# Benchmark Summary ## Full Checkpoint | Suite | F1 | Examples/s | | --- | ---: | ---: | | Irish core | 0.9935 | 37.7313 | | Irish extended | 1.0000 | 22.0184 | | Demographic holdout v2 | 0.9988 | 58.4569 | | Gov contact policy v1 | 0.9836 | 22.1918 | | Gov chatbot red-team v2 | 0.9726 | 30.4434 | | Gov chatbot gap holdout v2 | 0.9969 | 49.7465 | | Multilingual PPSN overall | 0.9222 | 57.2856 | | Multilingual PPSN label-only | 0.9940 | — | ## ONNX q8 | Suite | F1 | Examples/s | | --- | ---: | ---: | | Irish core | 0.9935 | 61.4970 | | Irish extended | 1.0000 | 98.7502 | | Demographic holdout v2 | 0.9988 | 90.7807 | | Gov contact policy v1 | 0.9836 | 29.7205 | | Gov chatbot red-team v2 | 0.9726 | 13.0422 | | Gov chatbot gap holdout v2 | 0.9969 | 30.7289 | | Multilingual PPSN overall | 0.9222 | 33.5447 | | Multilingual PPSN label-only | 0.9940 | — | ## Comparison | Artifact | Core F1 | Demographic Holdout v2 F1 | Gov Contact Policy v1 F1 | Multilingual F1 | Core examples/s | | --- | ---: | ---: | ---: | ---: | ---: | | ContextPII q8 | 0.9935 | 0.9988 | 0.9836 | 0.9222 | 61.4970 | | GlobalPointer rc4 q8 | 1.0000 | 0.6938 | 0.7843 | 0.9333 | 221.5743 | | DiffMask rc6 q8 | 0.9733 | — | — | 0.9274 | 130.3415 |