Proposed Kurdish term — under reviewزاراوەی پێشنیارکراو — لە ژێر پێداچوونەوەدایە
Definitionپێناسە
Testing an AI model, a or an on a fixed set of examples whose good answers are known, and scoring the results, so that you can compare versions instead of guessing. Public benchmarks are shared evaluations.
Left, an advert and a leaderboard. Right, your own test of 20 Kurdish questions: model A scores 17, model B 14, model C 11 (invented scores).چەپ، ڕیکلام و خشتەی پلەبەندی. ڕاست، تاقیکردنەوەی خۆت بە 20 پرسیاری کوردی: مۆدێلی A 17، مۆدێلی B 14، مۆدێلی C 11 (نمرەی دانراو).
Score every model on the same questions with known good answers.هەموو لەسەر هەمان پرسیار بە وەڵامی باشی زانراو نمرە بدە.CêIT
Exampleنموونە
Before changing her prompt, Dilan keeps 20 real student questions with good answers and checks both versions against them.
Other Kurdish renderings in useوەرگێڕانە کوردییەکانی تر
هەڵسەنگاندنی مۆدێلەکە، هەڵسەنگاندنی مۆدێلی
Note on the Kurdish termتێبینی لەبارەی زاراوە کوردییەکە
Proposed: هەڵسەنگاندن (evaluation, as in the registry's test-and-evaluation and evaluation-function) + مۆدێل. Kept apart from the hardware benchmark «پێوەری کارایی» (speed tests). Developers say "evals".