Direct answer
When the on-device AI task is interrupted, persist only the task state that can be safely resumed and restart everything else explicitly, without inferring additional meaning beyond what is stated.
Evidence to collect
The claim becomes reviewable only when the following evidence is attached to the same artifact and test run:
- Durable task schema
- Interruption fixtures
- Resume and restart outcomes
A runtime comparison is publishable only when artifact identity, build configuration, selected backend, task fixture, and failures are recorded.
Implementation workflow
Record and compare a durable task schema, interruption fixtures, and the resume and restart outcomes, keeping the evidence items aligned so their connections can be reviewed against the recorded observations.
- Freeze the product task and minimum device tier.
- Choose the runtime and supported artifact format.
- Pin artifact and runtime revisions.
- Wrap native state behind an application-owned interface.
- Test load, representative work, cancellation, close, and fallback.
- Publish only the behavior reproduced by the recorded configuration.
Failure patterns to prevent
- Persisting native handles
- Treating partial output as complete
Avoid persisting native handles and avoid treating partial output as complete; limit any release statement to wording grounded in the recorded evidence associated with the noted boundary identifiers.
Minimum reproducibility record
| Layer | Record |
|---|---|
| Device | Manufacturer, model, chipset, RAM class, operating-system build |
| Software | Application version and git commit |
| Model | Family, variant, revision, format, file length, hash, quantization |
| Runtime | Name, revision, requested backend, observed backend evidence |
| Workload | Fixture revision, input hash, prompt hash, output policy |
| Outcome | Completed, failed, cancelled, fallback, and privacy-safe diagnostics |
Release checklist
- ☐ The primary query is answered without an unsupported number.
- ☐ Every artifact and runtime is pinned.
- ☐ The representative task and failure policy are explicit.
- ☐ Lifecycle, cancellation, cleanup, and fallback are tested.
- ☐ User-content and network boundaries are documented.
- ☐ Result wording applies only to the recorded configuration.
- ☐ The page links to raw method or evidence when results are added.
Sources and related evidence
Chinese deployment and troubleshooting content is organized in the .奇连 AI 端侧专题.