arXivSoosung Kim, Minjae Park, Eui-Young Chung, Jaeyong ChungWed, Jul 1, 2026, 8:25 AM PDT
score 17.0
New quantization method shrinks AI model memory below 1 bit per token
Original: GSRQ: Gain-Shape Residual Quantization for Sub-1-bit KV Cache
Source: arxiv.org ↗
Writing ELI5 summary…