[Submitted on 7 Jul 2026]

View PDF HTML (experimental)

Abstract:The CoTFormer architecture formalizes Chain-of-Thought as a form of recurrent latent computation, preserving intermediate states as attendable representations to mimic explicit reasoning traces. In this work, we evaluate CoTFormer and its structural variants across perplexity and compute efficiency metrics. Furthermore, we extend evaluation to controlled algorithmic settings to determine whether this recurrent framework improves out-of-distribution generalisation on inductive reasoning tasks.

Submission history

From: Aras Kavuncu [view email]
[v1] Tue, 7 Jul 2026 19:30:40 UTC (2,410 KB)