What are the three stages of model training?

• A. Initialization, training, deployment
• B. Data collection, model tuning, evaluation
• C. Pre-training, fine-tuning, and alignment
• D. Training, testing, validation

Answer: (C)

In modern large language model development, training is viewed as three connected stages: pre-training, fine-tuning, and alignment.

Pre-training builds broad language capability by learning from a vast, diverse text corpus. The model learns to predict the next word, which teaches it general grammar, facts, and reasoning patterns.

Fine-tuning then specializes that base model. It trains on carefully selected datasets that teach the model to follow instructions, perform specific tasks, or operate well in a particular domain.

Alignment comes after, guiding the model to behave in desirable and safe ways. This often involves techniques like reinforcement learning from human feedback to align outputs with human preferences and safety constraints.

Other options describe related parts of the process or post-training stages, but they don’t capture the typical three-stage progression used to build capable, task-ready, and responsibly behaving models.