Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
LLMs can achieve superior KV cache compression by intelligently allocating quantization budgets between layer-wise and token-wise strategies, outperforming uniform or single-granularity approaches.
