Search papers, labs, and topics across Lattice.
1
0
2
Untied embeddings, not weight tying, unlock superior performance and memory efficiency for decoder-only LLMs under DP-SGD.