Causal Language Modeling

Same as autoregressive language modeling: predict the next token given previous tokens. "Causal" means the prediction doesn't depend on future context (which would be physically unrealistic for generation).