Search papers, labs, and topics across Lattice.
1
0
2
FIBER achieves up to 2.25x speedup in LLM serving by decoupling tensor computation from register ownership, revolutionizing GPU efficiency.
