Evidence-aware tool
Can my device run this model?
Enter measured values from your exact artifact and task. If measurements are missing, the tool returns a test plan instead of inventing compatibility.
Enter measurements to begin
The tool will separate observed facts from evidence still missing.
- Observed representative peak
- Arithmetic headroom vs physical RAM
A successful observed run applies only to the entered artifact, runtime, backend, device, OS, and task. It is not a universal device-support claim.
What a real compatibility record needs
Record device, chipset, OS build, app build, model artifact, runtime revision, execution backend, task input, output policy, memory timeline, sustained sequence, thermal context, failures, and fallback. Use the benchmark guide for the full method.