Semformer: Transformer Language Models with Semantic Planning
This addresses shortcut learning in language models, which can compromise accuracy, though it appears incremental as it builds on existing Transformer architectures.
The paper tackles the problem of shortcut learning in Transformer language models by introducing Semformer, which incorporates semantic planning tokens to guide response generation, achieving near-perfect performance in graph path-finding and showing efficacy in perplexity and downstream tasks.
Next-token prediction serves as the dominant component in current neural language models. During the training phase, the model employs teacher forcing, which predicts tokens based on all preceding ground truth tokens. However, this approach has been found to create shortcuts, utilizing the revealed prefix to spuriously fit future tokens, potentially compromising the accuracy of the next-token predictor. In this paper, we introduce Semformer, a novel method of training a Transformer language model that explicitly models the semantic planning of response. Specifically, we incorporate a sequence of planning tokens into the prefix, guiding the planning token representations to predict the latent semantic representations of the response, which are induced by an autoencoder. In a minimal planning task (i.e., graph path-finding), our model exhibits near-perfect performance and effectively mitigates shortcut learning, a feat that standard training methods and baseline models have been unable to accomplish. Furthermore, we pretrain Semformer from scratch with 125M parameters, demonstrating its efficacy through measures of perplexity, in-context learning, and fine-tuning on summarization tasks.