Search papers, labs, and topics across Lattice.
1
0
3
Diffusion Transformers waste up to 66% of their conditional embedding space without sacrificing generation quality, hinting at opportunities for more efficient conditioning.
