Search papers, labs, and topics across Lattice.
1
0
3
LLMs can get a free performance boost: decoupling compute and capacity within each layer lets you beat standard transformers at the same FLOPs.