Cropin Technology unveiled 'akṣara', the first micro language model for climate-smart agriculture

'Akṣara', the first micro language model (µ-LM) designed especially for climate-smart agriculture, was recently unveiled by Bengaluru-based agtech company Cropin Technologies. Built around the basic model of Mistral, the tool is made available to the public under the Apache 2.0 License. 

"Akṣara" aims to revolutionize farming methods by enabling researchers, developers, and other relevant parties to produce scalable artificial intelligence solutions. In the Global South in particular, these solutions seek to promote regenerative agriculture methods, improve food security, and adapt to climatic problems.

The 'akṣara' micro model is initially designed to manage data pertaining to nine important crops, encompassing five countries in the Indian subcontinent, such as paddy, wheat, maize, and soybeans. To reduce its environmental impact, it uses Quantization and Low-Rank Adapters (QLoRA) to compress data from 16 bits to 4 bits. Because of the large reduction in processing and storage resource requirements, this approach is energy- and cost-efficient. 

Cropin highlights the significance of ethical AI and makes sure that 'akṣara' complies with responsible AI guidelines by taking into account input from conversations with Google's Responsible AI team. The methodology encourages sustainable measures like soil and water conservation and is economical, scalable, and bias-free. By making AI technologies more accessible to a wider range of users—from agronomists to small-town farmers—this open-source project seeks to democratize access to these tools.

Although 'akṣara' only supports English at the moment, plans are in place to expand its functionality to several languages and different crops in the upcoming six months.