Base Model

A base language model is a model that has undergone large-scale pretraining but not instruction fine-tuning. It possesses language understanding and generation capabilities, yet it does not naturally follow human instructions. Such models are typically trained on broad text corpora to predict the next token, acquiring general linguistic knowledge and patterns. Because they lack alignment or supervised instruction tuning, their outputs may not reliably conform to task-oriented requests, and they usually serve as the starting point for subsequent fine-tuning or alignment stages.