Skip to content

Autoregressive Language Modeling {Self-Sup}

Description

In autoregressive language modeling, which is used in models such as GPT, the model predicts the next word in a sentence given all the preceding words. It's trained to maximize the likelihood of a word given its previous words in the sentence.

Formula

The objective of an autoregressive language model is to maximize:

where w_i is the current word, w1,...., and wiโˆ’1 are the previous words, and ฮธ represents the model parameters.