Cropin launches micro language model to drive climate-smart agriculture

Bengaluru-based agritech firm Cropin Technology announced the launch of an open-source micro language model built on Mistral’s foundation model on Tuesday. Dubbed ‘Aksara’, the new solution is designed to scale up major crops in South Asia, aiming for sustainable and energy-efficient farming practices while offering versatile language support.

Speaking to TechCircle, Krishna Kumar, co-founder & CEO of Cropin, said that the goal is to democratise access to digital technologies and modernise agricultural processes. Cropin aims to empower agricultural stakeholders, developers, and researchers to address global challenges like food security, climate change, resource conservation (water and soil), and regenerative agriculture practices by providing access to contextual, factual, and actionable information.

The first version will cover nine crops, including paddy, wheat, maize, sorghum, barley, cotton, sugarcane, soybean, and millet, across 5 countries in the Indian subcontinent. Kumar explained that these food crops collectively account for a substantial portion of the world's food requirements and are staple foods for the population in the global south.

Praveen Pankajakshan, Vice President of Data Science & AI at Cropin, explained the technology, stating that ‘akṣara’ is a frugal and scalable µ-LM built and fine-tuned on top of the Mistral-7B-v0.2 model, developed by Cropin and hosted on Hugging Face. Cropin compressed ‘akṣara’ into 4-bit from 16-bit using QLoRA (Quantisation and Low-Rank Adapters) to reduce the environmental impact of running large language models (LLMs).

Pankajakshan mentioned that the model outperforms GPT-4 Turbo by almost 40% on randomly selected test datasets, as measured by the ROUGE scoring algorithm. The model ensures that responses are factually relevant and brief while minimising compute and storage resource requirements.

The model was fine-tuned with over 5,000 high-quality question-response pairs specific to agriculture and more than 160k tokens in context. These numbers are expected to increase as more crops, geographic locations, and use cases are added. Pankajakshan emphasized that the model remains faithful to the question by using techniques like retrieval augmented generation (RAG) through cross-referencing expert knowledge bases.

In today's complex agricultural landscape, farmers and others involved in agriculture struggle to meet the world’s increasing demand for sufficient, safe, and nutritious food, Kumar added. Climate change is disrupting conventional agricultural practices, making existing knowledge impractical. Factors like irregular or extreme rainfall, unpredictable heat waves, and increased pest and disease attacks affect farmers' practices, reducing agricultural yield, productivity, and profitability.

Cropin aims to bridge this gap with ‘akṣara’ by leveraging GenAI to provide insights into modern farming practices, accurate information, and farm advisories. For example, it can suggest inputs to use for crops like rice or maize under specific agro-climatic conditions or provide climate-smart agri advisories.

Kumar said that this open-source initiative aims to support agronomists, agri-scientists, field staff, and extension workers, gradually extending services to farmers in multiple languages to provide local language support. The goal is to transform global food systems by equipping industry think tanks and researchers with the best decision-making tools and information, disseminating knowledge at the grassroots level.

This initiative follows the development of an AI-powered Food Security Decision Intelligence solution in partnership with Amazon Web Services (AWS) last month. Cropin is also seeking collaborations with industry players and academia to accelerate its vision of a more digitally empowered, data and AI-driven agriculture landscape worldwide, with ‘Akṣara’ marking a significant step in this journey.

Out of the world's 600 million farms, five out of six are small, covering less than two hectares each. Despite operating on only about 12% of agricultural land, these small farms produce roughly 35% of the world's food. The goal is to enable AI investments to have a significant impact and make AI accessible to everyone in the ecosystem, including academia, development agencies, governments, and agricultural enterprises, to empower farmers worldwide, starting with smallholders in the global south.

Cropin was founded in 2010 by CEO Krishna Kumar, with an initial focus on selling digital tools directly to farmers. Within a year, they gained traction with the Cropin Cloud Platform, which helps the food industry digitise its supply chains. Since then, the company has been leveraging Artificial Intelligence, satellite imagery, and a crop knowledge graph to democratise agricultural intelligence for farmers without the need for expensive field hardware.

In January 2023, Cropin raised ₹113 crore ($14 million) in funding from Google and the Japanese conglomerate JSR Corporation, along with existing investors ABC Impact and Chiratae Ventures. Kumar said that over the last 12 months, the company utilised the funds raised in the Series D round to expand Cropin Cloud, transforming it into an intelligent agriculture cloud platform to meet the increasing demand for digitisation and predictive intelligence in the global agriculture sector. The funds were also used to enhance go-to-market efforts and develop next-generation predictive intelligence solutions through the Cropin AI Labs.