Loopback Transformer beats baselines with fewer parameters
On ClimbMix it uses 40-75% of baseline Transformer parameters and posts higher mean accuracy at 1.4B and 3B.
3 stories, newest first · one card per story
On ClimbMix it uses 40-75% of baseline Transformer parameters and posts higher mean accuracy at 1.4B and 3B.
Block-wise grouping with wider residuals and learned projections improves FLOPs efficiency while preserving standard layer structure.
Structured residual paths in diffusion transformers cut FID and training iterations with almost no extra parameters.
Today 1.2K readers · 1.4K views · 193 reads · all time 203K views