Search papers, labs, and topics across Lattice.
1
0
2
7
A novel method reveals that the weights of LoRA fine-tuned models can directly identify harmful training content, bypassing the need for risky output generation.
