soketlabs/sarthi-agri-v1

Sarthi-agri-v1: Domain-Adapted Agricultural Advisory Model

soketlabs/sarthi-agri-v1 is a 27 billion parameter research-preview large language model developed by Soket AI Labs, fine-tuned from google/gemma-3-27b-it. This model is uniquely adapted for generating structured agricultural advisories, leveraging its base model's strong language understanding and instruction-following capabilities.

Key Capabilities

• Agricultural Reasoning: Optimized for taxonomy-driven reasoning across agronomic parameters such as crop type, growth stage, climate, soil properties, and regional conditions.
• Structured Analytical Thinking: Induces "thinking token generation" to improve structured reasoning from taxonomy-based agricultural inputs, leading to deeper analytical responses.
• Multilingual Advisory: Capable of generating farmer-friendly advisories, with a primary focus on Hindi, alongside English and Hinglish.
• Controlled Output: Utilizes explicit control tokens (<unused0>, <unused1>) to enforce structured reasoning and deterministic output formatting, aiding traceability and UI streaming.

Good for

• Agricultural Intelligence: Professionals, researchers, and developers working on agricultural intelligence and decision-support systems.
• Agronomy Workflows: Automating and assisting in agronomy workflows and rural advisory automation.
• Decision Support Systems: Providing structured, consistent, and accurate agricultural advisories based on detailed inputs.

Sarthi-agri-v1 was fine-tuned using LoRA on the soketlabs/Sarathi-AgriData dataset, which includes curated agricultural taxonomy, climate-conditioned crop advisory samples, and multi-language supervision. It is important to note that the model's outputs are informational and advisory only and do not replace certified agricultural experts.