KoMT-Bench: Korean Instruction-Following Benchmark for Language Models

KoMT-Bench is a benchmark dataset created by LGAI-EXAONE to evaluate the instruction-following capability of language models in Korean. It was developed by translating and culturally adapting the MT-Bench dataset, with expert linguist review. The dataset was last updated on HuggingFace on August 8, 2024.

Use Cases

• Benchmarking LLM performance on Korean instruction-following based on translated and culturally adapted questions.
• Evaluating cultural and linguistic nuance understanding in Korean language models.
• Training or fine-tuning models for improved Korean language comprehension and response generation.

Strengths

• Dataset is based on the established MT-Bench benchmark, providing a structured evaluation framework.
• Includes expert linguist review to ensure cultural and linguistic accuracy for Korean.
• Specifically designed to reflect Korean language characteristics and cultural nuances.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown, which may limit suitability assessment.

Provenance

Source

LGAI-EXAONE

Collection Method

Translated and culturally adapted from the MT-Bench dataset, with expert linguist review.

Time Range

Creation date is not specified; last platform update was 2024-08-08.

Freshness

Last updated 2024-08-08 00:42:43; freshness should be verified.

Geography

Focus is on the Korean language and cultural context.