Search papers, labs, and topics across Lattice.
3
5
3
5
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.
Forget scaling compute – the future of AI hinges on a 1000x leap in energy efficiency via tight AI+Hardware co-design over the next decade.
FlashAttention-4 shatters attention bottlenecks on Blackwell GPUs, achieving up to 71% hardware utilization and 2.7x speedups over Triton, thanks to innovations like software-emulated softmax and asynchronous MMA pipelines.
