A pretrained transformer shows similar surprisal-decay patterns for linear and exponentially spaced context sampling, suggesting inference-time computations can be separated from model architecture.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
q-bio.NC 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Dissociating model architectures from inference computations
A pretrained transformer shows similar surprisal-decay patterns for linear and exponentially spaced context sampling, suggesting inference-time computations can be separated from model architecture.