Beyond Perplexity: A Lightweight Benchmark for Knowledge Retention in Supervised Fine-Tuning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Shabgahi, Soheil Zibakhsh, Aghazadeh, Pedram, Koushanfar, Farinaz
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914238055841792
author Shabgahi, Soheil Zibakhsh
Aghazadeh, Pedram
Koushanfar, Farinaz
author_facet Shabgahi, Soheil Zibakhsh
Aghazadeh, Pedram
Koushanfar, Farinaz
contents Supervised Fine-Tuning (SFT) is a standard approach for injecting domain knowledge into Large Language Models (LLMs). However, relying on validation perplexity to monitor training is often insufficient, as it confounds stylistic mimicry with genuine factual internalization. To address this, we introduce the Knowledge Retention (KR) Test , a lightweight, corpus-grounded evaluation framework designed to distinguish factual learning from linguistics. KR-Test utilizes automatically generated contrastive examples to measure likelihood preferences for correct versus incorrect continuations, requiring no instruction tuning or generative decoding. We validate the framework's integrity through a "blind vs. oracle" baseline analysis. Furthermore, we demonstrate the diagnostic capabilities of KR-Test by analyzing the training dynamics of Low-Rank Adaptation (LoRA). By exposing the fine-grained dissociation between linguistic convergence and knowledge retention, KR-Test enhances the interpretability of fine-tuning dynamics.
format Preprint
id arxiv_https___arxiv_org_abs_2601_03505
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Beyond Perplexity: A Lightweight Benchmark for Knowledge Retention in Supervised Fine-Tuning
Shabgahi, Soheil Zibakhsh
Aghazadeh, Pedram
Koushanfar, Farinaz
Computation and Language
Artificial Intelligence
Supervised Fine-Tuning (SFT) is a standard approach for injecting domain knowledge into Large Language Models (LLMs). However, relying on validation perplexity to monitor training is often insufficient, as it confounds stylistic mimicry with genuine factual internalization. To address this, we introduce the Knowledge Retention (KR) Test , a lightweight, corpus-grounded evaluation framework designed to distinguish factual learning from linguistics. KR-Test utilizes automatically generated contrastive examples to measure likelihood preferences for correct versus incorrect continuations, requiring no instruction tuning or generative decoding. We validate the framework's integrity through a "blind vs. oracle" baseline analysis. Furthermore, we demonstrate the diagnostic capabilities of KR-Test by analyzing the training dynamics of Low-Rank Adaptation (LoRA). By exposing the fine-grained dissociation between linguistic convergence and knowledge retention, KR-Test enhances the interpretability of fine-tuning dynamics.
title Beyond Perplexity: A Lightweight Benchmark for Knowledge Retention in Supervised Fine-Tuning
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2601.03505