Weightless Fine-Tuning: Personalizing LLMs via Logit-Space Transport

2026-08-11

AI Summary

This work addresses the high cost of traditional supervised fine-tuning in personalized settings, where separate model weights must be stored and optimized for each user. The authors propose a training-free, decoding-time approach that simulates fine-tuning effects by transferring supervised residuals in logit space without modifying model parameters. Their method is the first to approximate the output distribution of supervised fine-tuning without any weight updates, leveraging a context-aware transfer operator constructed via dropout-induced cross-covariance estimation and replacing gradient-based updates with an inter-context perturbation propagation mechanism. Evaluated on three LaMP benchmarks, the approach achieves state-of-the-art average performance while incurring less than 7% of the computational overhead of fine-tuning, with logit shifts exhibiting a cosine similarity of 0.875 to those produced by full fine-tuning.

Abstract

Supervised fine-tuning (SFT) is a standard approach for adapting LLMs to a target distribution, but in settings such as personalization, where each author requires separate weight access, optimization, storage, and retraining, its costs become prohibitive. We propose Weightless Fine-Tuning (WFT), a training-free decoding-time method that approximates the distributional effect of SFT without weight updates. WFT computes supervised residuals on an author's training sequence and transports them to the current prompt through a cross-prefix transport operator estimated from dropout-induced cross-covariance. The operator captures how a perturbation at one context propagates to predictions at another, replacing gradient-based parameter updates with logit-space corrections. On three LaMP personalization benchmarks, WFT achieves the best average performance across datasets, matches or exceeds SFT on individual tasks, and outperforms other lightweight baselines on average. In a budget-controlled comparison, WFT approaches SFT performance using less than 7% of the effective computation. Logit-level analysis shows a cosine similarity of 0.875 between the logit shifts induced by WFT and SFT over 95% of the next-token probability mass, suggesting that WFT captures the distributional effect of supervised adaptation without modifying model weights.

Problem

• personalization
• supervised fine-tuning
• large language models
• weight adaptation
• computational cost

Innovation

• Weightless Fine-Tuning
• Logit-Space Transport
• Personalization
• Supervised Fine-Tuning
• Decoding-Time Adaptation