Agrim Gupta

PhD student, Stanford University

1 paper at NeurIPS 2025

OpenReview· Semantic Scholar· Google Scholar

Poster Session 4

Thursday, December 4, 2025 · 4:30 PM → 7:30 PM

Exploring Diffusion Transformer Designs via Grafting

#3601 · Keshigeyan Chandrasegaran, Michael Poli, Daniel Y Fu, Dongjun Kim, Lea M. Hadzic, Manling Li, Agrim Gupta, Stefano Massaroli, Azalia Mirhoseini, Juan Carlos Niebles, Stefano Ermon, Li Fei-Fei

We propose grafting, a simple approach to materialize new architectures by editing pretrained diffusion transformers. It enables architectural exploration under small compute budgets.