Search papers, labs, and topics across Lattice.
1
5
0
7
State-space models can achieve $1.8\times$ speculative decoding throughput and $1.47\times$ faster standard decoding by caching SSM inputs instead of their full state.
