Looped Diffusion Transformer模型发布
精选理由
新发布的Looped Diffusion Transformer模型比大6.5倍的模型表现更好，计算需求却低4.9倍，效率惊人。
Looped Diffusion Transformer通过在去噪步骤中重复运行共享Transformer块来扩展计算能力。该模型在T2I基准测试中超越了规模大6.5倍的模型，同时推理计算需求降低4.9倍。该模型在图像生成任务中表现出色，计算效率显著提升。
原文 · Aran Komatsuzaki (论文推介)
Looped Diffusion Transformer
- Scales computation by repeatedly running shared Transformer blocks within each denoising step - Surpasses a model 6.5× larger across T2I benchmarks while requiring 4.9× lower inference compute
https://t.co/hw1EA2K6Rk https://t.co/dwYJvqHymL
