Search papers, labs, and topics across Lattice.
IIT Bombay
1
0
2
8
MAESTRO prunes MoE models more effectively by leveraging the interdependencies of expert routing, achieving superior performance retention and consistency across tasks.