20, training: Winogrande train (small), trainingLanguage: zu, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data49.1%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: aftable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: af, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.2%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: amtable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: am, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data49.9%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: bmtable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: bm, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.4%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: entable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: en, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: igtable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: ig, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.2%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: nsotable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: nso, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.3%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: sntable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: sn, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data49.7%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: sttable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: st, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.8%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: tntable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: tn, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.8%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: tstable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: ts, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.5%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: xhtable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: xh, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.8%
Llama 3 70B ITLoRA, dataset: MMLU College Medicine dev + test + val, language: zutable: A.21, training: MMLU College Medicine dev + test + val, trainingLanguage: zu, task: Winogrande, language: am, shots: 5· Bridging the Gap (arXiv:2412.12417v1) data50.8%

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)