New method cuts AI model size without hurting accuracy

Original: Q-Strata: Hierarchical Bit Allocation for Mixed-Precision Quantization of Mixture-of-Experts LLMs

Source: arxiv.org ↗