act_order) vs. Salient-Channel AWQEmergent activation outliers in Transformer blocks (20x+ magnitude in <3% of channels) cause catastrophic output MSE degradation under naive Round-to-Nearest (RTN) 4-bit quantization.
We implement both MLSys 2024 Best Paper AWQ (grid-searched per-channel equivalence scaling $s = \bar{|X|}^\gamma$) and ICLR 2023 GPTQ using exact float64 Cholesky factorization of $H^{-1} = R^\top R$ paired with descending Hessian curvature ordering (act_order=True).
AWQ reduces 4-bit output MSE by 4.54x (+6.57 dB SQNR) via hardware-friendly diagonal scaling.GPTQ (Cholesky H^-1 + act_order) reduces 4-bit output MSE by 25.49x (21.22 dB -> 35.28 dB SQNR) by actively compensating quantization error across correlated unquantized columns.