KissanAI/Dhenu2-In-Llama3.1-8B-Instruct is an 8 billion parameter agricultural language model built on the Llama3.1 architecture, developed by KissanAI. Optimized for India's diverse agricultural practices, it provides actionable insights and knowledgeable responses tailored to Indian farmers, policymakers, and agri-businesses. This model excels at developing comprehensive advisory applications for crop management, pest control, and resource optimization. It was trained on over 1.5 million instructions covering more than 4,000 agricultural topics, utilizing real and synthetic conversations, mobile extension service logs, and localized studies.
Loading preview...
Model Overview
KissanAI's Dhenu2-In-Llama3.1-8B-Instruct is an 8 billion parameter agricultural language model based on the Llama3.1 architecture. Released on October 24, 2024, this model is specifically designed to address the unique needs of the Indian agricultural sector.
Key Capabilities
- Agricultural Advisory: Provides real-time advice on crop management, pest control, and resource optimization.
- Contextual Understanding: Meticulously trained on India's diverse agricultural practices and localized studies.
- Data-Driven Insights: Leverages a comprehensive dataset including over 1.5 million instructions from real and synthetic conversations, covering more than 4,000 agricultural topics.
Training and Evaluation
The model underwent full fine-tuning combined with LoRA using NVIDIA A100 GPUs and DeepSpeed for distributed training. Flash attention mechanisms were implemented for efficiency. Evaluation involved both human assessment by agricultural experts for relevance and accuracy, and synthetic evaluation by other LLMs for consistency.
Intended Use Cases
- Building advisory applications for farmers.
- Supporting informed decision-making for policymakers and agri-businesses.
Limitations
Dhenu2 India 8B is highly specialized. Its performance may be less effective outside agricultural contexts, emphasizing the need for contextually relevant applications to maintain accuracy.