Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
Most LLM unlearning dissolves the moment a model is compressed for production, but isolating updates to high-significance layers prevents forgotten data from resurfacing under 4-bit quantization.
