Research map / Architectures and efficiency / Quantization
PTQ4ARVG: Post-Training Quantization for AutoRegressive Visual Generation Models
Xuewen Liu, Zhikai Li, Jing Zhang, Mengjuan Chen, Jianquan Li, Qingyi Gu
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 GPS scaling (ablation) | SmoothQuant baseline · RAR-B W6A6 | FID: 63.77 -> 36.51 | Improved |
| add STWQ on top of GPS (ablation) | SmoothQuant+GPS · RAR-B W6A6 | FID: 36.51 -> 6.67 | Improved |
| add DGC calibration on top of GPS and STWQ (ablation) | PTQ4ARVG · RAR-B W6A6 | FID: 6.67 -> 5.13 | Improved |
| add DGC on top of GPS and STWQ (ablation) | PTQ4ARVG · VAR-d16 W6A6 | FID: 8.75 -> 8.34 | Improved |
| use GPS instead of SmoothQuant scaling (ablation) | PTQ4ARVG · RAR-B W6A6 | FID: 10.26 -> 5.13 | Improved |
| use GPS instead of RepQ* scaling (ablation) | PTQ4ARVG · RAR-B W6A6 | FID: 33.79 -> 5.13 | Improved |
| use GPS instead of OS+ scaling (ablation) | PTQ4ARVG · RAR-B W6A6 | FID: 7.71 -> 5.13 | Improved |
| use GPS instead of SmoothQuant then RepQ* scaling (ablation) | PTQ4ARVG · RAR-B W6A6 | FID: 7.43 -> 5.13 | Improved |
| use static instead of dynamic token-wise quantization (ablation) | SmoothQuant · VAR-d16 W6A6 | FID: 30.14 -> 10.41 | Improved |
| add dynamic token-wise quantization (ablation) | SmoothQuant · VAR-d16 W6A6 | FID: 18.54 -> 30.14 | Hurt |
| add static token-wise quantization (ablation) | SmoothQuant · VAR-d16 W6A6 | FID: 18.54 -> 10.41 | Improved |
| use static instead of dynamic token-wise quantization (ablation) | SmoothQuant 8-bit deployment · VAR-d16 batch 100 | speedup: 2.457x -> 2.922x | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- QuaRot — QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMs (NeurIPS 2024) · better on 5 of 5 shared results
- OmniQuant — OmniQuant: Omnidirectionally Calibrated Quantization for Large Language Models (ICLR 2024) · better on 4 of 4 shared results
- SmoothQuant — SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models (ICML 2023) · better on 2 of 2 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
- PermuQuant: Lowering Per-Group Quantization Error by Reordering Channels for Diffusion Models (NeurIPS 2026 · similarity 0.90)
- Q-ARVD: Quantizing Autoregressive Video Diffusion Models (NeurIPS 2026 · similarity 0.90)
- Understand Before You Generate: Self-Guided Training for Autoregressive Image Generation (NeurIPS 2025 · similarity 0.88)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.88)
- Channel-wise Vector Quantization (NeurIPS 2026 · similarity 0.87)
- TTS-VAR: A Test-Time Scaling Framework for Visual Auto-Regressive Generation (NeurIPS 2025 · similarity 0.87)
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.
