Skip to content

Benchmark

Macedonian LM Evaluation Harness HellaSwag accuracy

Multilingual benchmarks compare model performance across languages. Higher scores are better.

Models and systems measured
12
Scores recorded
12
Sources
1
IQ weight
No weight published

Score distribution

0.385
Mean
0.4
Median
0.065
Std dev
0.27
10th percentile
0.45
90th percentile
0.263
Min
0.2630.466
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.

Model or systemScore
domestic-yak-basetable: 4, task: HellaSwag, stage: domestic-yak-base· Macedonian Foundation Model (arXiv:2506.09560v1) data0.43
domestic-yak-instructtable: 4, task: HellaSwag, stage: domestic-yak-instruct· Macedonian Foundation Model (arXiv:2506.09560v1) data0.45
domestic-yak†8Btable: 3, task: HSwag· Macedonian Foundation Model (arXiv:2506.09560v1) data0.448
Llama 3.18Btable: 3, task: HSwag· Macedonian Foundation Model (arXiv:2506.09560v1) data0.374
Llama 3.1Llama-3.1-8B-Instructtable: 4, task: HellaSwag, stage: Llama 3.1· Macedonian Foundation Model (arXiv:2506.09560v1) data0.37
Llama 3.21Btable: 3, task: HSwag· Macedonian Foundation Model (arXiv:2506.09560v1) data0.27
Llama 3.370Btable: 3, task: HSwag· Macedonian Foundation Model (arXiv:2506.09560v1) data0.466
Mistral7Btable: 3, task: HSwag· Macedonian Foundation Model (arXiv:2506.09560v1) data0.372
Mistral Nemo12Btable: 3, task: HSwag· Macedonian Foundation Model (arXiv:2506.09560v1) data0.4
MKLLM†7Btable: 3, task: HSwag· Macedonian Foundation Model (arXiv:2506.09560v1) data0.433
Phi-3.5-mini3.8Btable: 3, task: HSwag· Macedonian Foundation Model (arXiv:2506.09560v1) data0.263
Qwen2.57Btable: 3, task: HSwag· Macedonian Foundation Model (arXiv:2506.09560v1) data0.339

Data credits: Stefan Krsteski, Matea Tashkovska, Borjan Sazdov, Hristijan Gjoreski and Branislav Gerazov. Towards Open Foundation Language Model and Corpus for Macedonian, arXiv:2506.09560v1. CC BY 4.0 article numeric/protocol extract; restructured, printed values unchanged. No task, corpus, weight or raw survey rights asserted. (licence)