Understand the deployment responsibilities that move into the app.
1. Runtime selection
Task APIs hide much of model management. A runtime such as LiteRT-LM gives more control but makes model compatibility, delivery and hardware behavior explicit engineering work.
2. Model artifacts
Weights, tokenizer and configuration form a compatible package. A file with the right extension is not necessarily supported by the chosen runtime. Pin and test the complete package.
3. Offline promise
A feature is only offline-ready after required assets are available. Explain first-download requirements and preserve a non-AI experience before setup completes.
Worked scenario
An app opens without internet after installation but has never downloaded the model. The offline feature is not ready yet.
Apply it
Compare a task-specific API and a custom local runtime for the same classification task.
Check your understanding
You can name the assets and runtime assumptions required for offline operation. Explain the decision and show evidence from your implementation or design. If you cannot demonstrate it yet, revisit the relevant section before continuing.