Search papers, labs, and topics across Lattice.
1
0
3
2
Forget slow rotations: IsoQuant's quaternion-based approach warps RotorQuant in LLM KV cache compression, delivering up to 6x speedups on synthetic data.
