Radar ·
LiFT: shared DiT core for iterative image inference
Radar item: LiFT repeatedly applies a shared Diffusion Transformer core and, according to the publication, outperforms a dense DiT on ImageNet 256×256 with about 60% fewer parameters.
According to the Radar item dated 6 October 2026, LiFT repeatedly applies a shared Diffusion Transformer core along a continuous depth path. The publication reports that, on ImageNet 256×256, the method outperforms a dense DiT baseline with about 60% fewer parameters. According to the source, the approach scales inference-time computation without retraining the model.
The topic is architecture of diffusion models for images, not a Rota Nacional feature. The platform does not offer this model and does not claim to reproduce the method. To verify, consult the original publication cited in the Radar source and check the benchmark figures, the metric used and the experimental conditions before quoting them. If you use AI to study the material, do not paste personal data or internal organizational documents into prompts.
Original source: LiFT usa um núcleo DiT compartilhado para inferência iterativa