ECHO
ECHO
> 看提示是因为输入尺寸不匹配的导致的，确定没有修改过代码吗 没有修改过，这行print语句是唯一自己写的代码
I solved the problem by delete the "tts_models/multilingual/multi-dataset/xtts_v2" model path and install tts0.22.0 ,it will redownload the model.Then you can get the speakers_xtts.pth I don't know why sometimes it didn't...
@jhj7905 when I run the inference code I get an AttributeError: 'TemporalBasicTransformerBlock' object has no attribute 'time_mix_inner_dim', do you know how to solve it
@jklj077 加载的时候就不均匀，加载的时候0卡大概占7G剩下3张卡占15G。用vllm启动就不会出现这种情况
so how to replace sd1.5 with sdxl,the text_emb size is different
看训练日志在第200000步时loss从0.031178424，提升到了5.4469901991 [log_20240222084743.txt](https://github.com/yerfor/GeneFacePlusPlus/files/14382799/log_20240222084743.txt)
`python inference/app_genefacepp.py --a2m_ckpt=checkpoints/audio2motion_vae --head_ckpt= --torso_ckpt=checkpoints/motion2video_nerf/obama_torso` 就是按照gradio的启动命令填的。 我感觉是训练过程中出的问题，日志里的validation结果中生成的图片就没有身体。
我使用py310重新编译paddle2.4.2仍然显示相同问题，是对python版本另有要求吗