Understanding the Collapse of LLMs in Model Editing

• URL: http://arxiv.org/abs/2406.11263v2
• Date: Mon, 30 Sep 2024 06:37:03 GMT
• Title: Understanding the Collapse of LLMs in Model Editing
• Authors: Wanli Yang, Fei Sun, Jiajun Tan, Xinyu Ma, Du Su, Dawei Yin, Huawei Shen,
• Abstract summary: We study the root causes of such collapse. We propose a simple yet effective approach: uniformly using prefixed keys during editing phase and adding prefixes during testing phase.
• Score: 37.429695927372755
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Despite significant progress in model editing methods, their application in real-world scenarios remains challenging as they often cause large language models (LLMs) to collapse. Among them, ROME is particularly concerning, as it could disrupt LLMs with only a single edit. In this paper, we study the root causes of such collapse. Through extensive analysis, we identify two primary factors that contribute to the collapse: i) inconsistent handling of prefixed and unprefixed keys in the parameter update equation may result in very small denominators, causing excessively large parameter updates; ii) the subject of collapse cases is usually the first token, whose unprefixed key distribution significantly differs from the prefixed key distribution in autoregressive transformers, causing the aforementioned issue to materialize. To validate our findings, we propose a simple yet effective approach: uniformly using prefixed keys during editing phase and adding prefixes during testing phase to ensure the consistency between training and testing. The experimental results show that the proposed solution can prevent model collapse while maintaining the effectiveness of the edits.

Related papers

• MEMOIR: Lifelong Model Editing with Minimal Overwrite and Informed Retention for LLMs [82.34547399693966]
  Existing methods for lifelong model editing compromise generalization, interfere with past edits, or fail to scale to long editing sequences.<n>We propose MEMOIR, a novel scalable framework that injects knowledge through a residual memory.<n>MeMOIR confines each edit to a distinct subset of the memory parameters, minimizing interference among edits.
  arXiv  Detail & Related papers  (2025-06-09T16:16:42Z)
• Multimodal LLM-Guided Semantic Correction in Text-to-Image Diffusion [52.315729095824906]
  MLLM Semantic-Corrected Ping-Pong-Ahead Diffusion (PPAD) is a novel framework that introduces a Multimodal Large Language Model (MLLM) as a semantic observer during inference.<n>It performs real-time analysis on intermediate generations, identifies latent semantic inconsistencies, and translates feedback into controllable signals that actively guide the remaining denoising steps.<n>Extensive experiments demonstrate PPAD's significant improvements.
  arXiv  Detail & Related papers  (2025-05-26T14:42:35Z)
• REACT: Representation Extraction And Controllable Tuning to Overcome Overfitting in LLM Knowledge Editing [42.89229070245538]
  We introduce REACT, a framework for precise and controllable knowledge editing.<n>In the initial phase, we utilize tailored stimuli to extract latent factual representations.<n>In the second phase, we apply controllable perturbations to hidden states using the obtained vector with a magnitude scalar.
  arXiv  Detail & Related papers  (2025-05-25T01:57:06Z)
• Learning Where to Edit Vision Transformers [27.038720045544867]
  We propose a locate-then-edit approach for editing vision Transformers (ViTs) in computer vision. We first address the where-to-edit challenge by meta-learning a hypernetwork on CutMix-augmented data. To validate our method, we construct an editing benchmark that introduces subpopulation shifts towards natural underrepresented images and AI-generated images.
  arXiv  Detail & Related papers  (2024-11-04T10:17:40Z)
• Keys to Robust Edits: from Theoretical Insights to Practical Advances [20.10464264597003]
  Large language models (LLMs) have revolutionized knowledge storage and retrieval, but face challenges with conflicting and outdated information. This work investigates the cause of failures in locate-and-edit methods, offering theoretical insights into their key-value modeling. Empirical analysis reveals that keys used by current methods fail to meet robustness and specificity requirements.
  arXiv  Detail & Related papers  (2024-10-12T02:54:12Z)
• Deep Autoregressive Models as Causal Inference Engines [38.26602521505842]
  We propose an autoregressive (AR) causal inference framework capable of handling complex confounders and sequential actions.<n>Our approach accomplishes this using em sequencification, which transforms data from an underlying causal diagram into a sequence of tokens.<n>We demonstrate that an AR model adapted for CI is efficient and effective in various complex applications such as navigating mazes, playing chess endgames, and evaluating the impact of certain keywords on paper acceptance rates.
  arXiv  Detail & Related papers  (2024-09-27T09:37:09Z)
• ELDER: Enhancing Lifelong Model Editing with Mixture-of-LoRA [55.697627106315004]
  Large language models (LLMs) require model editing to efficiently update specific knowledge within them and avoid factual errors. Previous approaches manage sequential edits by freezing original parameters and discretely allocating new parameters for each knowledge update. We propose ELDER, a novel approach to create a continuous association between data and adapters.
  arXiv  Detail & Related papers  (2024-08-19T02:27:00Z)
• Set-Based Prompting: Provably Solving the Language Model Order Dependency Problem [18.020492646988746]
  We present Set-Based Prompting, a technique that guarantees the output of an LLM will not have order dependence on a specified set of sub-sequences. Despite our inputs being out of distribution, the impact on expected accuracy is small, where the expectation is over the order of uniformly chosen shuffling of the candidate responses.
  arXiv  Detail & Related papers  (2024-06-04T16:09:13Z)
• Alice in Wonderland: Simple Tasks Showing Complete Reasoning Breakdown in State-Of-the-Art Large Language Models [13.532180752491954]
  We demonstrate a dramatic breakdown of function and reasoning capabilities of state-of-the-art models trained at the largest available scales. The breakdown is dramatic, as models show strong fluctuations across even slight problem variations that should not affect problem solving. We take these initial observations to stimulate urgent re-assessment of the claimed capabilities of current generation of Large Language Models.
  arXiv  Detail & Related papers  (2024-06-04T07:43:33Z)
• Disperse-Then-Merge: Pushing the Limits of Instruction Tuning via Alignment Tax Reduction [75.25114727856861]
  Large language models (LLMs) tend to suffer from deterioration at the latter stage ofSupervised fine-tuning process. We introduce a simple disperse-then-merge framework to address the issue. Our framework outperforms various sophisticated methods such as data curation and training regularization on a series of standard knowledge and reasoning benchmarks.
  arXiv  Detail & Related papers  (2024-05-22T08:18:19Z)
• The Butterfly Effect of Model Editing: Few Edits Can Trigger Large Language Models Collapse [58.0132400208411]
  Even a single edit can trigger model collapse, manifesting as significant performance degradation in various benchmark tasks. benchmarking Large Language Models after each edit is impractically time-consuming and resource-intensive. We have utilized GPT-3.5 to develop a new dataset, HardEdit, based on hard cases.
  arXiv  Detail & Related papers  (2024-02-15T01:50:38Z)
• Hard Nominal Example-aware Template Mutual Matching for Industrial Anomaly Detection [74.9262846410559]
  textbfHard Nominal textbfExample-aware textbfTemplate textbfMutual textbfMatching (HETMM) textitHETMM aims to construct a robust prototype-based decision boundary, which can precisely distinguish between hard-nominal examples and anomalies.
  arXiv  Detail & Related papers  (2023-03-28T17:54:56Z)
• Mutual Exclusivity Training and Primitive Augmentation to Induce Compositionality [84.94877848357896]
  Recent datasets expose the lack of the systematic generalization ability in standard sequence-to-sequence models. We analyze this behavior of seq2seq models and identify two contributing factors: a lack of mutual exclusivity bias and the tendency to memorize whole examples. We show substantial empirical improvements using standard sequence-to-sequence models on two widely-used compositionality datasets.
  arXiv  Detail & Related papers  (2022-11-28T17:36:41Z)
• Memory-Based Model Editing at Scale [102.28475739907498]
  Existing model editors struggle to accurately model an edit's intended scope. We propose Semi-Parametric Editing with a Retrieval-Augmented Counterfactual Model (SERAC) SERAC stores edits in an explicit memory and learns to reason over them to modulate the base model's predictions as needed.
  arXiv  Detail & Related papers  (2022-06-13T23:40:34Z)

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.