AI Potluck
Contribute
Back to Gap Map Model components / Language-specific datasets

Typhoon-S sovereign capability dataset

SCB 10X
open / Overall score: 1.7

The Typhoon-S sovereign capability dataset holds the Thai training and test files behind the region-specific part of the Typhoon-S post-training study. Legal-domain files derive from NitiBench and WangchanX-Legal-ThaiCCL-RAG, split into reinforcement-learning, pretraining, supervised and agent sets, and general-domain files derive from MIRAGE. The release totals about 30,000 rows in Thai and English and comes from the Typhoon team.

Openness

4 high confidence
4.0
license
odc-by(card metadata
access
public(ungated)
dataset_card
partial(lists files, example counts and original sources, but not how the splits were derived)

The files download without a gate under a permissive license and name the datasets they were cut from. The card lists files rather than explaining how they were built, so the method has to be read from the Typhoon-S report.

Adoption

1 high confidence
1.0

Adoption is Hugging Face downloads of the repository. It cannot separate reproductions of the Typhoon-S experiments from other uses.

Capability

2 medium confidence
2.0

This release packages training and test splits cut from existing Thai legal and retrieval datasets for the Typhoon-S study. It is small and derivative and serves that one project's experiments, unlike WangchanThaiInstruct, a general Thai instruction set written by people across four domains.

Verified 2026-09-24