AI Paper Atlas

Research map / Architectures and efficiency / Quantization

FourTune: Towards Fully 4-Bit Efficient Post-Training for Diffusion Models

Bowen Xue, Zihan Min, Xingyang Li, Zhekai Zhang, Haocheng Xi, Lvmin Zhang, Maneesh Agrawala, Jun-Yan Zhu, Song Han, Yujun Lin, Muyang Li

ICML 2026 · Architectures and efficiency › Quantization · approach: Post-training quantization of ViTs · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
quantize weights, activations and gradients to 4 bit (W4A4G4) instead of BF16 LoRAFourTune · FLUX.1-dev customization, end-to-endmemory / speed: 29.35 -> 14.51 GB; 2.27x speedup on RTX 5090Improved
W4A4G4 vs NF4 QLoRA training speedFourTune · FLUX.1-dev customization, RTX 5090speedup: 2.79xImproved
add frozen numerical stabilizer branch to naive W4A4G4 (ablation)FourTune · knowledge distillation gradient normgradient stabilityImproved
replace group-wise 1x16 with block-wise 16x16 quantization (ablation)FourTune · FLUX.1-dev reconstruction, MJHQ-30KLPIPS / PSNR: LPIPS 0.203 -> 0.227; PSNR 21.5 -> 20.4Mixed
lower activation and gradient precision from 16 to 4 bit (W4A16G16 -> W4A4G4) (ablation)FourTune · RL on FLUX.1-dev, Aesthetic ScoreAes: 6.0816 -> 6.1779No clear effect
fuse LoRA-related kernels vs unfused 4-bit baseline (ablation)FourTune · DiT training step latency RTX 5090speedup: 1.82xImproved
additionally fuse MLP components (ablation)FourTune · DiT training step latency RTX 5090speedup: 1.10x additionalImproved
use SparseLoRA instead of W4A4G4 for 4-step distillationSparseLoRA · pi-Flow distillation, HPSv2 promptsFID: 15.50 -> 202.51Hurt
W4A4G4 vs BF16 LoRA on customizationFourTune · Customization similarity, FLUX.1-devsimilarity: 0.771 -> 0.783 (Identity)No clear effect
W4A4G4 vs BF16 full finetune on RLFourTune · SRPO RL, FLUX.1-devAes: 6.3447 -> 6.3119No clear effect
W4A4G4 vs BF16 LoRA on Qwen-Image Subject customizationFourTune · Customization, Qwen-Imageprompt following: 0.999 -> 0.983Mixed

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy