Formalising the Logit Shift Induced by LoRA: A Technical Note
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866913053944053760 |
|---|---|
| author | Shi, Xiang Cheng, Shuaizhi Li, Mingwei |
| author_facet | Shi, Xiang Cheng, Shuaizhi Li, Mingwei |
| contents | This technical note provides a first-order formalisation of the logit shift and fact-margin change induced by Low-Rank Adaptation (LoRA). Using a first-order Fréchet approximation around the base model trajectory, we show that the multi-layer LoRA effect can be decomposed into a linear summation of layerwise contributions and a higher-order remainder term representing inter-layer coupling. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_20313 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Formalising the Logit Shift Induced by LoRA: A Technical Note Shi, Xiang Cheng, Shuaizhi Li, Mingwei Machine Learning Artificial Intelligence This technical note provides a first-order formalisation of the logit shift and fact-margin change induced by Low-Rank Adaptation (LoRA). Using a first-order Fréchet approximation around the base model trajectory, we show that the multi-layer LoRA effect can be decomposed into a linear summation of layerwise contributions and a higher-order remainder term representing inter-layer coupling. |
| title | Formalising the Logit Shift Induced by LoRA: A Technical Note |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2604.20313 |