Sara Chrouf / Glossary
Glossary 語
Each term gets a small animated explainer and one honest opinion. Complexity explained without flattening; motion without spectacle.
Tokenization التقطيع 分
The model never reads your word; it reads how the vocabulary decided to cut it. My philosophy: the cut is a policy. In Kuwain we renegotiated it by expanding the vocabulary, and the whole model changed temperament.
Fertility الخصوبة 率
Tokens per word. Sounds cosmetic; prices context, latency, and money. My philosophy: fertility is the exchange rate of a language inside a model, and nobody voted on it.
Diacritization التشكيل 点
Restoring the short vowels the page omits. My philosophy: tashkeel is compressed grammar; teaching a model to put it back is free supervision, and Sadeed is the proof I co-signed.
Attention الانتباه 注
Every token asks every other token how much it matters right now. My philosophy: attention is a budget, in models and in people; what you attend to, you become fluent in.
Embedding التمثيل 図
Words become coordinates; meaning becomes neighborhood. My philosophy: an embedding space is a city plan for a language. Ask which neighborhoods Arabic got, and how far the dialects live from the center.
Benchmark contamination تلوث المعيار 汚
When test material leaks into training data, the exam grades memory, not ability. My philosophy: a score without a provenance protocol is a rumor with decimals, which is why our benchmarks publish theirs.