Chrome's On-Device AI Model Requires 20GB Storage Space

Chrome's move toward integrated on-device AI signals major browser vendors' commitment to local inference, but the 20GB storage requirement highlights critical hardware constraints practitioners must consider. While this represents Google's confidence in delivering useful models on-device, the significant storage footprint raises questions about feasibility on devices with limited storage capacity, particularly in developing regions and older hardware.

This development matters because it demonstrates the scale at which commercial-grade on-device models operate. The 20GB likely includes multiple model formats, quantisation variants, and safety infrastructure rather than a single model file. Understanding these requirements helps local LLM practitioners make informed decisions about model selection, quantisation strategies, and hardware targeting for their own deployments.

The ability to disable or control this feature is crucial, as it reflects user concerns about storage and privacy. This emphasises an ongoing challenge in local inference: balancing model capability with hardware constraints, a problem that quantisation, pruning, and efficient architecture design aim to solve.

Read the full article on Google News (Ubergizmo).

Source: Google News (Ubergizmo) · Relevance: 8/10