Compression method shrinks memory of looped AI models

Original: LoopVQ: Sub 1-bit KV Cache Quantization for Looped Transformers

Source: arxiv.org ↗