AI Dictionary › AI Fundamentals

Logits

Logit

Logits are the raw numerical scores a model produces before turning them into probabilities. They are unconstrained values, positive or negative, expressing how much the model favors each option: a high logit signals strong preference, a low one the opposite. They're like the raw scores a judge assigns before normalizing them into a final verdict: informative, but not yet readable as probabilities. To convert them into comparable percentages that sum to one, the softmax function is applied. In a language model there's one logit for every possible next token in the vocabulary.

Definition

Working directly with logits is useful in many practical settings: it lets you measure model confidence, apply filtering rules, compute the loss function during training in a numerically stable way, and control generation. Understanding them helps distinguish what the model 'thinks' internally from the final probability shown.

Related terms

• Softmax
• Loss Function
• Temperature
• Token
• Inference

More in AI Fundamentals

• LLM (Large Language Model)
• Token
• Context Window
• Temperature
• AI Hallucination
• Fine-tuning
• RAG (Retrieval Augmented Generation)
• Embedding

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.