LLM Performance History

Per-nightly end-to-end inference performance on the AMD Ryzen AI (Krackan Point, NPU2) benchmark runner. Each point is one nightly build, labeled by date; hover to see the commit SHA and verify status. A red ✗ marker flags a nightly whose correctness verify failed (the marker shape, not just its color, signals the failure).

Click a model in a legend to toggle its line.