Sarthi Agridata: 220,222 Hindi Agri-Advisory Synthetic Examples

220,222 synthetic data points for agricultural advisory tasks, generated by Google Gemini 2.5 Flash. The dataset is designed for instruction tuning and chain-of-thought reasoning, with Hindi as the target output language and English used for internal reasoning and metadata. Soketlabs published the dataset on Hugging Face, with a last update recorded on January 16, 2026.

Use Cases

• Train instruction-following models for agricultural advice based on the described task category.
• Develop chain-of-thought reasoning models for agricultural problem-solving based on the dataset's design.
• Fine-tune language models for Hindi-language agricultural Q&A based on the target output language.
• Benchmark synthetic data generation methods for domain-specific tasks based on the use of a source model.

Strengths

• Contains 220,222 total data points, providing a substantial volume for training.
• Dataset is structured for chain-of-thought and instruction tuning, a specific and useful task format.
• Explicitly targets Hindi as the output language, addressing a specific linguistic need.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.
• The synthetic nature of the data may introduce artifacts not present in real-world advisory exchanges.

Provenance

Source

Soketlabs via Hugging Face.

Collection Method

Synthetically generated by Google Gemini 2.5 Flash.

Freshness

Last updated 2026-01-16 11:58:24; freshness should be verified.