Post rl language models adapt to new tasks while keeping reasoning skills
Reasoning-Preserving Fine-Tuning of Post-RL LLMs with Null-Basis LoRA
Artificial Intelligence
Summary
Fine-tuning large language models after training them with reinforcement learning to boost reasoning skills is useful but can accidentally erase those skills when adapting to new tasks. The authors found that reasoning abilities are stored in special low-dimensional spaces inside the model, leaving room to learn new things without overwriting key reasoning parts. They created a method called Null-Basis LoRA that adapts the model while protecting its reasoning ability by updating only in safe areas. Their tests show this approach keeps the original reasoning strength and still learns new tasks well.
What this means in practice
- •For machine learning engineers: Safely adapt post-RL language models to new domains without losing reasoning performance using a parameter-efficient method.
- •For ai product developers: Develop conversational AI that retains advanced reasoning after being updated with new knowledge or behaviors.$Commercial implications: Enables maintaining reasoning capabilities in AI products after fine-tuning, improving reliability for commercial applications.
Authors
Wenzhi Fang, Nicholas Tzou, Lazar Valkov, Srinivas Chappidi
Abstract
Reinforcement learning (RL)-based post-training has become an effective approach for eliciting reasoning capabilities in large language models (LLMs). However, adapting post-RL models to new knowledge domains or behaviors through subsequent supervised fine-tuning (SFT) can severely overwrite these capabilities. Existing approaches mitigate such forgetting through experience replay, specialized initialization, or constrained optimization using gradient projection, but either provide limited preservation or incur substantial training overhead. Our analysis shows that reasoning activations concentrate in low-dimensional subspaces, leaving substantial null-space capacity for adaptation, and that the corresponding approximate null spaces can be reliably estimated from a modest number of examples. Motivated by these observations, we propose Null-Basis Low-Rank Adaptation (NB-LoRA), a parameter-efficient method for adapting post-RL LLMs while preserving their acquired reasoning ability. We formulate reasoning retention as a layer-wise hidden-state preservation constraint and construct a fixed approximate null basis from reasoning activations. LoRA updates are then reparameterized through this basis, enforcing the preservation constraint throughout fine-tuning. Extensive experiments across multiple RL-trained LLMs and diverse downstream tasks show that NB-LoRA matches standard LoRA in adaptation performance, maintains reasoning accuracy near pre-fine-tuning levels, and generalizes this preservation to held-out reasoning benchmarks.