Global tech leaders are shifting their focus toward smaller, more efficient language models to save costs and energy.

Although more parameters mean better performance in artificial intelligence (AI) applications, the world‘s tech industry is now turning to more efficient and cost-effective models amid concerns about the massive cost and energy use associated with large language models (LLMs).

Microsoft Corp. announced on Tuesday that small language models (sLMs) with just 3.8 billion parameters, or approximately one-fifth of those for OpenAI’s GPT-3.5’s. More parameters mean better performance in artificial intelligence (AI) applications, but the world‘s tech industry is now turning to more efficient and cost-effective models as part of their effort to save cost and energy necessary to operate larger solutions.

Phi-3 mini, as a family of open models and the most capable, cost-effective SLM currently available, boasts its smallest and strongest capabilities. Microsoft said two additional models, Phi-3-small (7B) and Phi-3-medium (14B), will be available via the Azure AI model catalog.

Despite its diminutive size, Phi-3 mini packs a punch, excelling in language processing, learning, and coding tasks. Despite its compact nature, it can handle 128,000 tokens, which is roughly equivalent to 64 A4 pages. The newest model enables users to upload reports, make inquiries, and do more tasks while the smaller model also offers reduced latency and enhanced privacy compared to previous cloud-based alternatives.

Meta Platforms, Inc., which owns Facebook and Instagram, has introduced two versions of its smaller Large Language Model, Llama 3. These versions feature 8 billion and 70 billion parameters respectively, making them more accessible for anyone to download and use.

By Lee Sang-duk, Lee Duk-joo Han Yubin

[ⓒ Pulse by Maeil Business Newspaper & mk.co.kr, All rights reserved]