WangchanThaiInstruct: Human-Authored Thai Instruction-Following Dataset

WangchanThaiInstruct is a human-authored Thai dataset designed to improve instruction-following in low-resource settings. It captures cultural and domain-specific nuances across four domains and seven task types. The dataset was authored by airesearch and was last updated on 2025-12-16.

Use Cases

• Evaluating instruction-following model performance based on the described cultural and domain-specific nuances.
• Training multitask models based on the dataset's coverage of seven task types.
• Benchmarking low-resource language models based on the dataset's focus on Thai language settings.
• Developing culture-aware AI assistants based on the human-authored, culturally nuanced instructions.

Strengths

• Dataset is human-authored, which likely improves quality and nuance.
• Covers four distinct domains and seven different task types.
• Specifically designed to capture cultural and domain-specific nuances.

Limitations

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

Provenance

Source

airesearch

Collection Method

Human-authored.

Freshness

Last updated 2025-12-16 11:18:27; freshness should be verified.