Research map / Image generation / Autoregressive and latent generation
FastVAR: Linear Visual Autoregressive Modeling via Cached Token Pruning
Hang Guo, Yawei Li, Taolin Zhang, Jiangshan Wang, Tao Dai, Shu-Tao Xia, Luca Benini
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 FastVAR cached token pruning | HART · GenEval 1024x1024 | speedup / Overall: 1.0x -> 1.5x, 0.51 -> 0.51 | Improved |
| add FastVAR cached token pruning | Infinity · GenEval 1024x1024 | speedup / Overall: 2.7x speedup, 0.73 -> 0.72 | Improved |
| add FastVAR cached token pruning | HART · MJHQ30K people | FID: 30.61 -> 28.19 | Improved |
| add FastVAR cached token pruning | Infinity · MJHQ30K food | FID: 31.55 -> 32.54 | Hurt |
| replace ToMe token merging with FastVAR pruning (ablation) | HART · MJHQ30K FID | FID: 35.22 -> 28.19 | Improved |
| prune tokens at small scales [16,21] (75%) instead of large scales [48,64] (ablation) | HART · MJHQ30K | FID: 28.56 -> 34.27 | Hurt |
| prune last scales [48,64] at 75% instead of [27,36] (ablation) | HART · Speedup | speedup: 1.14x -> 1.64x | Improved |
| use cache from step K-N instead of K-N-3 (ablation) | HART · MJHQ30K | FID: 40.61 -> 28.19 | Improved |
| raise last-scale pruning ratio from 75% to 100% on HART (ablation) | HART · MJHQ30K | FID: 28.19 -> 48.54 | Hurt |
| combine FastVAR with FlashAttention instead of FastVAR alone (ablation) | Infinity · 1024x1024 efficiency | speedup: 2.1x -> 2.7x | Improved |
| use FastVAR on a smaller class-conditional VAR (d=24) on ImageNet | VAR d=24 · ImageNet 256x256 | FID: 2.29 -> 2.64 | Hurt |
| increase pruning ratio (ablation) | HART · MJHQ30K | FID | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Infinity — Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis (CVPR 2025) · better on 5 of 6 shared results
- HART — HART: Efficient Visual Generation with Hybrid Autoregressive Transformer (ICLR 2025) · better on 5 of 6 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
- FasterVAR: Plug-and-Play Acceleration for Visual Autoregressive Models (ICML 2026 · similarity 0.93)
- LazyVAR: Accelerating Visual Autoregressive Models via Scale-wise Token Pruning and Parallel Group Decoding (CVPR 2026 · similarity 0.92)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.90)
- FlexVAR: Flexible Visual Autoregressive Modeling without Residual Prediction (NeurIPS 2025 · similarity 0.89)
- Memory-Efficient Visual Autoregressive Modeling with Scale-Aware KV Cache Compression (NeurIPS 2025 · similarity 0.89)
- Parallelized Autoregressive Visual Generation (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.
