Chinese contentLangDiT：为连续扩散语言模型重新设计的 Transformer 架构
扩散语言模型一直借用图像那套 DiT 架构，这篇论文专门为文本重设计了 LangDiT，GenPPL 降到 11.33，做生成模型方向的可以看看架构细节。
#GenPPL
2 stories · 1 in 7 days · 1 in 30 days
Stories, products and papers tagged "GenPPL", newest first.
Oct 16
Sep 2
Committed Reveal Sampling for Diffusion Models
CRS improves text generation quality without additional training, offering a practical method for better perplexity scores across different computational budgets.