Skip to content
certsim
About

Glossary

Key terms, each linked to its source.

42 results · content in English

Hyperparameter tuning

Compares training configurations using a validation metric; hyperparameters are chosen outside the learned model weights.

Related: Feature engineering

Context window

The amount of input and output context a model can handle in a request, measured in tokens.

Temperature

A sampling control: lower values tend toward more predictable output; higher values increase variation.

Related: Top P · Top K

Top P

Nucleus sampling restricts token choices to a set reaching a cumulative probability threshold.

Related: Temperature · Top K

Few-shot prompting

Includes several examples of the desired task in the prompt without updating model weights.

Indirect prompt injection

Malicious instructions arrive through external content such as retrieved documents or tool output.

LLM-as-a-judge

Uses a language model and a scoring rubric to evaluate outputs; calibrate its judgments against human review.

Related: ROUGE · BLEU · BERTScore

Faithfulness

Whether an answer is supported by its retrieved context, distinct from whether the context itself is factually correct.

Related: Context relevance

Fairness

Addresses harmful differences in treatment or outcomes across people or groups.

Data residency

A requirement governing the geographic locations where data is stored or processed.