Cerence Inc. introduced CaLLM?? Edge, its pioneering, automotive-grade, embedded small language model (SLM). This new SLM is available within the existing Cerence solutions portfolio and will power its next-generation AI assistant platform, enabling an intelligent, seamless user experience regardless of connectivity.

Developed and optimized in partnership with Microsoft, CaLLM Edge is available directly to Cerence's automaker customers as well as in the Microsoft Azure AI model catalog. CaLLM (Cerence Automotive Large Language Model) Edge is fine-tuned on Microsoft's Phi-3 family of small language models, using Cerence's extensive automotive dataset to deliver highly specialized AI that can handle a variety of automotive use cases. With 3.8 billion parameters, 4k context size, and 4-bit quantization, this model fits comfortably when embedded in the automotive headunit.

Its core capabilities include implicit and explicit car control commands (for example, temperature, windows and doors, seat position) and point-of-interest search and navigation, as well as conversational interaction like, "What's the most popular movie ever shot in Hollywood?" followed by "Can you tell me more about the plot?" Compatible with major automotive platforms, CaLLM Edge is availability in both embedded-only deployments, meaning it can function independently without any connectivity, as well as hybrid or cloud-first deployments in which the SLM serves as one method of answering queries and as backup when connectivity is lost. For users, this means always-on access to key generative AI-style features and information, even when not connected to the cloud, as well as improved data privacy, with data staying on board in the car rather than being sent to the cloud. For automakers, CaLLM Edge delivers not only improved assistant performance, but also cost efficiency - by leveraging a fully embedded SLM model, OEMs can keep costs under control while still delivering a generative AI-based experience for their drivers.