Q-BERT: Hessian Based Ultra Low Precision Quantization of BERT---Q-BERT：基于Hessian的超低精度BERT...

论文标题：Q-BERT: Hessian Based Ultra Low Precision Quantization of BERT---Q-BERT：基于Hessian的超低精度BERT量化

• Abstract
• 1 Related Work
• • Model compression模型压缩
  • Compressed NLP model
• 2 Methodology
• • 2.1 Quantization process
  • 2.2 Mixed precision quantization
  • 2.3 Group-wise Quantization
• 3 Experiment
• • 3.1 Main Results
  • 3.2 Effects of group-wise quantization--群体量化效果
• 4