Does the model predict the next whole word each time?
Not necessarily. Tokens can be shorter than words or include punctuation and other text units. The tokenizer determines which units the model predicts.
Explore This Question
Search for additional explanations using these tools:
More FAQs in How does an LLM predict the next token in a sequence?
The prediction depends on the broader available context, including earlier tokens and instructions. Different contexts can give the same word different roles and meanings.
A logit is an unnormalized score produced by the model. A normalization operation such as softmax converts the set of scores into probabilities.
Still have questions?
Browse all our FAQs or reach out to our support team