Logarithmic quantization makes language models run faster on consumer GPUs

Original: $\text{Log}_\text{b}$Quant: Quantizing Language Models in Logarithmic Space

Source: arxiv.org ↗