0cde695b83bd186c1fd456302888454c-Supplemental-Conference.pdf
–Neural Information Processing Systems
LoRA[13] We use a rank of4 with initialization scale of0.01 and update all the attention and feedforwardmodule. Table 4: Per-dataset accuracies for the PEFT methods we consider when addingLUL and LLN.
Neural Information Processing Systems
Feb-7-2026, 10:53:38 GMT