paper-with-me

Papers

Localizing RL-Induced Tool Use to a Single Crosscoder Feature

2026-06-25 · Andrii Shportko, Shubham Bhokare, Ahmed Zeyad A Alzahrani, Bowen Cheng, Gustavo Mercier, Jessica Hullman arxiv

Fine-tuning through RL reshapes the internal representations of language models to enable agentic behaviors such as tool use, yet the mechanistic basis of these changes remains poorly understood. While RL substantially improves structured tool-call generation, it is unclear which features emerge, which are preserved, and whether identified features can be leveraged for retraining-free behavioral control. In this work, we show that $\textit{Dedicated Feature Crosscoders (DFC)}$ isolate a compact set of RL-specific features that mediate tool-calling capability in $\texttt{Qwen2.5-3B}$. Across a $48$-crosscoder hyperparameter sweep, encode-decode reconstruction improves the RL model's tool correctness by $+31.1 \pm {9.7}$ pp and passively transfers tool-calling ability to the frozen base model by $+6.8 \pm 5.0$ pp which we call a $\textit{capability spillover}$. Our findings show that DFC partitioning concentrates RL-introduced capability into a minimal, steerable feature set that enables runtime behavioral control of agentic LLMs.

📄 PDF Abstract BibTeX arXiv:2606.26474

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Interactions Between Crosscoder Features: A Compact Proofs Perspective

2026-06-08 · Dmitry Manning-Coe, Thomas Read, Anna Soligo, Oliver Clive-Griffin 외 arxiv

Dictionary learning methods like Sparse Autoencoders (SAEs) and crosscoders attempt to explain a model by decomposing its activations into independent features. Interactions between features hence induce errors in the re…

Group Crosscoders for Mechanistic Analysis of Symmetry

2024-10-31 · Liv Gorton

We introduce group crosscoders, an extension of crosscoders that systematically discover and analyse symmetrical features in neural networks. While neural networks often develop equivariant representations without explic…

Dictionary Learning

fmxcoders: Factorized Masked Crosscoders for Cross-Layer Feature Discovery

2026-05-10 · Andreas D. Demou, Panagiotis Koromilas, James Oldfield, Yannis Panagakis 외 arxiv

Many features in pretrained Transformers span multiple layers: they emerge through stages of inference, persist in the residual stream, or are built jointly by parallel MLPs. Crosscoders (namely, sparse dictionaries trai…

Delta-Crosscoder: Robust Crosscoder Model Diffing in Narrow Fine-Tuning Regimes

2026-02-16 · Aly Kassem, Thomas Jiralerspong, Negar Rostamzadeh, Golnoosh Farnadi arxiv

Model diffing methods aim to identify how fine-tuning changes a model's internal representations. Crosscoders approach this by learning shared dictionaries of interpretable latent directions between base and fine-tuned m…

Cross-Architecture Model Diffing with Crosscoders: Unsupervised Discovery of Differences Between LLMs

2026-02-12 · Thomas Jiralerspong, Trenton Bricken arxiv

Model diffing, the process of comparing models' internal representations to identify their differences, is a promising approach for uncovering safety-critical behaviors in new models. However, its application has so far …