EdgeDiT: Hardware-Aware Diffusion Transformers for Efficient On-Device Image Generation
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917369189761024 |
|---|---|
| author | Kodavanti, Sravanth Arveti, Manjunath Vajrala, Sowmya Miriyala, Srinivas R, Vikram N |
| author_facet | Kodavanti, Sravanth Arveti, Manjunath Vajrala, Sowmya Miriyala, Srinivas R, Vikram N |
| contents | Diffusion Transformers (DiT) have established a new state-of-the-art in high-fidelity image synthesis; however, their massive computational complexity and memory requirements hinder local deployment on resource-constrained edge devices. In this paper, we introduce EdgeDiT, a family of hardware-efficient generative transformers specifically engineered for mobile Neural Processing Units (NPUs), such as the Qualcomm Hexagon and Apple Neural Engine (ANE). By leveraging a hardware-aware optimization framework, we systematically identify and prune structural redundancies within the DiT backbone that are particularly taxing for mobile data-flows. Our approach yields a series of lightweight models that achieve a 20-30% reduction in parameters, a 36-46% decrease in FLOPs, and a 1.65-fold reduction in on-device latency without sacrificing the scaling advantages or the expressive capacity of the original transformer architecture. Extensive benchmarking demonstrates that EdgeDiT offers a superior Pareto-optimal trade-off between Frechet Inception Distance (FID) and inference latency compared to both optimized mobile U-Nets and vanilla DiT variants. By enabling responsive, private, and offline generative AI directly on-device, EdgeDiT provides a scalable blueprint for transitioning large-scale foundation models from high-end GPUs to the palm of the user. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_28405 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | EdgeDiT: Hardware-Aware Diffusion Transformers for Efficient On-Device Image Generation Kodavanti, Sravanth Arveti, Manjunath Vajrala, Sowmya Miriyala, Srinivas R, Vikram N Computer Vision and Pattern Recognition Artificial Intelligence Diffusion Transformers (DiT) have established a new state-of-the-art in high-fidelity image synthesis; however, their massive computational complexity and memory requirements hinder local deployment on resource-constrained edge devices. In this paper, we introduce EdgeDiT, a family of hardware-efficient generative transformers specifically engineered for mobile Neural Processing Units (NPUs), such as the Qualcomm Hexagon and Apple Neural Engine (ANE). By leveraging a hardware-aware optimization framework, we systematically identify and prune structural redundancies within the DiT backbone that are particularly taxing for mobile data-flows. Our approach yields a series of lightweight models that achieve a 20-30% reduction in parameters, a 36-46% decrease in FLOPs, and a 1.65-fold reduction in on-device latency without sacrificing the scaling advantages or the expressive capacity of the original transformer architecture. Extensive benchmarking demonstrates that EdgeDiT offers a superior Pareto-optimal trade-off between Frechet Inception Distance (FID) and inference latency compared to both optimized mobile U-Nets and vanilla DiT variants. By enabling responsive, private, and offline generative AI directly on-device, EdgeDiT provides a scalable blueprint for transitioning large-scale foundation models from high-end GPUs to the palm of the user. |
| title | EdgeDiT: Hardware-Aware Diffusion Transformers for Efficient On-Device Image Generation |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2603.28405 |