54 / 66 · 09 Mobile & Cross-Platform Testing · On-Device ML Model Testing← prev⊞ allnext →☰ Read as one page
10.2What to Test for On-Device ML
| Test Area | What to Verify | Example Assertion |
|---|---|---|
| Model loading | Model initializes without crash on target devices | Load time < 2s on Tier 1 devices |
| Inference latency | Prediction completes within acceptable time | p95 < 100ms on mid-range devices |
| Accuracy on device | Model produces same results as server-side | Agreement > 99% on validation set |
| Memory footprint | Model does not cause OOM on constrained devices | Peak memory < 100MB |
| Battery impact | Continuous inference does not drain battery | < 5% per hour of active use |
| Fallback behavior | App works when model fails to load | Graceful degradation to server-side API |
| Model update | OTA model updates apply correctly | New model version loads after app restart |
| Concurrent inference | Multiple inference requests do not crash | 10 concurrent requests complete |