Transferring Extreme Subword Style Using Ngram Model-Based Logit Scaling

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Messner, Craig, Lippincott, Tom
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908837898878976
author Messner, Craig
Lippincott, Tom
author_facet Messner, Craig
Lippincott, Tom
contents We present an ngram model-based logit scaling technique that effectively transfers extreme subword stylistic variation to large language models at inference time. We demonstrate its efficacy by tracking the perplexity of generated text with respect to the ngram interpolated and original versions of an evaluation model. Minimizing the former measure while the latter approaches the perplexity of a text produced by a target author or character lets us select a sufficient degree of adaptation while retaining fluency.
format Preprint
id arxiv_https___arxiv_org_abs_2503_08550
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Transferring Extreme Subword Style Using Ngram Model-Based Logit Scaling
Messner, Craig
Lippincott, Tom
Computation and Language
We present an ngram model-based logit scaling technique that effectively transfers extreme subword stylistic variation to large language models at inference time. We demonstrate its efficacy by tracking the perplexity of generated text with respect to the ngram interpolated and original versions of an evaluation model. Minimizing the former measure while the latter approaches the perplexity of a text produced by a target author or character lets us select a sufficient degree of adaptation while retaining fluency.
title Transferring Extreme Subword Style Using Ngram Model-Based Logit Scaling
topic Computation and Language
url https://arxiv.org/abs/2503.08550