Local Models
AI models that are deployed and run locally on a user's hardware, as opposed to cloud-based models.
Learn
When to use it
Use local models when data privacy concerns or network latency make cloud-based solutions impractical. Local models allow for direct control over data and processing, enabling tasks like offline language translation or on-device image recognition without sending data to external servers.
Quick example
In Apple's Core ML, developers can deploy models directly on iOS devices for tasks like image classification or natural language processing. Here, Core ML acts as the framework that facilitates running these local models, ensuring that the processing happens entirely on the user's device without relying on cloud services.
input data → Core ML (local model) → output
Ecosystem
Local models often interact with device-specific APIs and hardware accelerators to optimize performance. They sit alongside other on-device processing components, and the model's architecture must be compatible with the local environment.
input data → local model → device APIs/hardware → output
Misconceptions
| Misconception | Rebuttal |
|---|---|
| Local models are always faster | They can be slower if not optimized for the hardware |
| They require no internet | Some features might still need online access for updates |
| Local models are less powerful | They can be as powerful, depending on the hardware |
Trade-offs
- Data privacy — limited by device storage and processing power
- Latency reduction — requires careful optimization for hardware
- Offline capability — may lack real-time updates or cloud features