What does Mid-training mean? Definition | AI Understanding

Definition

An intermediate training phase between pretraining and post-training, often used for capability or domain adjustments.

Related terms

Training steps applied after pretraining, such as instruction tuning, preference optimization, and safety tuning.

A collection of structured or unstructured examples used for training, validation, or testing.

A held-out dataset used to measure model quality after training.

Continuing training on domain-specific data to adapt a pre-trained model to a specific task.

Initial large-scale model training on broad data before downstream adaptation.

A model that scores outputs based on preference signals, often used in RLHF pipelines.

Learn more in our free guides

AI Training Lost in the Middle Effect Pseudo-Labeling and Self-Training ELECTRA Pretraining

See also

Watermarking Citations Self-Play Factuality RAG-Fusion GSM8K Multi-Query Retrieval HumanEval

Browse the full AI Glossary Explore all free AI guides