Research map / Architectures and efficiency / Quantization
Shift-and-Sum Quantization for Visual Autoregressive Models
Jaehyeon Moon, Bumsub Ham
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add shift-and-sum quantization (ablation) | BRECQ on VAR-d16 · ImageNet W4/A6 | FID: 11.16 -> 10.15 | Improved |
| add calibration data resampling (ablation) | BRECQ on VAR-d16 · ImageNet W4/A6 | FID: 11.16 -> 10.19 | Improved |
| add both shift-and-sum and resampling (ablation) | BRECQ on VAR-d16 · ImageNet W4/A6 | FID: 11.16 -> 9.20 | Improved |
| add shift-and-sum quantization (ablation) | BRECQ on VAR-d20 · ImageNet W4/A6 | FID: 6.74 -> 6.10 | Improved |
| add calibration data resampling (ablation) | BRECQ on VAR-d20 · ImageNet W4/A6 | FID: 6.74 -> 5.86 | Improved |
| add both components (ablation) | BRECQ on VAR-d20 · ImageNet W4/A6 | FID: 6.74 -> 5.36 | Improved |
| raise attention threshold theta from 0.05 to 0.2 (fewer shifted tokens) (ablation) | Shift-and-sum VAR-d16 · ImageNet W4/A6 | FID: 10.15 -> 10.33 | Mixed |
| combine Ours with LiteVAR | Ours on VAR-d16 · ImageNet W4/A4 | FID: 24.57 -> 18.92 | Improved |
| combine Ours with LiteVAR | Ours on Infinity-2B · Text-to-image W4/A4 | ImageReward: 0.575 -> 0.748 | Improved |
| use Ours instead of BRECQ | VAR-d30 · ImageNet W4/A4 | FID: 15.71 -> 9.26 | Improved |
| use Ours instead of BRECQ at 8-bit | VAR-d16 ONNX CPU · throughput | image/sec: 0.2195 -> 0.2147 | No clear effect |
| use Ours instead of BRECQ at 6/6 bits | VAR-d30 · inpainting/outpainting/editing | LPIPS: 0.2852 -> 0.2843 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BRECQ — BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstruction (ICLR 2021) · better on 7 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VLM-PTQ: Efficient Post-Training Quantization for Large Vision-Language Models (CVPR 2026 · similarity 0.88)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.88)
- Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Generation (NeurIPS 2025 · similarity 0.88)
- Channel-wise Vector Quantization (NeurIPS 2026 · similarity 0.88)
- PermuQuant: Lowering Per-Group Quantization Error by Reordering Channels for Diffusion Models (NeurIPS 2026 · similarity 0.88)
- MBQ: Modality-Balanced Quantization for Large Vision-Language Models (CVPR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
