Reinforcement Learning Fine-Tunes a Sparse Subnetwork in Large Language Models

• URL: http://arxiv.org/abs/2507.17107v2
• Date: Mon, 28 Jul 2025 18:29:13 GMT
• Title: Reinforcement Learning Fine-Tunes a Sparse Subnetwork in Large Language Models
• Authors: Andrii Balashov,
• Abstract summary: It is often assumed thatReinforcement learning (RL) fine-tuning requires updating most of a model's parameters.<n>We call this phenomenon RL-induced parameter update sparsity.<n>We show that fine-tuning only this sparse subnetwork recovers full model performance and yields parameters nearly identical to the fully fine-tuned model.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Reinforcement learning (RL) is a key post-pretraining step for aligning large language models (LLMs) with complex tasks and human preferences. While it is often assumed that RL fine-tuning requires updating most of a model's parameters, we challenge this assumption with a surprising finding: RL fine-tuning consistently modifies only a small subnetwork (typically 5-30% of weights), leaving most parameters unchanged. We call this phenomenon RL-induced parameter update sparsity. It arises naturally, without any sparsity constraints or parameter-efficient tuning, and appears across multiple RL algorithms (e.g., PPO, DPO, SimPO, PRIME) and model families (e.g., OpenAI, Meta, and open-source LLMs). Moreover, the subnetworks updated by RL show substantial overlap across different seeds, datasets, and algorithms-far exceeding chance-suggesting a partially transferable structure in the pretrained model. We show that fine-tuning only this sparse subnetwork recovers full model performance and yields parameters nearly identical to the fully fine-tuned model. Our analysis suggests this sparsity emerges because RL operates near the model's original distribution, requiring only targeted changes. KL penalties, gradient clipping, and on-policy dynamics have limited effect on the sparsity pattern. These findings shed new light on how RL adapts models: not by shifting all weights, but by focusing training on a small, consistently updated subnetwork. This insight enables more efficient RL methods and reframes sparsity through the lens of the lottery ticket hypothesis.

Related papers

• RAST: Reasoning Activation in LLMs via Small-model Transfer [33.32587030836428]
  Reinforcement learning (RL) has become a powerful approach for improving the reasoning capabilities of large language models (LLMs)<n>Applying RL at scale remains intimidatingly resource-intensive, requiring multiple model copies and extensive GPU workloads.<n>We propose RAST, a simple yet effective method that transfers reasoning behaviors by injecting RL-induced probability adjustments from a small RL-trained model into larger models.
  arXiv  Detail & Related papers  (2025-05-30T17:57:08Z)
• Scaling Offline RL via Efficient and Expressive Shortcut Models [13.050231036248338]
  offline reinforcement learning (RL) remains challenging due to the iterative nature of their noise sampling processes.<n>We introduce Scalable Offline Reinforcement Learning (SORL), a new offline RL algorithm that leverages shortcut models to scale both training and inference.<n>We demonstrate that SORL achieves strong performance across a range of offline RL tasks and exhibits positive scaling behavior with increased test-time compute.
  arXiv  Detail & Related papers  (2025-05-28T20:59:22Z)
• Reinforcement Learning Finetunes Small Subnetworks in Large Language Models [27.55599230411277]
  Reinforcement learning (RL) yields substantial improvements in large language models downstream task performance and alignment with human values.<n>Surprisingly, such large gains result from updating only a small subnetwork comprising just 5 percent to 30 percent of the parameters.<n>We refer to this phenomenon as parameter update sparsity induced by RL.
  arXiv  Detail & Related papers  (2025-05-16T21:42:28Z)
• Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining [74.83412846804977]
  Reinforcement learning (RL)-based fine-tuning has become a crucial step in post-training language models.<n>We present a systematic end-to-end study of RL fine-tuning for mathematical reasoning by training models entirely from scratch.
  arXiv  Detail & Related papers  (2025-04-10T17:15:53Z)
• Scaling Offline Model-Based RL via Jointly-Optimized World-Action Model Pretraining [49.730897226510095]
  We introduce JOWA: Jointly-Reinforced World-Action model, an offline model-based RL agent pretrained on Atari games with 6 billion tokens data.<n>Our largest agent, with 150 million parameters, 78.9% human-level performance on pretrained games using only 10% subsampled offline data, outperforming existing state-of-the-art large-scale offline RL baselines by 31.6% on averange.
  arXiv  Detail & Related papers  (2024-10-01T10:25:03Z)
• REBEL: Reinforcement Learning via Regressing Relative Rewards [59.68420022466047]
  We propose REBEL, a minimalist RL algorithm for the era of generative models.<n>In theory, we prove that fundamental RL algorithms like Natural Policy Gradient can be seen as variants of REBEL.<n>We find that REBEL provides a unified approach to language modeling and image generation with stronger or similar performance as PPO and DPO.
  arXiv  Detail & Related papers  (2024-04-25T17:20:45Z)
• Recurrent Model-Free RL is a Strong Baseline for Many POMDPs [73.39666827525782]
  Many problems in RL, such as meta RL, robust RL, and generalization in RL, can be cast as POMDPs. In theory, simply augmenting model-free RL with memory, such as recurrent neural networks, provides a general approach to solving all types of POMDPs. Prior work has found that such recurrent model-free RL methods tend to perform worse than more specialized algorithms that are designed for specific types of POMDPs.
  arXiv  Detail & Related papers  (2021-10-11T07:09:14Z)
• Reinforcement Learning as One Big Sequence Modeling Problem [84.84564880157149]
  Reinforcement learning (RL) is typically concerned with estimating single-step policies or single-step models. We view RL as a sequence modeling problem, with the goal being to predict a sequence of actions that leads to a sequence of high rewards.
  arXiv  Detail & Related papers  (2021-06-03T17:58:51Z)
• MOPO: Model-based Offline Policy Optimization [183.6449600580806]
  offline reinforcement learning (RL) refers to the problem of learning policies entirely from a large batch of previously collected data. We show that an existing model-based RL algorithm already produces significant gains in the offline setting. We propose to modify the existing model-based RL methods by applying them with rewards artificially penalized by the uncertainty of the dynamics.
  arXiv  Detail & Related papers  (2020-05-27T08:46:41Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.