datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Indian_Climate_Disaster_data
BharatCRIC: Indian Climate Disaster Data (Heatwave Advisories + Scam Pairs)
Generated by scripts/grade_a_rebuild.py with blueprint grade_a_2026_04_30.
File
Use
instruction_dataset_main.jsonl
Full structured instruction upload (1680 rows)
instruction_dataset_smoke.jsonl
50-row smoke test, 5 languages x 5 formats x 2 rows
instruction_dataset_smoke_v2.jsonl
Same smoke set for the previous filename expected by notes
preference_pairs_scams.jsonl
Mirrored genuine-vs-scam… See the full description on the dataset page: https://huggingface.co/datasets/sahilmaniyar888/Indian_Climate_Disaster_data.safestep-disaster-instruction-base
SAFEstep baseline instruction dataset
This is the original 135-example SAFEstep baseline: 105 training examples and
30 held-out validation examples. It is published separately for reproducibility;
the later safestep-disaster-instruction repository supersedes it and includes
much broader natural-disaster, first-aid and biological-hazard coverage.
See DATASET_CARD.md for schema, generation method, intended use, limitations,
validation, attribution and safety boundaries.
This is a… See the full description on the dataset page: https://huggingface.co/datasets/0xKitkat/safestep-disaster-instruction-base.