litlas
ログインアカウント作成

Mass-Editing Memory in a Transformer

著者: Kevin Meng, Arnab Sen Sharma, Alex J. Andonian, Yonatan Belinkov, David Bau - International Conference on Learning Representations, ICLR 2023 被引用: 1,128

Recent work has shown exciting promise in updating large language models with new memories, so as to replace obsolete information or add specialized knowledge. However, this line of work is predominantly limited to updating single associations. We develop MEMIT, a method for directly updating a language model with many memories, demonstrating experimentally that it can scale up to thousands of associations for GPT-J (6B) and GPT-NeoX (20B), exceeding prior work by orders of magnitude. Our code and data are at https://memit.baulab.info.

HuggingFace'sTransformers…HuggingFace's Transformers: State-of-the-art Natural Language ProcessingDo Language Models HaveBeliefs? Methods for…Do Language Models Have Beliefs? Methods for Detecting, Updating, and Visualizing Model BeliefsCarbon Emissions andLarge Neural Network…Carbon Emissions and Large Neural Network TrainingTransformer Feed-ForwardLayers Build Prediction…Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary SpaceA Review on LanguageModels as Knowledge…A Review on Language Models as Knowledge BasesPaLM: Scaling LanguageModeling with PathwaysPaLM: Scaling Language Modeling with PathwaysAnalyzing Transformersin Embedding SpaceAnalyzing Transformers in Embedding SpaceEditing Large LanguageModels: Problems…Editing Large Language Models: Problems, Methods, and OpportunitiesCan We Edit MultimodalLarge Language Models?Can We Edit Multimodal Large Language Models?Backdoor ActivationAttack: Attack Large…Backdoor Activation Attack: Attack Large Language Models using Activation Steering for Safety-AlignmentPMET: Precise ModelEditing in a TransformerPMET: Precise Model Editing in a TransformerMELO: Enhancing ModelEditing with…MELO: Enhancing Model Editing with Neuron-Indexed Dynamic LoRALanguage ModelsRepresent Space and TimeLanguage Models Represent Space and TimeO-Edit: OrthogonalSubspace Editing for…O-Edit: Orthogonal Subspace Editing for Language Model Sequential EditingTrustLLM:Trustworthiness in Larg…TrustLLM: Trustworthiness in Large Language ModelsInterpreting ArithmeticMechanism in Large…Interpreting Arithmetic Mechanism in Large Language Models through Comparative Neuron AnalysisIn-Context Editing:Learning Knowledge from…In-Context Editing: Learning Knowledge from Self-Induced DistributionsRethinking Memory in AI:Taxonomy, Operations…Rethinking Memory in AI: Taxonomy, Operations, Topics, and Future DirectionsA Comprehensive Surveyin LLM(-Agent) Full…A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and DeploymentMass-Editing Memory in aTransformerMass-Editing Memory in a Transformer過去の参考文献中心の論文この論文を引用する論文古い新しい

ノードをクリックするとフォーカスを固定、空白をクリックすると本論文に戻ります。ホバーで一時的にプレビューできます。各ノードのページはタイトルから開けます。