GPTAQ: Efficient Finetuning-Free Quantization for Asymmetric Calibration
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:C2L22HAZrecord.jsonopen to challenge →
read the original abstract
We introduce GPTAQ, a novel finetuning-free quantization method for compressing large-scale transformer architectures. Unlike the previous GPTQ method, which independently calibrates each layer, we always match the quantized layer's output to the exact output in the full-precision model, resulting in a scheme that we call asymmetric calibration. Such a scheme can effectively reduce the quantization error accumulated in previous layers. We analyze this problem using optimal brain compression to derive a close-formed solution. The new solution explicitly minimizes the quantization error as well as the accumulated asymmetry error. Furthermore, we utilize various techniques to parallelize the solution calculation, including channel parallelization, neuron decomposition, and Cholesky reformulation for matrix fusion. As a result, GPTAQ is easy to implement, simply using 20 more lines of code than GPTQ but improving its performance under low-bit quantization. Remarkably, on a single GPU, we quantize a 405B language transformer as well as EVA-02, the rank first vision transformer that achieves 90% pretraining Imagenet accuracy. Code is available at Github.
This paper has not been read by Pith yet.
Forward citations
Cited by 5 Pith papers
-
Quantizing Time-Series Models As Dynamical Systems: Trajectory-Based Quantization Sensitivity Score
Introduces TQS metric and TQS-PTQ framework that uses dynamical-systems stability to enable a priori, calibration-free mixed-precision post-training quantization for time-series models.
-
Qift: Shift-Friendly No-Zero W2 Post-Training Quantization for Rotated W2A4/KV4 LLM Inference
Qift defines a fixed no-zero W2 level set for rotated weights that improves W2A4 perplexity and accuracy on LLaMA-2-7B and LLaMA-3.1-8B over the standard {-2,-1,0,1} set.
-
Rethinking Residual Errors in Compensation-based LLM Quantization
Redefining residual errors to include compensation-aware discrepancies and realigning calibration to full-precision outputs improves GPTQ and GPTAQ performance on LLMs.
-
MARR: Module-Adaptive Residual Reconstruction for Low-Bit Post-Training Quantization
MARR uses per-module adaptive residual scaling updated by PID feedback to balance error correction against Hessian-approximation bias in low-bit PTQ.
-
CAR-SAM: Cross-Attention Reconstruction for Post-Training Quantization of the Segment Anything Model
CAR-SAM introduces MatMul-Aware Compensation and Joint Cross-Attention Reconstruction to enable stable 4-bit post-training quantization of SAM, outperforming prior PTQ methods by 14.6% mAP on SAM-B and 6.6% on SAM-L.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.