Search papers, labs, and topics across Lattice.
1
0
3
Ditch the CUDA boilerplate: TiledAttention lets you rapidly prototype and tweak custom attention kernels directly from Python, unlocking faster iteration on novel SDPA architectures.