Uncovering Uncertainty in Transformer Inference

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Brothers, Greyson, Mannering, Willa, Tien, Amber, Winder, John
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929619861504000
author Brothers, Greyson
Mannering, Willa
Tien, Amber
Winder, John
author_facet Brothers, Greyson
Mannering, Willa
Tien, Amber
Winder, John
contents We explore the Iterative Inference Hypothesis (IIH) within the context of transformer-based language models, aiming to understand how a model's latent representations are progressively refined and whether observable differences are present between correct and incorrect generations. Our findings provide empirical support for the IIH, showing that the nth token embedding in the residual stream follows a trajectory of decreasing loss. Additionally, we observe that the rate at which residual embeddings converge to a stable output representation reflects uncertainty in the token generation process. Finally, we introduce a method utilizing cross-entropy to detect this uncertainty and demonstrate its potential to distinguish between correct and incorrect token generations on a dataset of idioms.
format Preprint
id arxiv_https___arxiv_org_abs_2412_05768
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Uncovering Uncertainty in Transformer Inference
Brothers, Greyson
Mannering, Willa
Tien, Amber
Winder, John
Computation and Language
Artificial Intelligence
68T50 (Primary), 68T07 (Secondary)
F.2.2; I.2.7
We explore the Iterative Inference Hypothesis (IIH) within the context of transformer-based language models, aiming to understand how a model's latent representations are progressively refined and whether observable differences are present between correct and incorrect generations. Our findings provide empirical support for the IIH, showing that the nth token embedding in the residual stream follows a trajectory of decreasing loss. Additionally, we observe that the rate at which residual embeddings converge to a stable output representation reflects uncertainty in the token generation process. Finally, we introduce a method utilizing cross-entropy to detect this uncertainty and demonstrate its potential to distinguish between correct and incorrect token generations on a dataset of idioms.
title Uncovering Uncertainty in Transformer Inference
topic Computation and Language
Artificial Intelligence
68T50 (Primary), 68T07 (Secondary)
F.2.2; I.2.7
url https://arxiv.org/abs/2412.05768