Knowledge editing has been proposed as an effective method for updating and correcting the internal knowledge of Large Language Models (LLMs). However, existing editing methods often struggle with complex tasks, such as multi-hop reasoning. In this paper, we identify and investigate the phenomenon of Editing Overfit, where edited models assign disproportionately high probabilities to the edit target, hindering the generalization of new knowledge in complex scenarios. We attribute this issue to the current editing paradigm, which places excessive emphasis on the direct correspondence between the input prompt and the edit target for each edit sample. To further explore this issue, we introduce a new benchmark, EVOKE (EValuation of Editing Overfit in Knowledge Editing), along with fine-grained evaluation metrics. Through comprehensive experiments and analysis, we demonstrate that Editing Overfit is prevalent in current editing methods and that common overfitting mitigation strategies are of limited effectiveness in knowledge editing. To overcome this, inspired by LLMs' knowledge recall mechanisms, we propose a new plug-and-play strategy called Learn to Inference (LTI), which introduce a Multi-stage Inference Constraint module to guide the edited models in recalling new knowledge similarly to how unedited LLMs leverage knowledge through in-context learning. Extensive experimental results across a wide range of tasks validate the effectiveness of LTI in mitigating Editing Overfit.
In recent years, with the maturation of large language model (LLM) technology and the emergence of high-quality programming code datasets, researchers have become increasingly confident in addressing the challenges of program synthesis automatically. However, since most of the training samples for LLMs are unscreened, it is inevitable that LLMs' performance may not align with real-world scenarios, leading to the presence of social bias. To evaluate and quantify the gender bias in code LLMs, we propose a dataset named CodeGenBias (Gender Bias in the Code Generation) and an evaluation metric called FB-Score (Factual Bias Score) based on the actual gender distribution of correlative professions. With the help of CodeGenBias and FB-Score, we evaluate and analyze the gender bias in eight mainstream Code LLMs. Previous work has demonstrated that model editing methods that perform well in knowledge editing have the potential to mitigate social bias in LLMs. Therefore, we develop a model editing approach named MG-Editing (Multi-Granularity model Editing), which includes the locating and editing phases. Our model editing method MG-Editing can be applied at five different levels of model parameter granularity: full parameters level, layer level, module level, row level, and neuron level. Extensive experiments not only demonstrate that our MG-Editing can effectively mitigate the gender bias in code LLMs while maintaining their general code generation capabilities, but also showcase its excellent generalization. At the same time, the experimental results show that, considering both the gender bias of the model and its general code generation capability, MG-Editing is most effective when applied at the row and neuron levels of granularity.
Large Language Models (LLMs) have recently revolutionized the NLP field, while they still fall short in some specific down-stream tasks. In the work, we focus on utilizing LLMs to perform machine translation, where we observe that two patterns of errors frequently occur and drastically affect the translation quality: language mismatch and repetition. The work sets out to explore the potential for mitigating these two issues by leveraging model editing methods, e.g., by locating Feed-Forward Network (FFN) neurons or something that are responsible for the errors and deactivating them in the inference time. We find that directly applying such methods either limited effect on the targeted errors or has significant negative side-effect on the general translation quality, indicating that the located components may also be crucial for ensuring machine translation with LLMs on the rails. To this end, we propose to refine the located components by fetching the intersection of the locating results under different language settings, filtering out the aforementioned information that is irrelevant to targeted errors. The experiment results empirically demonstrate that our methods can effectively reduce the language mismatch and repetition ratios and meanwhile enhance or keep the general translation quality in most cases.
This work explores sequential model editing in large language models (LLMs), a critical task that involves modifying internal knowledge within LLMs continuously through multi-round editing, each incorporating updates or corrections to adjust the model outputs without the need for costly retraining. Existing model editing methods, especially those that alter model parameters, typically focus on single-round editing and often face significant challenges in sequential model editing-most notably issues of model forgetting and failure. To address these challenges, we introduce a new model editing method, namely \textbf{N}euron-level \textbf{S}equential \textbf{E}diting (NSE), tailored for supporting sequential model editing. Specifically, we optimize the target layer's hidden states using the model's original weights to prevent model failure. Furthermore, we iteratively select neurons in multiple layers for editing based on their activation values to mitigate model forgetting. Our empirical experiments demonstrate that NSE significantly outperforms current modifying parameters model editing methods, marking a substantial advancement in the field of sequential model editing. Our code is released on \url{https://github.com/jianghoucheng/NSE}.
To ensure large language models contain up-to-date knowledge, they need to be updated regularly. However, model editing is challenging as it might also affect knowledge that is unrelated to the new data. State-of-the-art methods identify parameters associated with specific knowledge and then modify them via direct weight updates. However, these locate-and-edit methods suffer from heavy computational overhead and lack theoretical validation. In contrast, directly fine-tuning the model on requested edits affects the model's behavior on unrelated knowledge, and significantly damages the model's generation fluency and consistency. To address these challenges, we propose SAUL, a streamlined model editing method that uses sentence concatenation with augmented random facts for generation regularization. Evaluations on three model editing benchmarks show that SAUL is a practical and reliable solution for model editing outperforming state-of-the-art methods while maintaining generation quality and reducing computational overhead.
Large language models (LLMs) often exhibit hallucinations due to incorrect or outdated knowledge. Hence, model editing methods have emerged to enable targeted knowledge updates. To achieve this, a prevailing paradigm is the locating-then-editing approach, which first locates influential parameters and then edits them by introducing a perturbation. While effective, current studies have demonstrated that this perturbation inevitably disrupt the originally preserved knowledge within LLMs, especially in sequential editing scenarios. To address this, we introduce AlphaEdit, a novel solution that projects perturbation onto the null space of the preserved knowledge before applying it to the parameters. We theoretically prove that this projection ensures the output of post-edited LLMs remains unchanged when queried about the preserved knowledge, thereby mitigating the issue of disruption. Extensive experiments on various LLMs, including LLaMA3, GPT2-XL, and GPT-J, show that AlphaEdit boosts the performance of most locating-then-editing methods by an average of 36.4% with a single line of additional code for projection solely. Our code is available at: https://github.com/jianghoucheng/AlphaEdit.
Despite significant advancements in large language models (LLMs), the rapid and frequent integration of small-scale experiences, such as interactions with surrounding objects, remains a substantial challenge. Two critical factors in assimilating these experiences are (1) Efficacy: the ability to accurately remember recent events; (2) Retention: the capacity to recall long-past experiences. Current methods either embed experiences within model parameters using continual learning, model editing, or knowledge distillation techniques, which often struggle with rapid updates and complex interactions, or rely on external storage to achieve long-term retention, thereby increasing storage requirements. In this paper, we propose SELF-PARAM (Self-Updatable Large Language Models with Parameter Integration). SELF-PARAM requires no extra parameters while ensuring near-optimal efficacy and long-term retention. Our method employs a training objective that minimizes the Kullback-Leibler (KL) divergence between the predictions of an original model (with access to contextual information) and a target model (without such access). By generating diverse question-answer pairs related to the knowledge and minimizing the KL divergence across this dataset, we update the target model to internalize the knowledge seamlessly within its parameters. Evaluations on question-answering and conversational recommendation tasks demonstrate that SELF-PARAM significantly outperforms existing methods, even when accounting for non-zero storage requirements. This advancement paves the way for more efficient and scalable integration of experiences in large language models by embedding knowledge directly into model parameters.
Large Language Models (LLMs) require frequent updates to correct errors and keep pace with continuously evolving knowledge in a timely and effective manner. Recent research in it model editing has highlighted the challenges in balancing generalization and locality, especially in the context of lifelong model editing. We discover that inserting knowledge directly into the model often causes conflicts and potentially disrupts other unrelated pre-trained knowledge. To address this problem, we introduce UniAdapt, a universal adapter for knowledge calibration. Inspired by the Mixture of Experts architecture and Retrieval-Augmented Generation, UniAdapt is designed with a vector-assisted router that is responsible for routing inputs to appropriate experts. The router maintains a vector store, including multiple shards, to construct routing vectors based on semantic similarity search results. UniAdapt is fully model-agnostic and designed for seamless plug-and-play integration. Experimental results show that UniAdapt outperforms existing lifelong model editors and achieves exceptional results in most metrics.
Training large language models (LLMs) from scratch is an expensive endeavor, particularly as world knowledge continually evolves. To maintain relevance and accuracy of LLMs, model editing has emerged as a pivotal research area. While these methods hold promise, they can also produce unintended side effects. Their underlying factors and causes remain largely unexplored. This paper delves into a critical factor-question type-by categorizing model editing questions. Our findings reveal that the extent of performance degradation varies significantly across different question types, providing new insights for experimental design in knowledge editing. Furthermore, we investigate whether insights from smaller models can be extrapolated to larger models. Our results indicate discrepancies in findings between models of different sizes, suggesting that insights from smaller models may not necessarily apply to larger models. Additionally, we examine the impact of batch size on side effects, discovering that increasing the batch size can mitigate performance drops.
We find arithmetic ability resides within a limited number of attention heads, with each head specializing in distinct operations. To delve into the reason, we introduce the Comparative Neuron Analysis (CNA) method, which identifies an internal logic chain consisting of four distinct stages from input to prediction: feature enhancing with shallow FFN neurons, feature transferring by shallow attention layers, feature predicting by arithmetic heads, and prediction enhancing among deep FFN neurons. Moreover, we identify the human-interpretable FFN neurons within both feature-enhancing and feature-predicting stages. These findings lead us to investigate the mechanism of LoRA, revealing that it enhances prediction probabilities by amplifying the coefficient scores of FFN neurons related to predictions. Finally, we apply our method in model pruning for arithmetic tasks and model editing for reducing gender bias. Code is on https://github.com/zepingyu0512/arithmetic-mechanism.
There have been many studies on analyzing self-supervised speech Transformers, in particular, with layer-wise analysis. It is, however, desirable to have an approach that can pinpoint exactly a subset of neurons that is responsible for a particular property of speech, being amenable to model pruning and model editing. In this work, we identify a set of property neurons in the feedforward layers of Transformers to study how speech-related properties, such as phones, gender, and pitch, are stored. When removing neurons of a particular property (a simple form of model editing), the respective downstream performance significantly degrades, showing the importance of the property neurons. We apply this approach to pruning the feedforward layers in Transformers, where most of the model parameters are. We show that protecting property neurons during pruning is significantly more effective than norm-based pruning. The code for identifying property neurons is available at https://github.com/nervjack2/PropertyNeurons.
With recent advances in building foundation models for texts and video data, there is a surge of interest in foundation models for time series. A family of models have been developed, utilizing a temporal auto-regressive generative Transformer architecture, whose effectiveness has been proven in Large Language Models. While the empirical results are promising, almost all existing time series foundation models have only been tested on well-curated ``benchmark'' datasets very similar to texts. However, real-world time series exhibit unique challenges, such as variable channel sizes across domains, missing values, and varying signal sampling intervals due to the multi-resolution nature of real-world data. Additionally, the uni-directional nature of temporally auto-regressive decoding limits the incorporation of domain knowledge, such as physical laws expressed as partial differential equations (PDEs). To address these challenges, we introduce the Time Diffusion Transformer (TimeDiT), a general foundation model for time series that employs a denoising diffusion paradigm instead of temporal auto-regressive generation. TimeDiT leverages the Transformer architecture to capture temporal dependencies and employs diffusion processes to generate high-quality candidate samples without imposing stringent assumptions on the target distribution via novel masking schemes and a channel alignment strategy. Furthermore, we propose a finetuning-free model editing strategy that allows the seamless integration of external knowledge during the sampling process without updating any model parameters. Extensive experiments conducted on a varity of tasks such as forecasting, imputation, and anomaly detection, demonstrate the effectiveness of TimeDiT.
Large language models (LLMs) require model editing to efficiently update specific knowledge within them and avoid factual errors. Most model editing methods are solely designed for single-time use and result in a significant forgetting effect in lifelong editing scenarios, where sequential edits are conducted over time. Previous approaches manage sequential edits by freezing original parameters and discretely allocating new parameters for each knowledge update. However, these methods lack robustness to minor input variations due to the discrete mapping between data and parameters. To overcome this challenge, we propose ELDER, a novel approach to create a continuous association between data and adapters. ELDER integrates multiple LoRAs through a router network and is trained to establish a smooth data-adapter association, thereby enhancing the edit robustness and generalization of semantically equivalent inputs. To ensure inputs containing the same knowledge will be processed by the same LoRAs, we design a novel loss to guide the model link LoRA allocations with edit knowledge. Furthermore, we propose a deferral mechanism to retain the original LLM capabilities post-edit. Extensive experiments on GPT-2 XL and LLaMA2-7B demonstrate that ELDER effectively edits models in the lifelong setting, outperforming eight baselines while exhibiting strong scalability and preserving LLMs' general abilities on downstream tasks. Our code is available at https://github.com/JiaangL/ELDER.
Large language models (LLMs) face challenges with internal knowledge inaccuracies and outdated information. Knowledge editing has emerged as a pivotal approach to mitigate these issues. Although current knowledge editing techniques exhibit promising performance in single-hop reasoning tasks, they show limitations when applied to multi-hop reasoning. Drawing on cognitive neuroscience and the operational mechanisms of LLMs, we hypothesize that the residual single-hop knowledge after editing causes edited models to revert to their original answers when processing multi-hop questions, thereby undermining their performance in multihop reasoning tasks. To validate this hypothesis, we conduct a series of experiments that empirically confirm our assumptions. Building on the validated hypothesis, we propose a novel knowledge editing method that incorporates a Knowledge Erasure mechanism for Large language model Editing (KELE). Specifically, we design an erasure function for residual knowledge and an injection function for new knowledge. Through joint optimization, we derive the optimal recall vector, which is subsequently utilized within a rank-one editing framework to update the parameters of targeted model layers. Extensive experiments on GPT-J and GPT-2 XL demonstrate that KELE substantially enhances the multi-hop reasoning capability of edited LLMs.
Weight-preserving model editing techniques heavily rely on the scoping mechanism that decides when to apply an edit to the base model. These scoping mechanisms utilize distance functions in the representation space to ascertain the scope of the edit. In this work, we show that distance-based scoping functions grapple with lexical biases leading to issues such as misfires with irrelevant prompts that share similar lexical characteristics. To address this problem, we introduce, Projector Editor Networks for Model Editing (PENME),is a model editing approach that employs a compact adapter with a projection network trained via a contrastive learning objective. We demonstrate the efficacy of PENME in achieving superior results while being compute efficient and flexible to adapt across model architectures.
Large language models (LLMs) are trained on extensive text corpora, which inevitably include biased information. Although techniques such as Affective Alignment can mitigate some negative impacts of these biases, existing prompt-based attack methods can still extract these biases from the model's weights. Moreover, these biases frequently appear subtly when LLMs are prompted to perform identical tasks across different demographic groups, thereby camouflaging their presence. To address this issue, we have formally defined the implicit bias problem and developed an innovative framework for bias removal based on Bayesian theory, Bayesian-Theory based Bias Removal (BTBR). BTBR employs likelihood ratio screening to pinpoint data entries within publicly accessible biased datasets that represent biases inadvertently incorporated during the LLM training phase. It then automatically constructs relevant knowledge triples and expunges bias information from LLMs using model editing techniques. Through extensive experimentation, we have confirmed the presence of the implicit bias problem in LLMs and demonstrated the effectiveness of our BTBR approach.
Large language models (LLMs) have demonstrated remarkable capabilities. Their powerful generative abilities enable flexible responses based on various queries or instructions. Emerging as widely adopted generalists for diverse tasks, LLMs are still vulnerable to backdoors. This paper proposes an editing-based generative backdoor, named MEGen, aiming to create a customized backdoor for NLP tasks with the least side effects. In our approach, we first leverage a language model to insert a trigger selected on fixed metrics into the input, then design a pipeline of model editing to directly embed a backdoor into an LLM. By adjusting a small set of local parameters with a mini-batch of samples, MEGen significantly enhances time efficiency and achieves high robustness. Experimental results indicate that our backdoor attack strategy achieves a high attack success rate on poison data while maintaining the model's performance on clean data. Notably, the backdoored model, when triggered, can freely output pre-set dangerous information while successfully completing downstream tasks. This suggests that future LLM applications could be guided to deliver certain dangerous information, thus altering the LLM's generative style. We believe this approach provides insights for future LLM applications and the execution of backdoor attacks on conversational AI systems.
Model editing aims to correct outdated or erroneous knowledge in large models without costly retraining. Recent research discovered that the mid-layer representation of the subject's final token in a prompt has a strong influence on factual predictions, and developed Large Language Model (LLM) editing techniques based on this observation. However, for Vision-LLMs (VLLMs), how visual representations impact the predictions from a decoder-only language model remains largely unexplored. To the best of our knowledge, model editing for VLLMs has not been extensively studied in the literature. In this work, we employ the contribution allocation and noise perturbation methods to measure the contributions of visual representations for token predictions. Our attribution analysis shows that visual representations in mid-to-later layers that are highly relevant to the prompt contribute significantly to predictions. Based on these insights, we propose VisEdit, a novel model editor for VLLMs that effectively corrects knowledge by editing intermediate visual representations in regions important to the edit prompt. We evaluated VisEdit using multiple VLLM backbones and public VLLM editing benchmark datasets. The results show the superiority of VisEdit over the strong baselines adapted from existing state-of-the-art editors for LLMs.
The advent of Large Language Models (LLMs) has garnered significant popularity and wielded immense power across various domains within Natural Language Processing (NLP). While their capabilities are undeniably impressive, it is crucial to identify and scrutinize their vulnerabilities especially when those vulnerabilities can have costly consequences. One such LLM, trained to provide a concise summarization from medical documents could unequivocally leak personal patient data when prompted surreptitiously. This is just one of many unfortunate examples that have been unveiled and further research is necessary to comprehend the underlying reasons behind such vulnerabilities. In this study, we delve into multiple sections of vulnerabilities which are model-based, training-time, inference-time vulnerabilities, and discuss mitigation strategies including "Model Editing" which aims at modifying LLMs behavior, and "Chroma Teaming" which incorporates synergy of multiple teaming strategies to enhance LLMs' resilience. This paper will synthesize the findings from each vulnerability section and propose new directions of research and development. By understanding the focal points of current vulnerabilities, we can better anticipate and mitigate future risks, paving the road for more robust and secure LLMs.
Large language models (LLMs) can often be made to behave in undesirable ways that they are explicitly fine-tuned not to. For example, the LLM red-teaming literature has produced a wide variety of 'jailbreaking' techniques to elicit harmful text from models that were fine-tuned to be harmless. Recent work on red-teaming, model editing, and interpretability suggests that this challenge stems from how (adversarial) fine-tuning largely serves to suppress rather than remove undesirable capabilities from LLMs. Prior work has introduced latent adversarial training (LAT) as a way to improve robustness to broad classes of failures. These prior works have considered untargeted latent space attacks where the adversary perturbs latent activations to maximize loss on examples of desirable behavior. Untargeted LAT can provide a generic type of robustness but does not leverage information about specific failure modes. Here, we experiment with targeted LAT where the adversary seeks to minimize loss on a specific competing task. We find that it can augment a wide variety of state-of-the-art methods. First, we use targeted LAT to improve robustness to jailbreaks, outperforming a strong R2D2 baseline with orders of magnitude less compute. Second, we use it to more effectively remove backdoors with no knowledge of the trigger. Finally, we use it to more effectively unlearn knowledge for specific undesirable tasks in a way that is also more robust to re-learning. Overall, our results suggest that targeted LAT can be an effective tool for defending against harmful behaviors from LLMs.