Yuxuan Zhang
Yuxuan Zhang
The FP8 we currently provide is converted through torchao, you can see it in readme, where the T5 part still retains BF16 and has not been converted
长度越长，占用的显存越多，你两张卡一共多少呢
这是正常的，但是很慢，因为已经开启了所有显存优化，如果你的GPU显存够，关闭cpu_load就行，cli_demo中有详细的解释。 readme中的数据是A100测的，如果你的GPU是T4，V100等，大概需要5-10分钟才能跑出一个视频
How long have you been waiting, how many datasets, I think this should be the time to process the datasets, can you check if htop shows any memory or CPU...
This job is not at the model level, you can check the implementation of `CogVideoXVideoToVideoPipeline` in diffusers
Yes, this is actually generating the next 6-second video, for each individual video, the duration does not exceed 6 seconds
Yes, this is actually generating the next 6-second video, for each individual video, the duration does not exceed 6 seconds
are cuda_12.5 compile with torch? abd what is your GPU
Thank you for your support. However, this might take some time as we currently do not have any related plans in the near future. Thank you for your understanding.