Alpha-VLLM / Lumina-T2X

Lumina-T2X is a unified framework for Text to Any Modality Generation
MIT License
1.82k stars 74 forks source link

有计划,讲SD3的训练配置跟论文中的对齐吗 #83

Open heart-du opened 1 week ago

gaopengpjlab commented 1 week ago

请详细描述一下需求?没看懂😂

heart-du commented 1 week ago

sd3原论文中,涉及到rec flow的SNR Samplers参数配置,还有多尺寸训练,我看代码里面暂时还没有适配

gaopengpjlab commented 1 week ago

Lumina-T2X通过1D-RoPE实现灵活的多尺度训练。

zhuole1025 commented 1 week ago

我们lumina的代码中也有sd3的lognorm snr schedule,之后会加入到sd3训练脚本中:https://github.com/Alpha-VLLM/Lumina-T2X/blob/2e7c7319b1b4b3dc7939f78bd0eeffa3c13822d2/lumina_next_t2i/transport/transport.py#L112

heart-du commented 1 week ago

ok,好的好的