Research map / Image generation / Fast diffusion sampling
Scaling Laws for Diffusion Transformers
Zhengyang Liang, Hao He, Ceyuan Yang, Bo Dai
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 |
|---|---|---|---|
| replace in-context conditioning with cross-attention conditioning | DiT scaling law · Scaling exponents (loss vs compute) | loss exponent: -0.0273 -> -0.0385 | Improved |
| replace in-context conditioning with cross-attention conditioning | DiT scaling law · Scaling exponents (allocation) | model exponent / data exponent: model 0.56 -> 0.54, data 0.43 -> 0.46 | Mixed |
| mix models of aspect ratio 32 and 64 instead of fixing ratio at 64 (ablation) | isoFLOP model fitting · Loss vs parameters at fixed compute | trend clarity | Hurt |
| evaluate scaling on out-of-domain COCO 2014 validation instead of in-domain data | DiT scaling law · COCO 2014 val | validation loss, VLB, likelihood, FID | No clear effect |
| extrapolate fitted law to 1.5e21 FLOPs with 958.3M-parameter model | DiT scaling law · Laion-Aesthetic subset | training loss | Improved |
Closest papers by meaning
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.92)
- Scaling Diffusion Transformers Efficiently via $\mu$P (NeurIPS 2025 · similarity 0.91)
- FasterDiT: Towards Faster Diffusion Transformers Training without Architecture Modification (NeurIPS 2024 · similarity 0.90)
- Dynamic Diffusion Transformer (ICLR 2025 · similarity 0.89)
- Layer- and Timestep-Adaptive Differentiable Token Compression Ratios for Efficient Diffusion Transformers (CVPR 2025 · similarity 0.89)
- Exploring Diffusion Transformer Designs via Grafting (NeurIPS 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.
