MetaGDPO: Alleviating Catastrophic Forgetting with Metacognitive Knowledge through Group Direct Preference Optimization

• URL: http://arxiv.org/abs/2511.12113v1
• Date: Sat, 15 Nov 2025 09:06:23 GMT
• Title: MetaGDPO: Alleviating Catastrophic Forgetting with Metacognitive Knowledge through Group Direct Preference Optimization
• Authors: Lanxue Zhang, Yuqiang Xie, Fang Fang, Fanglong Dong, Rui Liu, Yanan Cao,
• Abstract summary: Large Language Models demonstrate strong reasoning capabilities, which can be effectively compressed into smaller models.<n>Existing datasets and fine-tuning approaches still face challenges that lead to catastrophic forgetting.<n>We propose a comprehensive solution that alleviates catastrophic forgetting from both the data and fine-tuning approach perspectives.
• Score: 16.323544391974114
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models demonstrate strong reasoning capabilities, which can be effectively compressed into smaller models. However, existing datasets and fine-tuning approaches still face challenges that lead to catastrophic forgetting, particularly for models smaller than 8B. First, most datasets typically ignore the relationship between training data knowledge and the model's inherent abilities, making it difficult to preserve prior knowledge. Second, conventional training objectives often fail to constrain inherent knowledge preservation, which can result in forgetting of previously learned skills. To address these issues, we propose a comprehensive solution that alleviates catastrophic forgetting from both the data and fine-tuning approach perspectives. On the data side, we construct a dataset of 5K instances that covers multiple reasoning tasks and incorporates metacognitive knowledge, making it more tolerant and effective for distillation into smaller models. We annotate the metacognitive knowledge required to solve each question and filter the data based on task knowledge and the model's inherent skills. On the training side, we introduce GDPO (Group Direction Preference Optimization), which is better suited for resource-limited scenarios and can efficiently approximate the performance of GRPO. Guided by the large model and by implicitly constraining the optimization path through a reference model, GDPO enables more effective knowledge transfer from the large model and constrains excessive parameter drift. Extensive experiments demonstrate that our approach significantly alleviates catastrophic forgetting and improves reasoning performance on smaller models.

Related papers

• LLM-Inspired Pretrain-Then-Finetune for Small-Data, Large-Scale Optimization [7.8639568562295965]
  We consider small-data, large-scale decision problems in which a firm must make many operational decisions simultaneously.<n>We propose a pretrain-then-finetune approach built on a designed Transformer model to address this challenge.
  arXiv  Detail & Related papers  (2026-02-03T16:08:33Z)
• Parameter Importance-Driven Continual Learning for Foundation Models [5.471848114633189]
  Domain-specific post-training often causes catastrophic forgetting, making foundation models lose their general reasoning ability.<n>We introduce PIECE, a Importance Estimation-based Continual Enhancement method that preserves general ability while efficiently learning domain knowledge.<n>Our results highlight a practical path to scalable, domain-adaptive foundation models without catastrophic forgetting.
  arXiv  Detail & Related papers  (2025-11-19T12:07:53Z)
• Forgetting: A New Mechanism Towards Better Large Language Model Fine-tuning [51.92313556418432]
  Supervised fine-tuning (SFT) plays a critical role for pretrained large language models (LLMs)<n>We suggest categorizing tokens within each corpus into two parts -- positive and negative tokens -- based on whether they are useful to improve model performance.<n>We conduct experiments on well-established benchmarks, finding that this forgetting mechanism not only improves overall model performance and also facilitate more diverse model responses.
  arXiv  Detail & Related papers  (2025-08-06T11:22:23Z)
• iTool: Reinforced Fine-Tuning with Dynamic Deficiency Calibration for Advanced Tool Use [56.31110409360567]
  Augmenting large language models with external tools is a promising approach to enhance their capabilities.<n>We show that training gains significantly decay as synthetic data increases.<n>We propose an iterative reinforced fine-tuning strategy designed to alleviate this limitation.
  arXiv  Detail & Related papers  (2025-01-15T04:52:34Z)
• KBAlign: Efficient Self Adaptation on Specific Knowledge Bases [73.34893326181046]
  We present KBAlign, a self-supervised framework that enhances RAG systems through efficient model adaptation.<n>Our key insight is to leverage the model's intrinsic capabilities for knowledge alignment through two innovative mechanisms.<n> Experiments demonstrate that KBAlign can achieve 90% of the performance gain obtained through GPT-4-supervised adaptation.
  arXiv  Detail & Related papers  (2024-11-22T08:21:03Z)
• Unleashing LLM Reasoning Capability via Scalable Question Synthesis from Scratch [54.12139707822201]
  We propose ScaleQuest, a novel, scalable, and cost-effective data synthesis method.<n>By generating diverse questions from scratch, we produce a dataset of 1 million problem-solution pairs.<n>Our experiments demonstrate that models trained on our data outperform existing open-source datasets.
  arXiv  Detail & Related papers  (2024-10-24T12:42:04Z)
• POINTS: Improving Your Vision-language Model with Affordable Strategies [28.611705477757454]
  We train a robust baseline model using latest advancements in vision-language models. We filter pre-training data using perplexity, selecting the lowest perplexity data for training. During visual instruction tuning, we used model soup on different datasets when adding more datasets yielded marginal improvements.
  arXiv  Detail & Related papers  (2024-09-07T13:41:37Z)
• Towards Robust and Parameter-Efficient Knowledge Unlearning for LLMs [25.91643745340183]
  Large Language Models (LLMs) have demonstrated strong reasoning and memorization capabilities via pretraining on massive textual corpora.<n>This poses risk of privacy and copyright violations, highlighting the need for efficient machine unlearning methods.<n>We propose Low-rank Knowledge Unlearning (LoKU), a novel framework that enables robust and efficient unlearning for LLMs.
  arXiv  Detail & Related papers  (2024-08-13T04:18:32Z)
• Encapsulating Knowledge in One Prompt [56.31088116526825]
  KiOP encapsulates knowledge from various models into a solitary prompt without altering the original models or requiring access to the training data. From a practicality standpoint, this paradigm proves the effectiveness of Visual Prompt in data inaccessible contexts. Experiments across various datasets and models demonstrate the efficacy of the proposed KiOP knowledge transfer paradigm.
  arXiv  Detail & Related papers  (2024-07-16T16:35:23Z)
• Fantastic Gains and Where to Find Them: On the Existence and Prospect of General Knowledge Transfer between Any Pretrained Model [74.62272538148245]
  We show that for arbitrary pairings of pretrained models, one model extracts significant data context unavailable in the other. We investigate if it is possible to transfer such "complementary" knowledge from one model to another without performance degradation.
  arXiv  Detail & Related papers  (2023-10-26T17:59:46Z)
• Progressive reduced order modeling: empowering data-driven modeling with selective knowledge transfer [0.0]
  We propose a progressive reduced order modeling framework that minimizes data cravings and enhances data-driven modeling's practicality. Our approach selectively transfers knowledge from previously trained models through gates, similar to how humans selectively use valuable knowledge while ignoring unuseful information. We have tested our framework in several cases, including transport in porous media, gravity-driven flow, and finite deformation in hyperelastic materials.
  arXiv  Detail & Related papers  (2023-10-04T23:50:14Z)
• Towards Efficient Task-Driven Model Reprogramming with Foundation Models [52.411508216448716]
  Vision foundation models exhibit impressive power, benefiting from the extremely large model capacity and broad training data. However, in practice, downstream scenarios may only support a small model due to the limited computational resources or efficiency considerations. This brings a critical challenge for the real-world application of foundation models: one has to transfer the knowledge of a foundation model to the downstream task.
  arXiv  Detail & Related papers  (2023-04-05T07:28:33Z)
• Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation [26.744403789694758]
  We propose Data-Free Adversarial Robustness Distillation (DFARD) to train small, easily deployable, robust models without relying on data. Inspired by human education, we design a plug-and-play Interactive Temperature Adjustment (ITA) strategy to improve the efficiency of knowledge transfer.
  arXiv  Detail & Related papers  (2023-03-21T06:10:47Z)
• Dataless Knowledge Fusion by Merging Weights of Language Models [47.432215933099016]
  Fine-tuning pre-trained language models has become the prevalent paradigm for building downstream NLP models.<n>This creates a barrier to fusing knowledge across individual models to yield a better single model.<n>We propose a dataless knowledge fusion method that merges models in their parameter space.
  arXiv  Detail & Related papers  (2022-12-19T20:46:43Z)
• Large Language Models with Controllable Working Memory [64.71038763708161]
  Large language models (LLMs) have led to a series of breakthroughs in natural language processing (NLP) What further sets these models apart is the massive amounts of world knowledge they internalize during pretraining. How the model's world knowledge interacts with the factual information presented in the context remains under explored.
  arXiv  Detail & Related papers  (2022-11-09T18:58:29Z)
• Improving Sample Efficiency of Deep Learning Models in Electricity Market [0.41998444721319217]
  We propose a general framework, namely Knowledge-Augmented Training (KAT), to improve the sample efficiency. We propose a novel data augmentation technique to generate some synthetic data, which are later processed by an improved training strategy. Modern learning theories demonstrate the effectiveness of our method in terms of effective prediction error feedbacks, a reliable loss function, and rich gradient noises.
  arXiv  Detail & Related papers  (2022-10-11T16:35:13Z)

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.