SkillFactory: Self-Distillation For Learning Cognitive Behaviors

TL;DR

SkillFactory introduces a supervised fine-tuning method that uses self-generated samples to prime models for cognitive skills, enhancing their ability to generalize and be robust after reinforcement learning.

Contribution

The paper presents a novel fine-tuning approach that leverages self-sampled data to enable models to acquire cognitive skills before reinforcement learning.

Findings

SkillFactory improves model generalization to harder tasks post-RL.

Models trained with SkillFactory utilize cognitive skills effectively.

SkillFactory models show increased robustness to out-of-domain regressions.

Abstract

Reasoning models leveraging long chains of thought employ various cognitive skills, such as verification of their answers, backtracking, retrying by an alternate method, and more. Previous work has shown that when a base language model exhibits these skills, training that model further with reinforcement learning (RL) can learn to leverage them. How can we get models to leverage skills that aren't exhibited by base models? Our work, SkillFactory, is a method for fine-tuning models to roughly learn these skills during a supervised fine-tuning (SFT) stage prior to RL. Our approach does not rely on distillation from a stronger model, but instead uses samples from the model itself, rearranged to provide training data in the format of those skills. These "silver" SFT traces may be imperfect, but are nevertheless effective for priming a model to acquire skills during RL. Our evaluation shows that (1) starting from SkillFactory SFT initialization helps a model to generalize to harder variants of a task post-RL, despite lower performance pre-RL;(2) cognitive skills are indeed used by the model; (3) RLed SkillFactory models are more robust to regression on out-of-domain tasks than RLed base models. Our work suggests that inductive biases learned prior to RL help models learn robust cognitive skill use.