Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
Pruning LLMs can amplify biases, but Debias-SparseGPT shows that you can reduce these biases while maintaining performance, even under aggressive sparsity constraints.
