TumCCC

> even better, Sonic demo: https://github.com/jixiaozhong/Sonic webpage & demo: https://jixiaozhong.github.io/Sonic/ Tried. Indeed! Thanks for your sharing

大家好，我这里最近在迁移一个蒸馏的lora到diffsynth_studio上(https://github.com/ModelTC/Qwen-Image-Lightning)，现在权重key值匹配问题已经解决了，但是发现diffuser推理的脚本上好像还改了scheduler的配置，然后diffuser和diffsynth_studio的flowmatching scheduler实现不一样，想请教下大家有遇到过类似问题么？

I met some bugs "torch.utils.checkpoint.CheckpointError: torch.utils.checkpoint: Recomputed values for the following tensors have different metadata than during the forward pass." when I set --use_gradient_checkpointing_offload and --training_strategy "deepspeed_stage_3" for the training...

> how to infer？ > > for example, how to use lightning_logs/version_27/checkpoints/epoch=0-step=22.ckpt/zero_to_fp32.py? and how to load the converted ckpt? I tried this code, but I got the issue like RuntimeError:...