Evolving LLMs' Self-Refinement Capability via Synergistic Training-Inference Optimization

• URL: http://arxiv.org/abs/2502.05605v6
• Date: Sun, 26 Oct 2025 16:21:53 GMT
• Title: Evolving LLMs' Self-Refinement Capability via Synergistic Training-Inference Optimization
• Authors: Yongcheng Zeng, Xinyu Cui, Xuanfa Jin, Qirui Mi, Guoqing Liu, Zexu Sun, Mengyue Yang, Dong Li, Weiyu Ma, Ning Yang, Jian Zhao, Jianye Hao, Haifeng Zhang, Jun Wang,
• Abstract summary: Self-Refinement refers to a model's ability to revise its own responses to produce improved outputs.<n>EVOLVE is a framework for eliciting and tracking the evolution of Self-Refinement through iterative training.<n>We demonstrate the potential of leveraging Self-Refinement to achieve broader Self-Improvement of intrinsic model abilities.
• Score: 53.93621974137829
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Self-Refinement refers to a model's ability to revise its own responses to produce improved outputs. This capability can also serve as a fundamental mechanism for Self-Improvement, for example, by reconstructing datasets with refined results to enhance intrinsic model performance. However, our comprehensive experiments reveal that large language models (LLMs) show no clear evidence of inherent Self-Refinement and may even experience response quality degradation after Self-Refinement. To address this issue, we propose EVOLVE, a simple and effective framework for eliciting and tracking the evolution of Self-Refinement through iterative training. We first explore optimization methods during training to activate the model's Self-Refinement capability. Then, at inference, we investigate various generation strategies to further enhance and utilize Self-Refinement while supplying the necessary data for training. Through synergistic optimization of training and inference stages, we continually evolve the model's Self-Refinement ability, enabling it to better refine its own responses. Moreover, we demonstrate the potential of leveraging Self-Refinement to achieve broader Self-Improvement of intrinsic model abilities. Experiments show that the evolved Self-Refinement ability enables the Llama-3.1-8B base model to surpass GPT-4o, achieving 62.3% length-controlled and 63.3% raw win rates on AlpacaEval 2, and 50.3% on Arena-Hard. It also generalizes effectively to out-of-domain reasoning tasks, improving performance on mathematical reasoning benchmarks such as GSM8K and MATH.

Related papers

• Can Large Reasoning Models Self-Train? [51.0277533541394]
  We use majority voting as a simple self-feedback mechanism to study whether self-training can be sustained within reinforcement learning.<n>We find that this basic approach improves not only the model's reasoning performance, but also its capability of generating better quality feedback for the next RL iteration.<n>Yet our analysis also reveals a critical limitation of such a self-training paradigm - prolonged RL with self-reward leads to reward hacking, resulting in sudden and complete performance collapse.
  arXiv  Detail & Related papers  (2025-05-27T17:16:00Z)
• ReflectEvo: Improving Meta Introspection of Small LLMs by Learning Self-Reflection [60.75785864719726]
  We present a novel pipeline, ReflectEvo, to demonstrate that small language models (SLMs) can enhance meta introspection through reflection learning.<n>We construct ReflectEvo-460k, a large-scale, comprehensive, self-generated reflection dataset with broadened instructions and diverse multi-domain tasks.
  arXiv  Detail & Related papers  (2025-05-22T10:03:05Z)
• Think, Prune, Train, Improve: Scaling Reasoning without Scaling Models [1.96238419451815]
  Large language models (LLMs) have demonstrated strong capabilities in programming and mathematical reasoning tasks, but are constrained by limited high-quality training data. We introduce a scalable framework that iteratively fine-tunes models on their own reasoning traces, using ground-truth pruning to ensure high-quality training data. This approach yields improved performance: on GSM8K, Gemma2-2B achieves a Pass@1 of 57.6% (from 41.9%), Gemma2-9B reaches 82%, matching LLaMA-3.1-70B, and LLaMA-3.1-70B attains 91%, even surpassing GPT-4o
  arXiv  Detail & Related papers  (2025-04-25T06:48:55Z)
• S$^2$R: Teaching LLMs to Self-verify and Self-correct via Reinforcement Learning [51.84977135926156]
  We introduce S$2$R, an efficient framework that enhances LLM reasoning by teaching models to self-verify and self-correct during inference. Our results demonstrate that Qwen2.5-math-7B achieves an accuracy improvement from 51.0% to 81.6%, outperforming models trained on an equivalent amount of long-CoT distilled data.
  arXiv  Detail & Related papers  (2025-02-18T13:40:22Z)
• From Drafts to Answers: Unlocking LLM Potential via Aggregation Fine-Tuning [31.95005389919542]
  Scaling data and model size has been proven effective for boosting the performance of large language models.<n>In this work, we introduce Aggregation Fine-Tuning (AFT), a supervised finetuning paradigm.<n> Empirical evaluations on benchmark datasets show that AFT-trained models substantially outperform standard SFT.
  arXiv  Detail & Related papers  (2025-01-21T04:11:59Z)
• FORLAPS: An Innovative Data-Driven Reinforcement Learning Approach for Prescriptive Process Monitoring [3.4437362489150254]
  This study introduces an innovative evaluation model, benchmarking its performance against earlier works using nine publicly available datasets.<n>The proposed model, FORLAPS, demonstrated exceptional performance, outperforming existing state-of-the-art approaches in suggesting the most optimal policies or predicting the best next activities within a process trace.
  arXiv  Detail & Related papers  (2025-01-17T20:31:35Z)
• Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models [10.449015816015566]
  Self-improvement is a mechanism in Large Language Model (LLM) pre-training, post-training and test-time inference.<n>We provide a mathematical formulation for self-improvement, which is largely governed by a quantity which we formalize as the generation-verification gap.<n>We also examine when self-improvement is possible, an iterative self-improvement procedure, and ways to improve its performance.
  arXiv  Detail & Related papers  (2024-12-03T18:47:26Z)
• Self-Improvement in Language Models: The Sharpening Mechanism [70.9248553790022]
  We offer a new perspective on the capabilities of self-improvement through a lens we refer to as sharpening.<n>Motivated by the observation that language models are often better at verifying response quality than they are at generating correct responses, we formalize self-improvement as using the model itself as a verifier during post-training.<n>We analyze two natural families of self-improvement algorithms based on SFT and RLHF.
  arXiv  Detail & Related papers  (2024-12-02T20:24:17Z)
• Evolving Alignment via Asymmetric Self-Play [52.3079697845254]
  We introduce a general open-ended RLHF framework that casts alignment as an asymmetric game between two players.<n>This framework of Evolving Alignment via Asymmetric Self-Play (eva) results in a simple and efficient approach that can utilize any existing RLHF algorithm for scalable alignment.
  arXiv  Detail & Related papers  (2024-10-31T08:15:32Z)
• Just Say What You Want: Only-prompting Self-rewarding Online Preference Optimization [64.34767799614328]
  Current self-rewarding approaches rely heavily on the discriminator's judgment capabilities. We propose a novel, only-prompting self-rewarding online algorithm that generates preference datasets without relying on judgment capabilities.
  arXiv  Detail & Related papers  (2024-09-26T04:41:08Z)
• Self-Taught Evaluators [77.92610887220594]
  We present an approach that aims to im-proves without human annotations, using synthetic training data only. Our Self-Taught Evaluator can improve a strong LLM from 75.4 to 88.3 on RewardBench.
  arXiv  Detail & Related papers  (2024-08-05T17:57:02Z)
• Weak-to-Strong Reasoning [33.20094938292376]
  We introduce a progressive learning framework that enables the strong model to autonomously refine its training data. Our method significantly enhances the reasoning capabilities of Llama2-70b using three separate weak models. This work paves the way for a more scalable and sophisticated strategy to enhance AI reasoning powers.
  arXiv  Detail & Related papers  (2024-07-18T16:25:17Z)
• Self-training Language Models for Arithmetic Reasoning [0.0]
  We explore the potential of improving models' reasoning capabilities without new data. We find that models can substantially improve in both single-round (offline) and online self-training.
  arXiv  Detail & Related papers  (2024-07-11T11:06:05Z)
• Efficient Continual Pre-training by Mitigating the Stability Gap [68.49269649759005]
  We study the behavior of Large Language Models (LLMs) during continual pre-training. We propose three effective strategies to enhance LLM performance within a fixed compute budget. Our strategies improve the average medical task performance of the OpenLlama-3B model from 36.2% to 40.7% with only 40% of the original training budget.
  arXiv  Detail & Related papers  (2024-06-21T02:28:37Z)
• Self-Play Preference Optimization for Language Model Alignment [75.83359213697854]
  Recent advancements suggest that directly working with preference probabilities can yield a more accurate reflection of human preferences. We propose a self-play-based method for language model alignment, which treats the problem as a constant-sum two-player game. Our approach, dubbed Self-Play Preference Optimization (SPPO), utilizes iterative policy updates to provably approximate the Nash equilibrium.
  arXiv  Detail & Related papers  (2024-05-01T17:59:20Z)
• Monte Carlo Tree Search Boosts Reasoning via Iterative Preference Learning [55.96599486604344]
  We introduce an approach aimed at enhancing the reasoning capabilities of Large Language Models (LLMs) through an iterative preference learning process. We use Monte Carlo Tree Search (MCTS) to iteratively collect preference data, utilizing its look-ahead ability to break down instance-level rewards into more granular step-level signals. The proposed algorithm employs Direct Preference Optimization (DPO) to update the LLM policy using this newly generated step-level preference data.
  arXiv  Detail & Related papers  (2024-05-01T11:10:24Z)
• Direct Nash Optimization: Teaching Language Models to Self-Improve with General Preferences [21.5605000515622]
  This paper studies post-training large language models (LLMs) using preference feedback from an oracle to help a model iteratively improve over itself. We introduce Direct Nash Optimization (DNO), a provable and efficient algorithm that marries the simplicity and stability of contrastive learning with theoretical generality from optimizing general preferences. In our experiments, a resulting 7B parameter Orca-2.5 model achieves the state-of-the-art win-rate against GPT-4-Turbo of 33% on AlpacaE 2.0 (even after controlling for response length), an absolute gain of 26% (7% to 33%) over the initializing model
  arXiv  Detail & Related papers  (2024-04-04T17:56:41Z)
• Aligner: Efficient Alignment by Learning to Correct [10.056049435141645]
  We introduce Aligner, a model-agnostic, plug-and-play module that learns the correctional residuals between preferred and dispreferred answers. It can be applied to various open-source and API-based models with only one-off training, making it suitable for rapid iteration. Our experiments demonstrate performance improvements by deploying the same Aligner model across 11 different language models.
  arXiv  Detail & Related papers  (2024-02-04T09:24:51Z)
• Augmenting Unsupervised Reinforcement Learning with Self-Reference [63.68018737038331]
  Humans possess the ability to draw on past experiences explicitly when learning new tasks. We propose the Self-Reference (SR) approach, an add-on module explicitly designed to leverage historical information. Our approach achieves state-of-the-art results in terms of Interquartile Mean (IQM) performance and Optimality Gap reduction on the Unsupervised Reinforcement Learning Benchmark.
  arXiv  Detail & Related papers  (2023-11-16T09:07:34Z)
• DavIR: Data Selection via Implicit Reward for Large Language Models [62.59514469369608]
  DavIR is a model-based data selection method for post-training Large Language Models.<n>We show that 6% of Alpaca dataset selected with DavIR can steer both the LLaMA and Gemma model family to produce superior performance compared to the same models trained on the full 52K dataset.
  arXiv  Detail & Related papers  (2023-10-16T07:26:24Z)
• SELF: Self-Evolution with Language Feedback [68.6673019284853]
  'SELF' (Self-Evolution with Language Feedback) is a novel approach to advance large language models. It enables LLMs to self-improve through self-reflection, akin to human learning processes. Our experiments in mathematics and general tasks demonstrate that SELF can enhance the capabilities of LLMs without human intervention.
  arXiv  Detail & Related papers  (2023-10-01T00:52:24Z)

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.