QazUNTv2: Dataset of high school math problems on english and russian languages

This dataset is intended for the subsequent verification of the correctness of LLM (GPT-3.5 Turbo) generated responses to mathematics problems similar to those found in exams for graduate schools. The dataset includes problems and their types in both Russian and English, along with five options, manually solved answers, and detailed solutions in both languages. The primary goal is to analyze and compare LLM (GPT-3.5 Turbo)-generated answers with provided correct solutions.
The data has been collected and structured across the following sections of mathematics: Algebra, Probability and Logic. We ensured a comprehensive evaluation of GPT's capabilities in understanding and solving these problems. The dataset is divided into the following sections with the corresponding number of problems:
1. Algebra: 436 problems;
2. Logic: 312 problems;
3. Probability: 163 problems.
This dataset will facilitate a detailed assessment of GPT's performance in mathematical problem-solving across various domains.
For the future analysis, we also calculated quantity of tokens that may help to generate responses from ChatGPT-3.5 Turbo:
The English math problems comprise 37563 tokens.
The Russian math problems comprise 66406 tokens.
The average number of tokens per English task is approximately 39.96 and for the problems in Russian this number is approximately 70.64.