Search papers, labs, and topics across Lattice.
1
0
2
Quantization error isn't just about concentrating weights, it's also about aligning them with activations, and a simple transform can exploit this to boost 4-bit LLM performance.
