Editing Models with Task Arithmetic

2022/12/08 by Gabriel Ilharco, Marco Túlio Ribeiro, Ilharco, Gabriel +12 · 1 voice · 292 citations
Computer Science · #Adversarial Robustness in Machine Learning #Computation and Language (cs.CL) #Computer Vision and Pattern Recognition (cs.CV) #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning and Data Classification #cs.CL #cs.CV #cs.LG

paper · pdf · doi:10.48550/arxiv.2212.04089

openalex publication_date 2022/12/08 · arxiv published 2022/12/08 · arxiv updated 2023/03/31 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Changing how pre-trained models behave -- e.g., improving their performance on a downstream task or mitigating biases learned during pre-training -- is a common practice when developing machine learning systems. In this work, we propose a new paradigm for steering the behavior of neural networks, centered around task vectors. A task vector specifies a direction in the weight space of a pre-trained model, such that movement in that direction improves performance on the task. We build task vectors by subtracting the weights of a pre-trained model from the weights of the same model after fine-tuning on a task. We show that these task vectors can be modified and combined together through arithmetic operations such as negation and addition, and the behavior of the resulting model is steered accordingly. Negating a task vector decreases performance on the target task, with little change in model behavior on control tasks. Moreover, adding task vectors together can improve performance on multiple tasks at once. Finally, when tasks are linked by an analogy relationship of the form ``A is to B as C is to D", combining task vectors from three of the tasks can improve performance on the fourth, even when no data from the fourth task is used for training. Overall, our experiments with several models, modalities and tasks show that task arithmetic is a simple, efficient and effective way of editing models.

Cited by

• Rethinking Expert Training for Model Merging with Prompt Learning
• The Appeal and Reality of Recycling LoRAs with Adaptive Merging
• Mechanistic Evidence for Preserved-but-Misaligned Representations in Non-IID FedAvg
• Merge before Forget: A Single LoRA Continual Learning via Continual Merging
• Decomposing Task Vectors for Refined Model Editing
• GLUE: Gradient-free Learning to Unify Experts
• The Effectiveness of Approximate Regularized Replay for Efficient Supervised Fine-Tuning of Large Language Models
• Distributed Sparse Interventions in Language Models
• The Intruder Threshold: A Spectral Law for LoRA Fine-Tuning
• Forgetting Is Not a Fix: Path Dependence in Sequential Engram Editing
• Rare Word Recognition and Translation Without Fine-Tuning via Task Vector in Speech Models
• Model Merging via Multi-Teacher Knowledge Distillation
• The Erasure Illusion: Stress-Testing the Generalization of LLM Forgetting Evaluation
• MAGIC: Achieving Superior Model Merging via Magnitude Calibration
• Task Vector in TTS: Toward Emotionally Expressive Dialectal Speech Synthesis
• Bridging Training and Merging Through Momentum-Aware Optimization
• Feature-Selective Representation Misdirection for Machine Unlearning
• Bolmo: Byteifying the Next Generation of Language Models
• Task Matrices: Linear Maps for Cross-Model Finetuning Transfer
• GTR-Turbo: Merged Checkpoint is Secretly a Free Teacher for Agentic VLM Training
• Phase transitions reveal hierarchical structure in deep neural networks
• Shapley-based Data Valuation for LLM Alignment via Sequential Preference Optimization
• AP-BMM: Approximating Capability-Cost Pareto Sets of LLMs via Asynchronous Prior-Guided Bayesian Model Merging
• Exploring Test-time Scaling via Prediction Merging on Large-Scale Recommendation
• Each Prompt Matters: Scaling Reinforcement Learning Without Wasting Rollouts on Hundred-Billion-Scale MoE
• LUNE: Efficient LLM Unlearning via LoRA Fine-Tuning with Negative Examples
• ProSocialAlign: Preference Conditioned Test Time Alignment in Language Models
• EvoEdit: Lifelong Free-Text Knowledge Editing through Latent Perturbation Augmentation and Knowledge-driven Parameter Fusion
• Basis-Oriented Low-rank Transfer for Few-Shot and Test-Time Adaptation
• Stay Unique, Stay Efficient: Preserving Model Personality in Multi-Task Merging
• Elastic Mixture of Rank-Wise Experts for Knowledge Reuse in Federated Fine-Tuning
• From Coefficients to Directions: Rethinking Model Merging with Directional Alignment
• Group-Aware Partial Model Merging for Children's Automatic Speech Recognition
• A Systematic Study of In-the-Wild Model Merging for Large Language Models
• MortgageLLM: Domain-Adaptive Pretraining with Residual Instruction Transfer, Alignment Tuning, and Task-Specific Routing
• Towards Benign Memory Forgetting for Selective Multimodal Large Language Model Unlearning
• Merging without Forgetting: Continual Fusion of Task-Specific Models via Optimal Transport
• Understanding and Mitigating Over-refusal for Large Language Models via Safety Representation
• MergeVLA: Cross-Skill Model Merging Toward a Generalist Vision-Language-Action Agent
• Subtract the Corruption: Training-Data-Free Corrective Machine Unlearning using Task Arithmetic
• A Novel Hierarchical Integration Method for Efficient Model Merging in Medical LLMs
• MergeSlide: Continual Model Merging and Task-to-Class Prompt-Aligned Inference for Lifelong Learning on Whole Slide Images
• RETROFIT: Continual Learning with Controlled Forgetting for Binary Security Detection and Analysis
• Forgetting-MarI: LLM Unlearning via Marginal Information Regularization
• Defending Unauthorized Model Merging via Dual-Stage Weight Protection
• EnchTable: Unified Safety Alignment Transfer in Fine-tuned Large Language Models
• LT-Soups: Bridging Head and Tail Classes via Subsampled Model Soups
• Continual Unlearning for Text-to-Image Diffusion Models: A Regularization Perspective
• You Had One Job: Per-Task Quantization Using LLMs' Hidden Representations
• DRAGON: Guard LLM Unlearning in Context via Negative Detection and Reasoning
• Steering Language Models with Weight Arithmetic
• Model Merging Improves Zero-Shot Generalization in Bioacoustic Foundation Models
• Shared Parameter Subspaces and Cross-Task Linearity in Emergently Misaligned Behavior
• T3: Test-Time Model Merging in VLMs for Zero-Shot Medical Imaging Analysis
• MIN-Merging: Merge the Important Neurons for Model Merging
• LAMP: Data-Efficient Linear Affine Weight-Space Models for Parameter-Controlled 3D Shape Generation and Extrapolation
• Multi-Task Vehicle Routing Solver via Mixture of Specialized Experts under State-Decomposable MDP
• Model Merging with Functional Dual Anchors
• PLAN: Proactive Low-Rank Allocation for Continual Learning
• Continual Knowledge Adaptation for Reinforcement Learning
• HIDISC: A Hyperbolic Framework for Domain Generalization with Generalized Category Discovery
• Reasoning Distillation and Structural Alignment for Improved Code Generation
• Forgetting to Forget: Attention Sink as A Gateway for Backdooring LLM Unlearning
• Adaptive Minds: Empowering Agents with LoRA-as-Tools
• Directional Reasoning Injection for Fine-Tuning MLLMs
• Backdoor Unlearning by Linear Task Decomposition
• Purifying Task Vectors in Knowledge-Aware Subspace for Model Merging
• Can MLLMs Absorb Math Reasoning Abilities from LLMs as Free Lunch?
• Towards Reversible Model Merging For Low-rank Weights
• REAP the Experts: Why Pruning Prevails for One-Shot MoE compression
• Towards Robust Knowledge Removal in Federated Learning with High Data Heterogeneity
• Weight Weaving: Parameter Pooling for Data-Free Model Merging
• Research in Collaborative Learning Does Not Serve Cross-Silo Federated Learning in Practice
• Exploring and Leveraging Class Vectors for Classifier Editing
• Enhancing Faithfulness in Abstractive Summarization via Span-Level Fine-Tuning
• Towards Efficient Multimodal Unified Reasoning Model via Model Merging
• Backdoor Vectors: a Task Arithmetic View on Backdoor Attacks and Defenses
• FlyLoRA: Boosting Task Decoupling and Parameter Efficiency via Implicit Rank-Wise Mixture-of-Experts
• Transmuting prompts into weights
• Gradient-Sign Masking for Task Vector Transport Across Pre-Trained Models
• MixReasoning: Switching Modes to Think
• Refusal Falls off a Cliff: How Safety Alignment Fails in Reasoning?
• Boomerang Distillation Enables Zero-Shot Model Size Interpolation
• Learning to Interpret Weight Differences in Language Models
• How does the optimizer implicitly bias the model merging loss landscape?
• MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering
• BiasFreeBench: a Benchmark for Mitigating Bias in Large Language Model Responses
• Expert Merging: Model Merging with Unsupervised Expert Alignment and Importance-Guided Layer Chunking
• TDHook: A Lightweight Framework for Interpretability
• Understanding the Dilemma of Unlearning for Large Language Models
• Real-Aware Residual Model Merging for Deepfake Detection
• Model Merging Scaling Laws in Large Language Models
• Bridging the Knowledge-Prediction Gap in LLMs on Multiple-Choice Questions
• Merge Now, Regret Later: The Hidden Cost of Model Merging is Adversarial Transferability
• Toward a Holistic Approach to Continual Model Merging
• Temporal Generalization: A Reality Check
• Dual-Space Smoothness for Robust and Balanced LLM Unlearning
• Context Parametrization with Compositional Adapters
• The Thinking Spectrum: An Empirical Study of Tunable Reasoning in LLMs through Model Merging
• Closing the Oracle Gap: Increment Vector Transformation for Class Incremental Learning
• SFT Doesn't Always Hurt General Capabilities: Revisiting Domain-Specific Fine-Tuning in LLMs
• Null-Space Filtering for Data-Free Continual Model Merging: Preserving Transparency, Promoting Fidelity
• Faster, Smaller, and Smarter: Task-Aware Expert Merging for Online MoE Inference
• Unifying Adversarially Robust Model Experts in Vision-Language Models
• SkillSmith: Learning to Compose Parametric Skills and Textual Knowledge
• Asymmetric Collapse in Model Merging: When Refusal Over- writes Recognition
• Dense Supervision, Sparse Updates: On the Sparsity and Geometry of On-Policy Distillation
• Transporting Task Vectors across Different Architectures without Training
• AMELIA: A Family of Multi-task End-to-end Language Models for Argumentation
• LLM-based Agents Suffer from Hallucinations: A Survey of Taxonomy, Methods, and Directions
• Accurate and Efficient Low-Rank Model Merging in Core Space
• SEQR: Secure and Efficient QR-based LoRA Routing
• nDNA -- the Semantic Helix of Artificial Cognition
• Variational Task Vector Composition
• Local Mechanisms of Compositional Generalization in Conditional Diffusion
• HAM: Hierarchical Adapter Merging for Scalable Continual Learning
• Harnessing Optimization Dynamics for Curvature-Informed Model Merging
• Continually Adding New Languages to Multilingual Language Models
• Delta Activations: A Representation for Finetuned Large Language Models
• Modular Embedding Recomposition for Incremental Learning
• Unlearning That Lasts: Utility-Preserving, Robust, and Almost Irreversible Forgetting in LLMs
• Surrogate Benchmarks for Model Merging Optimization
• Reasoning Vectors: Transferring Chain-of-Thought Capabilities via Task Arithmetic
• On Task Vectors and Gradients
• Model Unmerging: Making Your Models Unmergeable for Secure Model Sharing
• Improving Fisher Information Estimation and Efficiency for LoRA-based LLM Unlearning
• Rethinking Layer-wise Model Merging through Chain of Merges
• Lethe: Purifying Backdoored Large Language Models with Knowledge Dilution
• PSO-Merging: Merging Models Based on Particle Swarm Optimization
• UNIFORM: Unifying Knowledge from Large-scale and Diverse Pre-trained Models
• Think in Blocks: Adaptive Reasoning from Direct Response to Deep Reasoning
• Efficient Multi-Source Knowledge Transfer by Model Merging
• Learn Faster and Remember More: Balancing Exploration and Exploitation for Continual Test-time Adaptation
• Cost-Aware Contrastive Routing for LLMs
• Rethinking Safety in LLM Fine-tuning: An Optimization Perspective
• MedSAMix: A Training-Free Model Merging Approach for Medical Image Segmentation
• VisCodex: Unified Multimodal Code Generation via Merging Vision and Coding Models
• Hierarchical Adaptive networks with Task vectors for Test-Time Adaptation
• Low-Rank Expert Merging for Multi-Source Domain Adaptation in Person Re-Identification
• Sculpting Margin Penalty: Intra-Task Adapter Merging and Classifier Calibration for Few-Shot Class-Incremental Learning
• Learning from Oblivion: Predicting Knowledge Overflowed Weights via Retrodiction of Forgetting
• RegMean++: Enhancing Effectiveness and Generalization of Regression Mean for Model Merging
• Industrial LLM-based Code Optimization under Regulation: A Mixture-of-Agents Approach
• Welcome New Doctor: Continual Learning with Expert Consultation and Autoregressive Inference for Whole Slide Image Analysis
• DisTaC: Conditioning Task Vectors via Distillation for Robust Model Merging
• Watch the Weights: Unsupervised monitoring and control of fine-tuned LLMs
• Forgetting of task-specific knowledge in model merging-based continual learning
• Modular Delta Merging with Orthogonal Constraints: A Scalable Framework for Continual and Reversible Model Composition
• Uncertainty-driven Embedding Convolution
• CLoRA: Parameter-Efficient Continual Learning with Low-Rank Adaptation
• Fishers for Free? Approximating the Fisher Information Matrix by Recycling the Squared Gradient Accumulator
• modelDNA: Calibrated Lineage Verification and Merge Decomposition from Sampled Weight Fingerprints
• KAT-Coder-V2.5 Technical Report
• MeMo: Memory as a Model
• Darwin Family: MRI-Trust-Weighted Evolutionary Merging for Training-Free Scaling of Language-Model Reasoning
• Look the Other Way: Designing 'Positive' Molecules with Negative Data via Task Arithmetic
• Harmonizing and Merging Source Models for CLIP-based Domain Generalization
• Merging Smarter, Generalizing Better: Enhancing Model Merging on OOD Data
• Do Concept Replacement Techniques Really Erase Unacceptable Concepts?
• What Should LLMs Forget? Quantifying Personal Data in LLMs for Right-to-Be-Forgotten Requests
• SoK: Machine Unlearning for Large Language Models
• FlexOlmo: Open Language Models for Flexible Data Use
• Exploring Sparse Adapters for Scalable Merging of Parameter Efficient Experts
• Interaction-Merged Motion Planning: Effectively Leveraging Diverse Motion Datasets for Robust Planning
• Transferring Visual Explainability of Self-Explaining Models to Prediction-Only Models without Additional Training
• Addressing The Devastating Effects Of Single-Task Data Poisoning In Exemplar-Free Continual Learning
• Speaker-agnostic Emotion Vector for Cross-speaker Emotion Intensity Control
• Eigenvoice Synthesis based on Model Editing for Speaker Generation
• CLUES: Collaborative High-Quality Data Selection for LLMs via Training Dynamics
• Prompting as Scientific Inquiry
• TuCo: Measuring the Contribution of Fine-Tuning to Individual Responses of LLMs
• DC-TTA: Divide-and-Conquer Framework for Test-Time Adaptation of Interactive Segmentation
• DuET: Dual Incremental Object Detection via Exemplar-Free Task Arithmetic
• Revisiting the Past: Data Unlearning with Model State History
• Multiple Streams of Knowledge Retrieval: Enriching and Recalling in Transformers
• GPTailor: Large Language Model Pruning Through Layer Cutting and Stitching
• BLUR: A Bi-Level Optimization Approach for LLM Unlearning
• Command-V: Pasting LLM Behaviors via Activation Profiles
• SE-Merging: A Self-Enhanced Approach for Dynamic Model Merging
• CultureMERT: Continual Pre-Training for Cross-Cultural Music Representation Learning
• Latent Concept Disentanglement in Transformer-based Language Models
• Subspace-Boosted Model Merging
• Learning-Time Encoding Shapes Unlearning in LLMs
• From Memorization to Parameter Interference: How Overtraining Experts Harms Model Merging
• SEAT: Sparse Entity-Aware Tuning for Knowledge Adaptation while Preserving Epistemic Abstention
• Knowledge Adaptation as Posterior Correction
• Unlearning Isn't Invisible: Detecting Unlearning Traces in LLMs from Model Outputs
• Transferring Linear Features Across Language Models With Model Stitching
• The Butterfly Effect: Neural Network Training Trajectories Are Highly Sensitive to Initial Conditions
• Reasoning Model Unlearning: Forgetting Traces, Not Just Answers, While Preserving Reasoning Skills
• A correlation-permutation approach for speech-music encoders model merging
• Model Organisms for Emergent Misalignment
• UCD: Unlearning in LLMs via Contrastive Decoding
• Lifting Data-Tracing Machine Unlearning to Knowledge-Tracing for Foundation Models
• Dynamic Mixture of Progressive Parameter-Efficient Expert Library for Lifelong Robot Learning
• Do LLMs Really Forget? Evaluating Unlearning with Knowledge Correlation and Confidence Awareness
• Bring Reason to Vision: Understanding Perception and Reasoning through Model Merging
• StatsMerging: Statistics-Guided Model Merging via Task-Specific Teacher Distillation
• Out-of-Distribution Graph Models Merging
• Adaptive Task Vectors for Large Language Models
• FedRPCA: Enhancing Federated LoRA Aggregation Using Robust PCA
• Assembly of Experts: Linear-time construction of the Chimera LLM variants with emergent and adaptable behaviors
• On Fairness of Task Arithmetic: The Role of Task Vectors
• Towards Unified Modeling in Federated Multi-Task Learning via Subspace Decoupling
• Continual Learning in Vision-Language Models via Aligned Model Merging
• Unlearned but Not Forgotten: Data Extraction after Exact Unlearning in LLM
• Towards Minimizing Feature Drift in Model Merging: Layer-wise Task Vector Fusion for Adaptive Knowledge Integration
• Navigating the Accuracy-Size Trade-Off with Flexible Model Merging
• Two Is Better Than One: Rotations Scale LoRAs
• Decom-Renorm-Merge: Model Merging on the Right Space Improves Multitasking
• Learning Composable Chains-of-Thought
• Update Your Transformer to the Latest Release: Re-Basin of Task Vectors
• Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning
• Multi-objective Large Language Model Alignment with Hierarchical Experts
• Concept Reachability in Diffusion Models: Beyond Dataset Constraints
• Universal Reasoner: A Single, Composable Plug-and-Play Reasoner for Frozen LLMs
• On the Emergence of Linear Analogies in Word Embeddings
• Knowledge Grafting of Large Language Models
• Reward Model Overoptimisation in Iterated RLHF
• CapBencher: Give Your LLM Benchmark a Built-in Alarm for Test-Set Overfitting
• Training-Free Reasoning and Reflection in MLLMs
• When Are Concepts Erased From Diffusion Models?
• CodeMerge: Codebook-Guided Model Merging for Robust Test-Time Adaptation in Autonomous Driving
• DUSK: Do Not Unlearn Shared Knowledge
• Covert Attacks on Machine Learning Training in Passively Secure MPC
• Merge to Mix: Mixing Datasets via Model Merging
• Model Merging is Secretly Certifiable: Non-Vacuous Generalisation Bounds for Low-Shot Learning
• Decouple and Orthogonalize: A Data-Free Framework for LoRA Merging
• SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignment
• Local Mixtures of Experts: Essentially Free Test-Time Training via Model Merging
• Activation-Guided Consensus Merging for Large Language Models
• Safety Subspaces are Not Linearly Distinct: A Fine-Tuning Case Study
• Context-Free Synthetic Data Mitigates Forgetting
• Text Generation Beyond Discrete Token Sampling
• Latent Flow Transformer
• Shadow-FT: Tuning Instruct Model via Training on Paired Base Model
• GUARD: Generation-time LLM Unlearning via Adaptive Restriction and Detection
• Distilling a speech and music encoder with task arithmetic
• Scalable Strategies for Continual Learning with Replay
• Do different prompting methods yield a common task representation in language models?
• MINGLE: Mixture of Null-Space Gated Low-Rank Experts for Test-Time Continual Model Merging
• Parameter Efficient Continual Learning with Dynamic Low-Rank Adaptation
• Model Merging in Pre-training of Large Language Models
• Cross-Model Transfer of Task Vectors via Few-Shot Orthogonal Alignment
• Exploring Criteria of Loss Reweighting to Enhance LLM Unlearning
• MergeBench: A Benchmark for Merging Domain-Specialized LLMs
• Dynamic Base model Shift for Delta Compression
• RanDeS: Randomized Delta Superposition for Multi-Model Compression
• Exploring Nonlinear Pathway in Parameter Space for Machine Unlearning
• A Modular Approach for Clinical SLMs Driven by Synthetic Data with Pre-Instruction Tuning, Model Merging, and Clinical-Tasks Alignment
• Layered Unlearning for Adversarial Relearning
• Trustworthiness Costs of Domain Adaptation in Small Language Models:A Cross-Architecture Empirical Study
• ABRA: Teleporting Fine-Tuned Knowledge Across Domains for Open-Vocabulary Object Detection
• CAT Merging: A Training-Free Approach for Resolving Conflicts in Model Merging
• Noise-Consistent Siamese-Diffusion for Medical Image Synthesis and Segmentation
• WaterDrum: Watermarking for Data-centric Unlearning Metric
• StereoVGGT: A Training-Free Visual Geometry Transformer for Stereo Vision
• Colluding LoRA: A Compositional Vulnerability in LLM Safety Alignment
• Unlearning Sensitive Information in Multimodal LLMs: Benchmark and Attack-Defense Evaluation
• Lightweight Visual Reasoning for Socially-Aware Robots
• STRIDE: Training Data Attribution via Sparse Recovery from Subset Perturbations
• Bus-Conditioned Zero-Shot Trajectory Generation via Task Arithmetic
• GDI-Bench: A Benchmark for General Document Intelligence with Vision and Reasoning Decoupling
• MetaMoE: Diversity-Aware Proxy Selection for Privacy-Preserving Mixture-of-Experts Unification
• UniDetox: Universal Detoxification of Large Language Models via Dataset Distillation
• Weak-to-Strong Generalization via Direct On-Policy Distillation
• DanceOPD: On-Policy Generative Field Distillation
• Neglected Free Lunch from Post-training: Progress Advantage for LLM Agents
• Trait-space Monitoring for Emergent Misalignment During Supervised Finetuning
• Scalable Token-Level Hallucination Detection in Large Language Models
• Adaptive Helpfulness-Harmlessness Alignment with Preference Vectors
• Unified Multi-Task Learning & Model Fusion for Efficient Language Model Guardrailing
• χ₀: Resource-Aware Robust Manipulation via Taming Distributional Inconsistencies
• Controlling Repetition in Protein Language Models
• Per-parameter Task Arithmetic for Unlearning in Large Language Models
• SFT Conflicts, RL Coexists: A Theoretical and Empirical Analysis of Multi-Task Learning for LLMs
• EvolveNet: Collaborative Harness Evolution for Agent Self-Improvement
• When Do PEFT Adaptations Leak Structure? Measuring Black-Box Structural Bounds in Public-Base Model Services
• A Unified Model for Cross-Domain Clone Detection via Model Merging
• MergeSE: Post-Hoc Model Merging for Software Engineering Tasks Without Retraining
• Parameter-Efficient Checkpoint Merging via Metrics-Weighted Averaging
• Exact Unlearning of Finetuning Data via Model Merging at Scale
• MASS: MoErging through Adaptive Subspace Selection
• Advancing AI-assisted Hardware Design with Hierarchical Decentralized Training and Personalized Inference-Time Optimization
• Mitigating Parameter Interference in Model Merging via Sharpness-Aware Fine-Tuning
• TrustLoRA: Low-Rank Adaptation for Failure Detection under Out-of-distribution Data
• DMM: Building a Versatile Image Generation Model via Distillation-Based Model Merging
• CircuitSteer: Geometrically Aligned Multi-Layer Steering via Sparse Autoencoder Circuits
• Single-Input Multi-Output Model Merging: Leveraging Foundation Models for Dense Multi-Task Learning
• When is Task Vector Provably Effective for Model Editing? A Generalization Analysis of Nonlinear Transformers
• Leveraging Submodule Linearity Enhances Task Arithmetic Performance in LLMs
• LLM Unlearning Reveals a Stronger-Than-Expected Coreset Effect in Current Benchmarks
• Alleviating the Fear of Losing Alignment in LLM Fine-tuning
• LoRI: Reducing Cross-Task Interference in Multi-Task Low-Rank Adaptation
• FedMerge: Federated Personalization via Model Merging
• SEA-LION: Southeast Asian Languages in One Network

Discussions

• Editing Models with Task Arithmetic (2022) [hn, 1 points, 0 comments]

Related

• Decomposing Task Vectors for Refined Model Editing
• When is Task Vector Provably Effective for Model Editing? A Generalization Analysis of Nonlinear Transformers
• Efficient Model Editing with Task-Localized Sparse Fine-tuning
• Escaping Optimization Stagnation: Taking Steps Beyond Task Arithmetic via Difference Vectors
• On Task Vectors and Gradients