801: Merged LLMs Are Smaller And More Capable — with Arcee AI's Mark McQuade and Charles Goddard

Future of SLMs

The future of AI is leaning towards smaller, specialized language models (SLMs) that offer efficiency and cost-effectiveness. Charles Goddard explains that SLMs are more compact and cheaper to train, yet they can be equally or more powerful for specific tasks compared to larger foundational models 1. This shift allows for models to be run on edge devices, enhancing accessibility and reducing dependency on large-scale infrastructure. Jon Krohn highlights how Arcee AI's RC Spark model, with only 7 billion parameters, can outperform much larger models on certain benchmarks 2.

Smaller language models are the future, offering a more efficient and scalable solution for specific use cases.

--- Jon Krohn

This trend is driven by the need for models that are tailored to specific applications, providing a more targeted and efficient approach to AI deployment 3.

   

Cost-Effective AI Models

Smaller AI models are not only efficient but also significantly reduce costs, making them ideal for businesses with specific needs. Jon Krohn notes that running a 7 billion parameter model on a personal GPU can save up to 90% in costs compared to using closed-source models 4. This cost-effectiveness is crucial for enterprises that require models tailored to their data without the overhead of large-scale models. Charles Goddard emphasizes that these smaller models can be fine-tuned to outperform larger models in specific tasks, offering flexibility and power 5.

The ability to run a 7 billion parameter model efficiently on your own infrastructure is a game-changer for enterprises.

--- Charles Goddard

This approach allows companies to leverage AI without incurring the high costs associated with larger, less specialized models.