Stabilizing LLM Supervised Fine-Tuning via Explicit Distributional Control

• URL: http://arxiv.org/abs/2605.04468v1
• Date: Wed, 06 May 2026 03:48:15 GMT
• Title: Stabilizing LLM Supervised Fine-Tuning via Explicit Distributional Control
• Abstract summary: Post-training large language models (LLMs) often suffer from catastrophic forgetting.<n>Recent evidence suggests this phenomenon is primarily driven by excessive distributional drift during optimization.<n>We propose Anchored Learning, a framework that explicitly controls distributional updates during offline fine-tuning.
• Score: 30.25101185177597
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Post-training large language models (LLMs) often suffers from catastrophic forgetting, where improvements on a target objective degrade previously acquired capabilities. Recent evidence suggests that this phenomenon is primarily driven by excessive distributional drift during optimization. Motivated by this perspective, we propose Anchored Learning, a simple framework that explicitly controls distributional updates during offline fine-tuning via a dynamically evolving moving anchor. Instead of matching a fixed reference distribution, the anchor interpolates between the current model and a frozen reference to construct an intermediate target that the model distills toward, transforming global fine-tuning into a sequence of local trust-region updates in distribution space. Theoretically, we prove this anchor-based update admits a linear KL-divergence upper bound per iteration, ensuring a stable transition between model distributions. Extensive experiments on iGSM, MedCalc, and IFEval show that Anchored Learning consistently lies on the Pareto frontier of gain-stability trade-offs, achieving near-optimal performance improvements while substantially reducing degradation compared to strong baselines. For example, while standard SFT suffers from over 53% performance degradation on iGSM and MedCalc, Anchored Learning slashes this drop to under 5% while maintaining near-optimal gains (e.g., 75.2% on iGSM).

Related papers

• OmniISR: A Unified Framework for Centralized and Federated Learning via Intermediate Supervision and Regularization [58.03221830946145]
  We propose OmniISR, a unified framework that fuses pure CL, pure FL, and hybrid CL-FL training modes.<n>We show that OmniISR consistently improves model performance in both centralized and federated paradigms.
  arXiv  Detail & Related papers  (2026-05-19T04:13:27Z)
• L-Drive: Beyond a Single Mapping-Latent Context Drives Time Series Forecasting [6.726418089422619]
  We propose L-Drive, a change-aware forecasting framework.<n>It provides more timely change cues and improves adaptation to changing segments.<n>Experiments validate the effectiveness of L-Drive and show a better overall trade-off between forecasting accuracy and computational efficiency.
  arXiv  Detail & Related papers  (2026-05-18T01:25:59Z)
• Self-Distilled Trajectory-Aware Boltzmann Modeling: Bridging the Training-Inference Discrepancy in Diffusion Language Models [65.89572755202245]
  Diffusion Language Models (DLMs) offer stronger global awareness and highly parallel generation.<n>Post-training DLMs with standard Negative Evidence Lower Bound (NELBO)-based supervised fine-tuning remains inefficient.<n>We propose a self-distilled trajectory-based post-training framework that aligns training with the easy-to-hard structure of inference.
  arXiv  Detail & Related papers  (2026-05-12T09:39:06Z)
• Power Reinforcement Post-Training of Text-to-Image Models with Super-Linear Advantage Shaping [66.25536973294726]
  Post-training methods for text-to-image (T2I) models are prone to reward hacking.<n>Super-Linear Advantage Shaping (SLAS) reshapes the local policy space.<n>SLAS consistently surpasses the DanceGRPO baseline across multiple backbones and benchmarks.
  arXiv  Detail & Related papers  (2026-05-11T17:59:25Z)
• AnchorRefine: Synergy-Manipulation Based on Trajectory Anchor and Residual Refinement for Vision-Language-Action Models [60.04879435087352]
  Vision-language-action (VLA) policies generate actions within a single unified space.<n>We propose AnchorRefine, a hierarchical framework that factorizes VLA action modeling into trajectory anchor and residual refinement.<n> Experiments on LIBERO, CALVIN, and real-robot tasks demonstrate that AnchorRefine consistently improves both regression-based and diffusion-based VLA backbones.
  arXiv  Detail & Related papers  (2026-04-20T04:25:24Z)
• GIRL: Generative Imagination Reinforcement Learning via Information-Theoretic Hallucination Control [0.0]
  We introduce GIRL (Generative Imagination Reinforcement Learning), a latent world-model framework that addresses this failure mode with two key components.<n> GIRL reduces latent rollout drift by 38 to 61 percent across tasks relative to DreamerV3, improves return, and requires fewer environment interactions on long-horizon tasks.<n>A distilled-prior variant reduces overhead and improves computational efficiency relative to the full model.
  arXiv  Detail & Related papers  (2026-04-08T17:14:21Z)
• GRPO-Guard: Mitigating Implicit Over-Optimization in Flow Matching via Regulated Clipping [63.33669214116784]
  GRPO-Guard is a simple yet effective enhancement to existing GRPO frameworks.<n>It restores a balanced and step-consistent importance ratio, ensuring that PPO clipping properly constrains harmful updates.<n>It substantially mitigates implicit over-optimization without relying on heavy KL regularization.
  arXiv  Detail & Related papers  (2025-10-25T14:51:17Z)
• STABLE: Gated Continual Learning for Large Language Models [0.0]
  STABLE is a gated continual self editing framework that constrains forgetting during sequential updates.<n>Each candidate edit is evaluated against a stability budget using one of three metrics.<n>Experiments on the Qwen-2.5-7B model show that gating effectively mitigates forgetting while preserving adaptability.
  arXiv  Detail & Related papers  (2025-10-17T16:14:05Z)
• Iterative Refinement of Flow Policies in Probability Space for Online Reinforcement Learning [56.47948583452555]
  We introduce the Stepwise Flow Policy (SWFP) framework, founded on the key insight that discretizing the flow matching inference process via a fixed-step Euler scheme aligns it with the variational Jordan-Kinderlehrer-Otto principle from optimal transport.<n>SWFP decomposes the global flow into a sequence of small, incremental transformations between proximate distributions.<n>This decomposition yields an efficient algorithm that fine-tunes pre-trained flows via a cascade of small flow blocks, offering significant advantages.
  arXiv  Detail & Related papers  (2025-10-17T07:43:51Z)
• WSM: Decay-Free Learning Rate Schedule via Checkpoint Merging for LLM Pre-training [64.0932926819307]
  We present Warmup-Stable and Merge (WSM), a framework that establishes a formal connection between learning rate decay and model merging.<n>WSM provides a unified theoretical foundation for emulating various decay strategies.<n>Our framework consistently outperforms the widely-adopted Warmup-Stable-Decay (WSD) approach across multiple benchmarks.
  arXiv  Detail & Related papers  (2025-07-23T16:02:06Z)
• Global Variational Inference Enhanced Robust Domain Adaptation [7.414646586981638]
  We propose a framework that learns continuous, class-conditional global priors via variational inference to enable structure-aware cross-domain alignment.<n>GVI-DA minimizes domain gaps through latent feature reconstruction, and mitigates posterior collapse using global codebook learning with randomized sampling.<n>It further improves robustness by discarding low-confidence pseudo-labels and generating reliable target-domain samples.
  arXiv  Detail & Related papers  (2025-07-04T04:43:23Z)

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.