Nearly 9x training speedup: residual streams in DiT are becoming a convergence bottleneck

近9倍训练加速：DiT里的残差流，正在成为收敛瓶颈

Nanjing University LAMDA and Alibaba Intelligent Engine proposed DAR, a timestep-aware cross-layer routing method that replaces fixed residual accumulation in DiT; on ImageNet 256x256, it reduced SiT-XL/2 FID from 9.67 to 7.56 and reached baseline convergence quality with 8.75x fewer training iterations.

Why it matters: HKR-H/K/R all pass, but the topic is a narrow DiT training method rather than a broad model or product launch. Concrete ImageNet metrics and the Alibaba/LAMDA mechanism clear the featured bar, not the 78+ band.