Integer or Floating Point? New Outlooks for Low-Bit Quantization on Large Language Models

Open in new window