metadata

license: apache-2.0
language:
  - sr
task_categories:
  - question-answering

OZ Eval

Dataset Description

OZ Eval (sr. Opšte Znanje Evaluacija) dataset was created for the purposes of evaluating General Knowledge of LLM models in Serbian language. Data consists of 1k+ high-quality questions and answers which were used as part of entry exams at the Faculty of Philosophy and Faculty of Organizational Sciences, University of Belgrade. The exams test the General Knowledge of students and were used in the enrollment periods from 2003 to 2024.

This is a joint work with @Stopwolf!

Evaluation process

Models are evaluated by the following principle using HuggingFace's library lighteval. We supply the model with the following template:

Pitanje: {question}

Ponuđeni odgovori:
A. {option_a}
B. {option_b}
C. {option_c}
D. {option_d}
E. {option_e}

Krajnji odgovor:

We then compare likelihoods of each letter (A, B, C, D, E) and calculate the final accuracy. All evaluations are ran in a 0-shot manner using a chat template.

GPT-like models were evaluated by taking top 20 probabilities of the first output token, which were further filtered for letters A to E. Letter with the highest probability was taken as a final answer.

Exact code for the task will be posted here.

Evaluation results

Model	Accuracy		Stderr
GPT-4-0125-preview	0.9199	±	0.002
GPT-4o-2024-05-13	0.9196	±	0.0017
GPT-3.5-turbo-0125	0.8245	±	0.0016
Tito-7B-slerp	0.7099	±	0.0101
Qwen2-7B-instruct	0.673	±	0.0105
Yugo60-GPT	0.6411	±	0.0107
Hermes-2-Theta-Llama-3-8B	0.5852	±	0.011
Llama3-8B-Instruct	0.5274	±	0.0111
Hermes-2-Pro-Mistral-7B	0.5145	±	0.0112
Perucac-7B-slerp	0.4247	±	0.011
SambaLingo-Serbian-Chat	0.2802	±	0.01
Gemma-2-9B-it	0.2193	±	0.0092

Citation

@article{oz-eval,
  author    = "Stanivuk Siniša & Đorđević Milena",
  title     = "Opšte znanje LLM Eval",
  year      = "2024"
  howpublished = {\url{https://huggingface.co/datasets/DjMel/oz-eval}},
}