MGSM

Dataset Card for MGSM Dataset Summary 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. The same 250 problems from GSM8K are each translated via human annotators in 10 languages. The 10 languag

Papers using MGSM (6)

Layer-wise Swapping for Generalizable Multilingual Safety2026What Really Counts? Examining Step and Token Level Attribution in Multilingual CoT Reasoning2025MERLIN: Multi-Stage Curriculum Alignment for Multilingual Encoder-LLM Integration in Cross-Lingual Reasoning2025Type-Compliant Adaptation Cascades: Adapting Programmatic LM Workflows to Data2025Cross-Lingual Consistency: A Novel Inference Framework for Advancing Reasoning in Large Language Models2025Language Imbalance Driven Rewarding for Multilingual Self-improving2024