---
title: README
colorFrom: gray
colorTo: yellow
sdk: static
pinned: false
license: mit
---

# Diff Interpretation Tuning
<p align="center">
  <a href="https://arxiv.org/abs/2510.05092"><strong>Paper</strong></a> ·
  <a href="https://colab.research.google.com/drive/12YD_9GRT-y_hFOBqXzyI4eN_lJGKiXwN?usp=sharing#forceEdit=true&sandboxMode=true"><strong>Colab Demo Notebook</strong></a> ·
  <a href="https://github.com/Aviously/diff-interpretation-tuning"><strong>Code</strong></a>
</p>

This organization hosts the weight diffs, DIT adapters, and finetuning data used in paper [Learning to Interpret Weight Differences in Language Models (Goel et al. 2025)](https://arxiv.org/abs/2510.05092).
The paper introduces *Diff Interpretation Tuning*, a method that trains a LoRA adapter than can be applied to a model to get it to describe its own finetuning induced modifications.

<p align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/637bc0902d2d9c4f248736e8/JoleMfukliT7gY-jZAGd2.png" alt="Teaser image for Diff Interpretation Tuning" width="650"/>
</p>