Pith. sign in

REVIEW 7 cited by

Multi-Task Model Merging via Adaptive Weight Disentanglement

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2411.18729 v2 pith:4FSB426Z submitted 2024-11-27 cs.LG cs.CLcs.CV

classification cs.LGcs.CLcs.CV
keywords taskvectorsmodelperformancetasksmergingredundantadaptive
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Model merging has recently gained attention as an economical and scalable approach to incorporate task-specific weights from various tasks into a unified multi-task model. For example, in Task Arithmetic (TA), adding the fine-tuned weights of different tasks can enhance the model's performance on those tasks, while subtracting them leads to task forgetting. Although TA is highly effective, interference among task still hampers the performance of the merged model. Existing methods for handling conflicts between task generally rely on empirical selection, resulting in suboptimal performance. In this paper, we introduce an Adaptive Weight Disentanglement method. We begin by theoretically proving that task vectors employed in model merging should be orthogonal to minimize interference among tasks. Guided by this insight, we initialize redundant vectors such that, when subtracted from the original task vectors, the resulting vectors exhibit increased orthogonality. Additionally, we impose an norm constraint on the redundant vectors to preserve the performance of the task-specific models. Experimental results demonstrate the effectiveness of our proposed technique: it successfully extracts redundant vectors, and after their subtraction, the task vectors not only retain robust performance but also achieve superior fusion outcomes. Our code is available at \href{https://github.com/FarisXiong/AWD.git}{https://github.com/FarisXiong/AWD.git}.

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Understanding and Enforcing Weight Disentanglement in Task Arithmetic

    cs.AI 2026-04 unverdicted novelty 7.0 of 10

    Task-Feature Specialization explains weight disentanglement in task arithmetic and leads to orthogonality, which OrthoReg enforces to enhance performance of model composition methods.

  2. SAFE-Merge: Data-Free Continual Model Merging with General Knowledge Preservation

    cs.LG 2026-08 conditional novelty 6.0 of 10

    SAFE-Merge masks risk-prone parameter updates and recovers lost task information with a constrained low-rank correction, achieving the best H-score in data-free continual model merging benchmarks.

  3. Learning to Recover Task Experts from a Multi-Task Merged Model

    cs.AI 2026-06 unverdicted novelty 6.0 of 10

    ReTeX predicts additive offsets to undo merging interference and uses an SVD subspace signature task identifier to recover over 95% of expert performance while improving generalization to unseen tasks.

  4. PACT: Preserving Anchored Cores in Task-vectors for Model Merging

    cs.LG 2026-06 unverdicted novelty 6.0 of 10

    PACT preserves load-bearing wall dimensions from pre-trained weights inside task vectors to reduce conflicts and improve merged model performance.

  5. When Model Merging Breaks Routing: Training-Free Calibration for MoE

    cs.LG 2026-06 unverdicted novelty 6.0 of 10

    Merging breaks MoE routing via softmax sensitivity; HARC uses Hessian curvature for closed-form router calibration that improves merged model performance without retraining.

  6. Memory Grafting: Scaling Language Model Pre-training via Offline Conditional Memory

    cs.CL 2026-05 unverdicted novelty 6.0 of 10

    Memory Grafting improves language-model benchmarks by grafting offline hidden-state memory from a larger model into a recipient model using n-gram lookups and lightweight adapters, outperforming MoE and vanilla Engram...

  7. Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunities

    cs.LG 2024-08 accept novelty 4.0 of 10

    The paper introduces a new taxonomy for model merging methods and reviews their applications in LLMs, MLLMs, continual learning, multi-task learning, and other subfields while outlining open challenges.

Pith tools