Implicit Bias of Next-Token Prediction

arXiv:2402.18551v1 Announce Type: new
Abstract: Next-token prediction (NTP), the go-to training paradigm in training large language models, involves predicting the next token in a sequence. Departing from traditional one-hot classification, in NTP, multiple tokens with varying frequencies follow each given context. This work frames NTP training as cross-entropy minimization over distinct contexts, each associated with a sparse empirical probability vector across a finite vocabulary. It then addresses the following question: do gradient-based optimizers exhibit a bias towards solutions with specific …

abstract arxiv bias classification context cross-entropy cs.cl cs.lg entropy hot language language models large language large language models multiple next paradigm prediction stat.ml token tokens training type work