Step-wise Adaptive Integration of Supervised Fine-tuning and Reinforcement Learning for Task-Specific LLMs

• URL: http://arxiv.org/abs/2505.13026v2
• Date: Tue, 27 May 2025 07:16:10 GMT
• Title: Step-wise Adaptive Integration of Supervised Fine-tuning and Reinforcement Learning for Task-Specific LLMs
• Authors: Jack Chen, Fazhong Liu, Naruto Liu, Yuhan Luo, Erqu Qin, Harry Zheng, Tian Dong, Haojin Zhu, Yan Meng, Xiao Wang,
• Abstract summary: SASR is a step-wise adaptive hybrid training framework for large language models.<n>It unifies SFT and RL and dynamically balances the two throughout optimization.<n> Experimental results demonstrate that SASR outperforms SFT, RL, and static hybrid training methods.
• Score: 13.292104357930866
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) excel at mathematical reasoning and logical problem-solving. The current popular training paradigms primarily use supervised fine-tuning (SFT) and reinforcement learning (RL) to enhance the models' reasoning abilities. However, when using SFT or RL alone, there are respective challenges: SFT may suffer from overfitting, while RL is prone to mode collapse. The state-of-the-art methods have proposed hybrid training schemes. However, static switching faces challenges such as poor generalization across different tasks and high dependence on data quality. In response to these challenges, inspired by the curriculum learning-quiz mechanism in human reasoning cultivation, We propose SASR, a step-wise adaptive hybrid training framework that theoretically unifies SFT and RL and dynamically balances the two throughout optimization. SASR uses SFT for initial warm-up to establish basic reasoning skills, and then uses an adaptive dynamic adjustment algorithm based on gradient norm and divergence relative to the original distribution to seamlessly integrate SFT with the online RL method GRPO. By monitoring the training status of LLMs and adjusting the training process in sequence, SASR ensures a smooth transition between training schemes, maintaining core reasoning abilities while exploring different paths. Experimental results demonstrate that SASR outperforms SFT, RL, and static hybrid training methods.

Related papers

• Step-level Verifier-guided Hybrid Test-Time Scaling for Large Language Models [53.567732520169]
  Test-Time Scaling is a promising approach to progressively elicit the model's intelligence during inference.<n>In this paper, we focus on training-free TTS methods for reasoning.<n>We introduce a novel inference paradigm called Hybrid Test-Time Scaling.
  arXiv  Detail & Related papers  (2025-07-21T11:28:09Z)
• The Synergy Dilemma of Long-CoT SFT and RL: Investigating Post-Training Techniques for Reasoning VLMs [66.17068546293487]
  Large vision-language models (VLMs) increasingly adopt post-training techniques such as long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL) to elicit sophisticated reasoning.<n>We present a systematic investigation into the distinct roles and interplay of long-CoT SFT and RL across multiple multimodal reasoning benchmarks.<n>We find that SFT improves performance on difficult questions by in-depth, structured reasoning, but introduces verbosity and degrades performance on simpler ones.
  arXiv  Detail & Related papers  (2025-07-10T09:05:49Z)
• Reinforcement Fine-Tuning Enables MLLMs Learning Novel Tasks Stably [80.36077974826865]
  Post-training algorithms such as Supervised Fine-Tuning (SFT) and Reinforcement Fine-Tuning (RFT) are widely used to adapt multimodal large language models to downstream tasks.<n>We study the behavior of SFT and RFT on an open-source multimodal model, Qwen2.5-VL.<n>Our experiments reveal a sharp trade-off: SFT enables rapid task acquisition but leads to catastrophic forgetting, whereas RFT learns more slowly on novel tasks but maintains prior knowledge.
  arXiv  Detail & Related papers  (2025-06-30T04:15:01Z)
• SRFT: A Single-Stage Method with Supervised and Reinforcement Fine-Tuning for Reasoning [20.442971494407896]
  Large language models (LLMs) have achieved remarkable progress in reasoning tasks, yet the optimal integration of Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) remains a fundamental challenge.<n>We propose Supervised Reinforcement Fine-Tuning (SRFT), a single-stage method that unifies both fine-tuning paradigms through entropy-aware weighting mechanisms.<n>Extensive experiments show that SRFT achieves 59.1% average accuracy, outperforming zero-RL methods by 9.0% on five mathematical reasoning benchmarks and 10.9% on three out-of-distribution benchmarks.
  arXiv  Detail & Related papers  (2025-06-24T16:31:37Z)
• Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections [65.36449542323277]
  We present a unified theoretical framework bridgingSupervised Fine-Tuning (SFT) and preference learning in Large Language Model (LLM) post-training.<n>We propose a simple yet effective learning rate reduction approach that yields significant performance improvements.
  arXiv  Detail & Related papers  (2025-06-15T05:42:29Z)
• PATS: Process-Level Adaptive Thinking Mode Switching [53.53401063490537]
  Current large-language models (LLMs) typically adopt a fixed reasoning strategy, either simple or complex, for all questions, regardless of their difficulty.<n>This neglect of variation in task and reasoning process complexity leads to an imbalance between performance and efficiency.<n>Existing methods attempt to implement training-free fast-slow thinking system switching to handle problems of varying difficulty, but are limited by coarse-grained solution-level strategy adjustments.<n>We propose a novel reasoning paradigm: Process-Level Adaptive Thinking Mode Switching (PATS), which enables LLMs to dynamically adjust their reasoning strategy based on the difficulty of each step, optimizing the balance between
  arXiv  Detail & Related papers  (2025-05-25T17:58:50Z)
• Efficient Reinforcement Finetuning via Adaptive Curriculum Learning [24.52451100497884]
  Reinforcement finetuning (RFT) has shown great potential for enhancing the mathematical reasoning capabilities of large language models (LLMs)<n>AdaRFT dynamically adjusts the difficulty of training problems based on the model's recent reward signals.<n>AdaRFT reduces training time by up to 2x and improves accuracy by a considerable margin, offering a more scalable and effective RFT framework.
  arXiv  Detail & Related papers  (2025-04-07T21:31:31Z)
• Simplify RLHF as Reward-Weighted SFT: A Variational Method [34.222095430239555]
  Reinforcement Learning from Human Feedback (RLHF) is crucial for aligning Large Language Models (LLMs) with human values.<n>We propose a novel simplification of RLHF from the perspective of variational inference.<n>We transform the alignment objective into a reward-driven supervised fine-tuning form to obtain noticeable improvement on training stability and effectiveness.
  arXiv  Detail & Related papers  (2025-02-16T07:22:00Z)
• Q-SFT: Q-Learning for Language Models via Supervised Fine-Tuning [62.984693936073974]
  Value-based reinforcement learning can learn effective policies for a wide range of multi-turn problems.<n>Current value-based RL methods have proven particularly challenging to scale to the setting of large language models.<n>We propose a novel offline RL algorithm that addresses these drawbacks, casting Q-learning as a modified supervised fine-tuning problem.
  arXiv  Detail & Related papers  (2024-11-07T21:36:52Z)
• FADAS: Towards Federated Adaptive Asynchronous Optimization [56.09666452175333]
  Federated learning (FL) has emerged as a widely adopted training paradigm for privacy-preserving machine learning. This paper introduces federated adaptive asynchronous optimization, named FADAS, a novel method that incorporates asynchronous updates into adaptive federated optimization with provable guarantees. We rigorously establish the convergence rate of the proposed algorithms and empirical results demonstrate the superior performance of FADAS over other asynchronous FL baselines.
  arXiv  Detail & Related papers  (2024-07-25T20:02:57Z)
• Online Merging Optimizers for Boosting Rewards and Mitigating Tax in Alignment [47.682736928029996]
  Large Language Models (LLMs) are designed to align with human-centric values while preventing the degradation of abilities acquired through Pre-training and Supervised Fine-tuning (SFT) In this paper, we show that interpolating RLHF and SFT model parameters can adjust the trade-off between human preference and basic capabilities, thereby reducing the alignment tax. It significantly enhances alignment reward while mitigating alignment tax, achieving higher overall performance across 14 benchmarks.
  arXiv  Detail & Related papers  (2024-05-28T07:53:40Z)
• Training Neural Networks from Scratch with Parallel Low-Rank Adapters [46.764982726136054]
  We introduce LoRA-the-Explorer (LTE), a novel bi-level optimization algorithm designed to enable parallel training of multiple low-rank heads across computing nodes. Our approach includes extensive experimentation on vision transformers using various vision datasets, demonstrating that LTE is competitive with standard pre-training.
  arXiv  Detail & Related papers  (2024-02-26T18:55:13Z)
• ATE-SG: Alternate Through the Epochs Stochastic Gradient for Multi-Task Neural Networks [44.99833362998488]
  This paper introduces novel alternate training procedures for hard- parameter sharing Multi-Task Neural Networks (MTNNs)<n>The proposed alternate training method updates shared and task-specific weights alternately through the epochs, exploiting the multi-head architecture of the model.<n> Empirical experiments demonstrate enhanced training regularization and reduced computational demands.
  arXiv  Detail & Related papers  (2023-12-26T21:33:03Z)
• Optimization-Derived Learning with Essential Convergence Analysis of Training and Hyper-training [52.39882976848064]
  We design a Generalized Krasnoselskii-Mann (GKM) scheme based on fixed-point iterations as our fundamental ODL module. Under the GKM scheme, a Bilevel Meta Optimization (BMO) algorithmic framework is constructed to solve the optimal training and hyper-training variables together.
  arXiv  Detail & Related papers  (2022-06-16T01:50:25Z)

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.