An Empirical Study and Theoretical Explanation on Task-Level Model-Merging Collapse
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866910047512035328 |
|---|---|
| author | Cao, Yuan Ran, Dezhi Guo, Yuzhe Wu, Mengzhou Chen, Simin Li, Linyi Yang, Wei Xie, Tao |
| author_facet | Cao, Yuan Ran, Dezhi Guo, Yuzhe Wu, Mengzhou Chen, Simin Li, Linyi Yang, Wei Xie, Tao |
| contents | Model merging unifies independently fine-tuned LLMs from the same base, enabling reuse and integration of parallel development efforts without retraining. However, in practice we observe that merging does not always succeed: certain combinations of task-specialist models suffer from catastrophic performance degradation after merging. We refer to this failure mode as merging collapse. Intuitively, collapse arises when the learned representations or parameter adjustments for different tasks are fundamentally incompatible, so that merging forces destructive interference rather than synergy. In this paper, we identify and characterize the phenomenon of task-level merging collapse, where certain task combinations consistently trigger huge performance degradation across all merging methods. Through extensive experiments and statistical analysis, we demonstrate that representational incompatibility between tasks is strongly correlated with merging collapse, while parameter-space conflict metrics show minimal correlation, challenging conventional wisdom in model merging literature. We provide a theoretical explanation on this phenomenon through rate-distortion theory with a dimension-dependent bound, establishing fundamental limits on task mergeability regardless of methodology. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_09463 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | An Empirical Study and Theoretical Explanation on Task-Level Model-Merging Collapse Cao, Yuan Ran, Dezhi Guo, Yuzhe Wu, Mengzhou Chen, Simin Li, Linyi Yang, Wei Xie, Tao Artificial Intelligence Model merging unifies independently fine-tuned LLMs from the same base, enabling reuse and integration of parallel development efforts without retraining. However, in practice we observe that merging does not always succeed: certain combinations of task-specialist models suffer from catastrophic performance degradation after merging. We refer to this failure mode as merging collapse. Intuitively, collapse arises when the learned representations or parameter adjustments for different tasks are fundamentally incompatible, so that merging forces destructive interference rather than synergy. In this paper, we identify and characterize the phenomenon of task-level merging collapse, where certain task combinations consistently trigger huge performance degradation across all merging methods. Through extensive experiments and statistical analysis, we demonstrate that representational incompatibility between tasks is strongly correlated with merging collapse, while parameter-space conflict metrics show minimal correlation, challenging conventional wisdom in model merging literature. We provide a theoretical explanation on this phenomenon through rate-distortion theory with a dimension-dependent bound, establishing fundamental limits on task mergeability regardless of methodology. |
| title | An Empirical Study and Theoretical Explanation on Task-Level Model-Merging Collapse |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2603.09463 |