Saved in:
Bibliographic Details
Main Authors: Pavlova, Elizabeth, Koroliuk, Mariia, Viswanathan, Karthik, Tice, Cameron, Young, Edward James, Radmard, Puria
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.21376
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914416395550720
author Pavlova, Elizabeth
Koroliuk, Mariia
Viswanathan, Karthik
Tice, Cameron
Young, Edward James
Radmard, Puria
author_facet Pavlova, Elizabeth
Koroliuk, Mariia
Viswanathan, Karthik
Tice, Cameron
Young, Edward James
Radmard, Puria
contents We propose a new architectural change, and post-training pipeline, for making LLMs more verbose reasoners by teaching a model to truncate forward passes early. We augment an existing transformer architecture with an early-exit mechanism at intermediate layers and train the model to exit at shallower layers when the next token can be predicted without deep computation. After a calibration stage, we incentivise the model to exit as early as possible while maintaining task performance using reinforcement learning. We provide preliminary results to this effect for small reasoning models, showing that they learn to adaptively reduce computations across tokens. We predict that, applied at the right scale, our approach can minimise the amount of excess computation that reasoning models have at their disposal to perform non-myopic planning using their internal activations, reserving this only for difficult-to-predict tokens.
format Preprint
id arxiv_https___arxiv_org_abs_2603_21376
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle A transformer architecture alteration to incentivise externalised reasoning
Pavlova, Elizabeth
Koroliuk, Mariia
Viswanathan, Karthik
Tice, Cameron
Young, Edward James
Radmard, Puria
Artificial Intelligence
We propose a new architectural change, and post-training pipeline, for making LLMs more verbose reasoners by teaching a model to truncate forward passes early. We augment an existing transformer architecture with an early-exit mechanism at intermediate layers and train the model to exit at shallower layers when the next token can be predicted without deep computation. After a calibration stage, we incentivise the model to exit as early as possible while maintaining task performance using reinforcement learning. We provide preliminary results to this effect for small reasoning models, showing that they learn to adaptively reduce computations across tokens. We predict that, applied at the right scale, our approach can minimise the amount of excess computation that reasoning models have at their disposal to perform non-myopic planning using their internal activations, reserving this only for difficult-to-predict tokens.
title A transformer architecture alteration to incentivise externalised reasoning
topic Artificial Intelligence
url https://arxiv.org/abs/2603.21376