
Developed by	KissanAI (https://kissan.ai)
Base model	`Qwen/Qwen3-4B`
Architecture	Qwen3 decoder-only causal-LM, 32 k context
Fine-tuning method	Supervised fine-tuning (SFT) via llama-factory
Languages	Primarily English + technical Indian-agricultural vocabulary
License	Apache 2.0 (inherits from base model)

Intended Use

Primary purpose
Assist farmers, agronomists and ag-tech developers with Climate-Resilient and Sustainable Agriculture (CRSA) recommendations tailored to Indian conditions (e.g., APCNF/organic practices, climate-smart cropping, pest IPM, soil/nutrient management).

Direct use examples

Decision-support micro-service answering agronomic queries.
Content generation for ag-extension material.

Out-of-scope uses

Any medical, legal, or financial advice.
Real-time critical decision making without human validation.
Disinformation, hateful or extremist content.

Training Data

Dataset	Size	Notes
`KissanAI/Thinking-climate-100k`	101 k multi-turn dialogues on climate-smart ag topics with thinking tags

The dataset is synthetic/aligned through “chain-of-thought + answer” format that explicitly separates the model’s private reasoning (<think> … </think>) from the final answer, reducing chain-of-thought leakage at inference time.

Bias, Risks & Limitations

May embed agronomic bias toward Indian Natural Farming practices (APCNF).
Climate data cited is static (2024) – cross-check against latest IMD advisories.
Still prone to LLM hallucinations; always validate high-stakes advice with qualified professionals.

Citation

@misc{KissanAI2025ThinkingDhenu1,
  title        = {ThinkingDhenu1-CRSA-India-preview},
  author       = {KissanAI},
  howpublished = {\url{https://huggingface.co/KissanAI/ThinkingDhenu1-CRSA-India-preview}},
  year         = {2025},
  note         = {Fine-tuned from Qwen3-4B on Indian climate-smart agriculture data.}
}

Contact

Contact Questions or feedback? Open an issue on the model repo.

Next steps you might consider

Add private eval numbers.
Specify dataset licences.