1 paper · 1 filter
Tianyu Wang, Ying Wang, Zhihao Liu +2
Sequential output generation with large-scale Transformer and diffusion decoders pays a memory cost that grows with sequence length, plus iterative per-step computation. Replacing…