Search papers, labs, and topics across Lattice.
K, with 70% token merging ratio on Stable Diffusion 2.0, BiGain increases classification accuracy by 7.15% while also improving FID for generation by 0.34 (1.85%). Our comprehensive analyses indicate that balanced spectral retention, preserving high-frequency detail alongside low/mid-frequency semantic content is a reliable design rule for token compression in diffusion models. To our knowledge, BiGain is the first framework to jointly study and advance both generation and classification under accelerated diffusion, supporting lower-cost deployment of dual-purpose generative systems. **footnotetext: Equal contribution.
2
0
5
3
Token compression in diffusion models no longer has to sacrifice classification accuracy for faster generation – BiGain boosts both.
Attention sinks, considered essential in autoregressive language models, turn out to be surprisingly prunable in diffusion language models, leading to better efficiency.
