Do language models plan ahead for future tokens?
Wu, Wilson, Morris, John X., Levine, Lionel
–arXiv.org Artificial Intelligence
Do transformers "think ahead" during inference at a given position? It is known transformers prepare information in the hidden states of the forward pass at $t$ that is then used in future forward passes $t+\tau$. We posit two explanations for this phenomenon: pre-caching, in which off-diagonal gradient terms present in training result in the model computing features at $t$ irrelevant to the present inference task but useful for the future, and breadcrumbs, in which features most relevant to time step $t$ are already the same as those that would most benefit inference at time $t+\tau$. We test these hypotheses by training language models without propagating gradients to past timesteps, a scheme we formalize as myopic training. In a synthetic data setting, we find clear evidence for pre-caching. In the autoregressive language modeling setting, our experiments are more suggestive of the breadcrumbs hypothesis.
arXiv.org Artificial Intelligence
Mar-31-2024
- Country:
- North America > United States
- Pennsylvania > Philadelphia County
- Philadelphia (0.04)
- New York > New York County
- New York City (0.04)
- Colorado > Boulder County
- Boulder (0.04)
- Pennsylvania > Philadelphia County
- Europe > Spain
- Catalonia > Barcelona Province > Barcelona (0.04)
- North America > United States
- Genre:
- Research Report > New Finding (0.46)
- Technology: