[KongchangAI]
ModelUUE的低秩变体

UUE-L

A low-rank variant of the UUE framework that uses low-rank decomposition to compress adapter parameter counts, reducing storage and computation overhead.

Timeline (last 90 days)

Nov 16

UUE-L is the low-rank variant of UUE that further compresses adapter parameters and suits scenarios requiring maintenance of multiple unlearning modules.

Unverified50%
Nov 16

Experiments on the TOFU and WMDP benchmarks show that UUE and UUE-L significantly outperform existing methods in unlearning effectiveness.

Unverified50%

All Facts (2)

Source Articles