Tag: Mixed Precision

All the papers with the tag "Mixed Precision".

• Pack-PTQ: Advancing Post-training Quantization of Neural Networks by Pack-wise Reconstruction

  grok-3-latest

  Score: 0.48

  Published:2025年5月1日 at  at 02:53

  #Neural Network, #Quantization, #Post-Training, #Mixed Precision, #Cross-Block Dependency

  本文提出 Pack-PTQ 方法，通过 Hessian-guided 打包机制和包级混合精度量化策略，显著提升低比特后训练量化的性能，同时捕捉跨块依赖性。