GPTAQ Neural Network Quantization Framework based on GPTQ With addition of: Activations quantization (RTN + weight reoptimization + Token-wise) Hessian Eigenvalues in sensitivity params Cross-layer equalization Algorithm Experiments