Resa: Transparent Reasoning Models via SAEs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Shangshang, Asilis, Julian, Akgül, Ömer Faruk, Bilgin, Enes Burak, Liu, Ollie, Fu, Deqing, Neiswanger, Willie
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913893453922304
author Wang, Shangshang
Asilis, Julian
Akgül, Ömer Faruk
Bilgin, Enes Burak
Liu, Ollie
Fu, Deqing
Neiswanger, Willie
author_facet Wang, Shangshang
Asilis, Julian
Akgül, Ömer Faruk
Bilgin, Enes Burak
Liu, Ollie
Fu, Deqing
Neiswanger, Willie
contents How cost-effectively can we elicit strong reasoning in language models by leveraging their underlying representations? We answer this question with Resa, a family of 1.5B reasoning models trained via a novel and efficient sparse autoencoder tuning (SAE-Tuning) procedure. This method first trains an SAE to capture reasoning abilities from a source model, and then uses the trained SAE to guide a standard supervised fine-tuning process to elicit such abilities in a target model, all using verified question-answer data without any reasoning traces. Notably, when applied to certain base models before further RL post-training, SAE-Tuning retains >97% of its RL-trained counterpart's reasoning performance while reducing training costs by >2000x to roughly \$1 and training time by >450x to around 20 minutes. Furthermore, when applied to lightly RL-trained models (e.g., within 1 hour on 2 GPUs), it enables reasoning performance such as 43.33% Pass@1 on AIME24 and 90% Pass@1 on AMC23 for only around \$1 additional cost. Surprisingly, the reasoning abilities extracted via SAEs are potentially both generalizable and modular. Generality means abilities extracted from one dataset still elevate performance on a larger and overlapping corpus. Modularity means abilities extracted from Qwen or Qwen-Math can be attached to the R1-Distill model at test time, without any retraining, and yield comparable gains. Extensive ablations validate these findings and all artifacts are fully open-sourced.
format Preprint
id arxiv_https___arxiv_org_abs_2506_09967
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Resa: Transparent Reasoning Models via SAEs
Wang, Shangshang
Asilis, Julian
Akgül, Ömer Faruk
Bilgin, Enes Burak
Liu, Ollie
Fu, Deqing
Neiswanger, Willie
Computation and Language
Artificial Intelligence
Machine Learning
How cost-effectively can we elicit strong reasoning in language models by leveraging their underlying representations? We answer this question with Resa, a family of 1.5B reasoning models trained via a novel and efficient sparse autoencoder tuning (SAE-Tuning) procedure. This method first trains an SAE to capture reasoning abilities from a source model, and then uses the trained SAE to guide a standard supervised fine-tuning process to elicit such abilities in a target model, all using verified question-answer data without any reasoning traces. Notably, when applied to certain base models before further RL post-training, SAE-Tuning retains >97% of its RL-trained counterpart's reasoning performance while reducing training costs by >2000x to roughly \$1 and training time by >450x to around 20 minutes. Furthermore, when applied to lightly RL-trained models (e.g., within 1 hour on 2 GPUs), it enables reasoning performance such as 43.33% Pass@1 on AIME24 and 90% Pass@1 on AMC23 for only around \$1 additional cost. Surprisingly, the reasoning abilities extracted via SAEs are potentially both generalizable and modular. Generality means abilities extracted from one dataset still elevate performance on a larger and overlapping corpus. Modularity means abilities extracted from Qwen or Qwen-Math can be attached to the R1-Distill model at test time, without any retraining, and yield comparable gains. Extensive ablations validate these findings and all artifacts are fully open-sourced.
title Resa: Transparent Reasoning Models via SAEs
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2506.09967