Search papers, labs, and topics across Lattice.
1
0
3
Halving the parameter count of LLMs without sacrificing performance is now possible with Hyperloop Transformers, thanks to looped layers and hyper-connected residual streams.
