Arabic AI Evaluation Toolkit

Compare Arabic model output with a reference using conservative normalization and transparent token-level metrics.

Reference text

Model prediction

Normalize Alef variants Normalize Yeh variants Normalize Teh Marbuta

Normalized reference

Normalized prediction

Exact match—

Precision—

Recall—

Token F1—