Logo Lanfrica

Multilingual Grade School Math Benchmark (MGSM)

Domain:

natural language processing

Record type:

dataset
Creator:
yuj
Host:
Multilingual Grade School Math Benchmark (MGSM) is a benchmark of grade-school math problems, proposed in the paper [Language models are multilingual chain-of-thought reasoners](Language Models are Multili…). The same 250 problems from [GSM8K](arxiv.org) are each translated via human annotators in 10 languages. The 10 languages are: - Spanish - French - German - Russian - Chinese - Japanese - Thai - Swahili - Bengali - Telugu You can find the input and targets for each of the ten languages (and English) as `.tsv` files. We also include few-shot exemplars that are also manually translated from each language in `exemplars.py`.