ytu-ce-cosmos/gsm8k_tr

---
license: apache-2.0
size_categories:
- 1K<n<10K
---
This is the Turkish version of the GSM8K dataset, one of the most widely used benchmark datasets today. You can find the original dataset here:
<br>
[https://huggingface.co/datasets/openai/gsm8k](https://huggingface.co/datasets/openai/gsm8k)
<br>
We translated the dataset into Turkish using the following methodology:

- The questions were translated into Turkish using DeepL.
- The answers were generated from Turkish questions using GPT-4o, as we believe this approach yields much better results compared to simply translating the answers from another language.

The following Turkish prompt was used to generate concise and human-readable answers.

- “Soruya kısa ve net bir yanıt ver, yanıtlarında ‘24 \times \frac’ gibi notasyonlar kullanma.”

_Translates to: Provide a short and clear answer; do not use notations like ‘24 \times \frac’ in your answers._

We strongly believe that this is the best Turkish localization of the GSM8K dataset.

### Contact
COSMOS AI Research Group, Yildiz Technical University Computer Engineering Department <br>
https://cosmos.yildiz.edu.tr/ <br>
[e-mail]