Context

The Union Minister for Electronics and IT has said that the future of Artificial Intelligence (AI) will be shaped by smaller, efficient models rather than extremely large systems.

About Small Language Models (SLMs)

• SLMs are compact artificial intelligence systems built on simpler neural network architectures that can understand and generate natural language, similar to Large Language Models (LLMs), but at a smaller scale.
• SLMs typically operate with several million to around 30 billion parameters, whereas LLMs often have hundreds of billions or even trillions of parameters.
• Nearly 95% of global AI workloads are currently handled by SLMs due to their efficiency and practicality.
• Prominent examples include Llama, Mistral, Gemma, and Granite, which are widely used across industries.

Advantages of SLMs over LLMs

• Cost Efficiency: Smaller model size reduces computational and energy requirements, significantly lowering development and deployment costs.
• On-Device Deployment: Well-suited for edge and on-device applications where connectivity, memory, and power availability are limited.
• Democratisation of AI: Enables wider participation by startups, academia, and smaller organisations, fostering diversity in AI development.
• Operational Benefits: Easier monitoring and maintenance, improved data privacy and security, lower infrastructure needs, reduced latency, and better performance in domain-specific applications.

Limitations of SLMs

• Lower Accuracy: Compared to LLMs, SLMs generally offer reduced accuracy and versatility for highly complex or open-ended tasks.
• Narrower Knowledge Base: Training on smaller and more specialised datasets limits their general knowledge and adaptability.
• Functional Constraints: SLMs exhibit lower creativity, weaker large-scale data analysis capability, and reduced contextual depth compared to larger models.

Sharing is caring!